Sie sind auf Seite 1von 150

Chapter 5: Modes and basis functions Chapter 5: Modes and basis functions In the previous Chapter, we introduced the

concept of propagation modes in an optical fiber. The concept of modes is one that appears in many aspects of physical systems. Modes occur in essentially all physical systems that oscillate, for example, or whenever we consider wave propagation. We find modes throughout classical physics, including applications in fields such as acoustics, mechanical structural engineering, and the electrical engineering of waves, signals, and oscillators. Modes can also have some quite remarkable and very useful mathematical properties. In addition, the concepts we find in modes and the mathematics surrounding them are the core of the mathematics of quantum mechanics. For all these reasons, it is well worth our time therefore to spend some time finding out more about modes and some of their mathematics. The terminology varies somewhat between different fields. In the classical physics of oscillations or waves, the term mode is common, and in mechanical oscillations modes are also known as normal modes. In the mathematics of modes, modes are more usually known as eigenfunctions or basis functions, and in quantum mechanics they are also known as states or basis states. In the case of quantum mechanics and atoms, they are also known as orbitals. The underlying concepts are the same in all these cases, however. Examples of modes The reader might ask What is a mode? Despite that fact that modes are very common in physics and engineering, it is difficult to find a broad definition of modes in text books. Most (and possibly all!) scientists and engineers who would insist that they know perfectly well what a mode is would be hard pressed to give a simple definition that was not rather mathematical. There is a precise mathematical answer, but it gives little direct physical insight, so we postpone it. Here we will discuss a few examples to get across the concepts of modes first. We are quite used to the idea of modes in things that oscillate. In this case, the concept of a mode is essentially the same as that of a resonance, a preference to oscillate at a specific resonant frequency. Many such oscillators have only one way or mode in which they can oscillate (at least for small amplitudes of oscillation), and that is associated with only one frequency. Examples of this kind include a pendulum (at least if it is constrained to move in only one direction, as it usually is in a clock), and a mass on a spring (again if it is constrained to move in only one direction). You may have played a note on a wine bottle by blowing across the top; in this case, the oscillation is basically the air in the neck of the bottle behaving like a mass that is bouncing up and down on spring that is given by compressing the volume of air in the body of the wine bottle. This kind of acoustic resonator is known as a Helmholtz resonator. The body of a guitar together with the hole in the front of it also forms a Helmholtz resonator, and this concept is extensively used in loudspeaker design, especially for boosting the bass response of loudspeakers. Many other oscillating systems have many different resonances or modes. A guitar string has a fundamental mode, but also has harmonics that can also be excited, at twice the fundamental frequency, three times the fundamental frequency, and so on. The same is true for wind instruments such as the flute. Bodies that are essentially more two dimensional, such as a gong or a cymbal, have a much more complicated spectrum of

EE41 Physics of Electrical Engineering

Chapter 5: Modes and basis functions resonances (which are in general not in integer ratios of frequencies), and three dimensional structures, such as a bell or a girder bridge, can have yet more complicated sets of possible resonances or modes. There are also electromagnetic resonators, basically metal cavities, that have specific resonant modes, and there are optical resonant cavities of various kinds, formed with mirrors, that also similarly have resonant modes. In all these cases of oscillating modes or resonances, there are at least two common features. One is that each resonance or mode corresponds to quite a distinct way in which the object oscillates; the pattern of oscillation of the object is quite unique to a given mode, being quite different from any other mode. As we will see later, the way in which the patterns of oscillation are different from one another can be quite specific mathematically; in a broad range of situations, they are orthogonal, and we will be introducing that concept as we describe modes. A second feature of oscillating modes is that, for each such resonance or mode, there is one well defined numerical quantity, usually the frequency, associated with it. A third important point about all such modes is that, at least if we take a loss-less and small amplitude idealization of the mode, once excited, the oscillation would stay of exactly the same form for ever, and everything that was oscillating would be oscillating at the same frequency in that mode. The other kind of mode that we have already encountered is that of propagating modes. In the case of waveguides, there are particular cross-sectional patterns (again usually all orthogonal to one another) that, again in a loss-less idealization, will propagate in exactly the same form for ever down such a waveguide. Associated with each such mode again is a specific numerical parameter, usually defined as a propagation constant or propagation velocity. We can see therefore that there is a strong mathematical similarity between oscillating modes and propagating modes; in both cases there is a specific mathematical function a mode form that is associated with a specific numerical parameter. People sometimes also talk about the modes of free space when they are talking about waves outside waveguides. Some such modes do not retain exactly the same form as they propagate (for example, laser beams do get somewhat larger in cross-section as they propagate), and might not seem to share the characteristic of a constancy of form with the other modes we have been discussing. We can, if we want, imagine that these are really the modes of a very large resonator, which could remove this formal difficulty. To proceed further in understanding modes, we need to analyze a few cases to begin to uncover the important mathematical properties and a tighter definition of what a mode is. Analysis of systems with modes Simple oscillator Perhaps the simplest system we can think of to illustrate modes is a simple harmonic oscillator, for example consisting of a mass on a spring, as shown in Fig.5.1. For extreme simplicity here, we imagine the mass can only move up and down, not sideways, we presume it cannot twist or rotate in any way, and we presume that there is no friction. The common dictionary definition of mode that is relevant here is that of a manner or way. There is only one way in which this mass can move, which is to oscillate up and down with a specific frequency .

EE41 Physics of Electrical Engineering

Chapter 5: Modes and basis functions

Spring, spring constant k Mass m


Fig.5.1 Mass on a spring, without friction, oscillating up and down. The analysis of this system is straightforward. Taking x as the position of the mass in the vertical direction, we have, starting from Newtons Second Law, for force F and acceleration a,

F = ma = m

d 2x dt 2

(5.1)

The force here is the force from the spring, which is, for a spring constant k
F = kx

(5.2)

This expresses the fact that the force from the ideal spring is linearly proportional to its extension, x, with proportionality constant k, the spring constant. Here we presume we are measuring x from the equilibrium position the mass would be at if it were not oscillating. The minus sign in (5.2) comes from the fact that the force is a restoring force, pushing the mass back towards x = 0, the equilibrium position. Putting (5.1) and (5.2) together, we get, for x as a function of time t,

d 2 x (t ) k = x (t ) 2 dt m

(5.3)

The solutions to Eq. (5.3) are well known, as is easily verified by substitution, to be of the form sin (t ) (or cos (t ) , or exp ( it ) ) where

2 =

k m

(5.4)

To describe the oscillation completely, we would need to specify an amplitude and a phase for the oscillation, but the only possible solution is a sinusoidal (or harmonic), vertical oscillation at (angular) frequency . That sinusoidal oscillation is the only possible mode of oscillation of this system. The concept of modes is most useful for linear systems, that is ones whose behavior is the same regardless of the amplitude of the oscillation. In Eq., if x ( t ) is a solution to the equation, so also is gx ( t ) , where g is any constant (as is easily verified by trying gx ( t ) in Eq. (5.3)), so this equation is linear in that sense. An example of a nonlinear equation would be one such as

EE41 Physics of Electrical Engineering

Chapter 5: Modes and basis functions

d 2x = cx 2 dt 2

(5.5)

where c is a constant. If some particular form x ( t ) is a solution of this equation, it is not in general true that bx ( t ) is a solution, as is easily checked by substitution. The fact that Eq. (5.3) is linear is a consequence of our choice of a linear spring, i.e., one whose force is linearly proportional to its extension. The rest of our discussion of modes will be restricted to linear systems and their associated linear equations.

Linear eigen equations


Note that the equation (5.3) of our linear oscillator is of the form Operator operating on the function x(t) = a constant times the function x(t) where here the operator 1 is d2/dt2 and the constant is -2. If we denote an operator by A and functions by x and y respectively, then we can introduce a mathematical notation, and write (5.6)

Ax = y

(5.7)

when we mean that operator A operating on function x gives rise to function y. With that notation, we can write Eq. (5.6) in general as
Ax = bx

(5.8)

where A represents the operator, x represents the function, and b represents a constant. The form of equation given by relation (5.6) (or (5.8)) is called an eigen equation 2 . Any constant for which such an equation holds is called an eigenvalue, and a function that is a solution for that eigenvalue is called an eigenfunction associated with that eigenvalue 3 . The particular eigen equation (5.3) for our physical problem of a mass on a spring, for a given k and m, only has one eigenvalue ( k / m ) and one eigenfunction. As we stated above, this particular oscillator is an example of an oscillator that only has one mode.

Linear operators
One final point we should note is that the operator d2/dt2 is itself a linear operator. The definition of a linear operator is that it should obey the relations (i) Operator operating on (b times x) = b times (Operator operating on x) (5.9)

1 2

An operator is something that turns one function into another function.

eigen is a German word with several possible translations into English depending in the context; one translation is own as in the sense of ones own. It is possible in general to have more than one eigenfunction associated with a given eigenvalue, in which case the eigenvalue is said to be degenerate. Incidentally, for the equations associated with linear physical problems, there can only be a finite number of such eigenfunctions associated with a given eigenvalue; that number is called the degeneracy of the eigenvalue.

EE41 Physics of Electrical Engineering

Chapter 5: Modes and basis functions for an arbitrary constant b and a function x, and (ii) Operator operating on (x1 + x2) = (Operator operating on x1) + (Operator operating on x2) for any two functions x1 and x2. By this definition, it is straightforward to see that d2/dt2 is linear. In our more abstract operator algebra, incidentally, we could write Eq. (5.9) as
Abx = bAx

(5.10)

(5.11)

and Eq. (5.10) as

A ( x1 + x 2 ) = Ax1 + Ax2

(5.12)

By now the reader can see that we are moving towards defining modes as the eigenfunctions of an operator, especially a linear operator, and that is indeed the case. Before tidying up some details on that definition, we will discuss another examples. This will also expose other important characteristics of modes. Coupled oscillator Now let us consider the situation of Fig.5.2. Here we have two masses and three springs. We are presuming here that the masses can only move from side to side, and that the motion is frictionless. The positions of the two masses relative to their equilibrium positions are x1 and x2 respectively. The masses and the spring constants are the same as before for the simple oscillator.

Identical springs x1 x2

Identical masses
Fig.5.2. Coupled oscillator made with two masses and three springs. We now can apply Newtons second law to each of the masses. This gives us two coupled differential equations. Note first that the length of the spring in the middle, relative to its equilibrium length, is x2 x1 . Hence we have, for the mass on the left,

d 2 x1 = kx1 + k ( x2 x1 ) = 2kx1 + kx2 dt 2

(5.13)

and for the mass on the right

EE41 Physics of Electrical Engineering

Chapter 5: Modes and basis functions

d 2 x2 = k ( x2 x1 ) = 2kx2 + kx1 dt 2

(5.14)

One of the intuitive ideas we have about a mode of oscillation is that, in such a mode, everything that is oscillating is oscillating at the same frequency. To look for modes in this problem, we therefore propose that there are (sinusoidal) oscillating solutions at angular frequency , and see if we can find any solutions. For solutions of the form sin (t ) (or cos (t ) , or exp ( it ) ), we therefore have, from (5.13) and (5.14) respectively
2 mx1 = 2kx1 + kx2

(5.15)

and
2 mx2 = 2kx2 + kx1

(5.16)

These equations can be written in a matrix form as


2k k k x1 x = 2m 1 2k x2 x2

(5.17)

i.e.,

x1 2 1 x1 1 2 x = x 2 2

(5.18)

where We can if we wish rewrite Eq. (5.18) as

2m
k

(5.19)

Ax = x

(5.20)

with

2 1 A= 1 2
and
x x = 1 x2

(5.21)

(5.22)

We see that we have reduced this problem to an eigen problem, with eigenvalue , which when written in the form of Eq. (5.20) is in exactly the same general form as we encountered in dealing with the simple, one-mass, oscillator (Eq. (5.8). Here, as before, the operator A is linear (in fact, matrices only ever represent linear operators). Now x is a vector, with its two elements representing the positions of the two masses.

EE41 Physics of Electrical Engineering

Chapter 5: Modes and basis functions

To solve the problem, we can now take the standard approach for finding the eigenvalues and eigenvectors of a matrix. Rewriting we have
2 1 1 x1 =0 2 x2

(5.23)

For Eq. (5.23) to hold, the determinant of the matrix must be zero, i.e.,

( 2 )( 2 ) 1 = 0
i.e., i.e.,

(5.24) (5.25)
(5.26)

2 4 + 3 = 0
= 2 1 2
1

i.e.,

2 =

2k m

1 1 2

(5.27)

From Eq. (5.23), substituting the two allowed values of , we can deduce the form of the associated eigenvectors, which are, respectively,

=1:

1 x1 = x2 , i.e., x 1

(5.28)

i.e., the two masses are oscillating in the same direction, with equal amplitude, with frequency = k / m , and

= 3;

1 x1 = x2 , i.e., x 1

(5.29)

i.e., the two masses are oscillating in opposite directions, with equal (but opposite) amplitudes, with frequency = 3k / m . Perhaps not surprisingly, then, now that we have two masses, we have discovered two different modes, a symmetric, lower frequency one, with both masses going in the same direction, and an antisymmetric, higher frequency one with them going in opposite directions. Orthogonality of modes Note one very important point about these two different eigenmodes; they are mathematically orthogonal. That is, if we take the vector dot or inner product of the two eigenvectors corresponding to the two modes of this system, we find the result is zero, i.e.,

[1 1]

1 =0 1

(5.30)

EE41 Physics of Electrical Engineering

Chapter 5: Modes and basis functions

If we had drawn the two vectors on a piece of paper, we would find they are at right angles to one another. This orthogonality property of modes in many classes of problems is a very important one. This orthogonality is also obvious if we plot these vectors on a two-dimensional plane, as shown in Fig.5.3.
0 or x2 1 direction

1 1 direction 1 or x1 0 direction

1 1 direction

Fig.5.3. Plot of vectors representing functions. Functions as vectors Note now that we are representing a function as a vector. The function here is a very simple one; it is the function that describes the positions of the masses at a given time, and it simply is a list of two numbers, the position of mass 1 and the position of mass 2. Any list of numbers can be written as a vector. In general, we can represent any function as simply a list of numbers, though for functions of continuous variables that list would be infinitely long. In general, though, we can if we want represent any function as a vector. We will give more examples of this idea below. Completeness Another very important point about these modes is that they can be used to describe any particular position of the two masses. This rather important property that sets of modes can have is called completeness. A particular position of the two masses, namely x1 = f and x1 = g can rather obviously be represented as

f 1 0 = f +g g 0 1
but it can equivalently be represented by

(5.31)

f ( f + g ) 1 ( f g ) 1 + g = 2 1 2 1

(5.32)

Again it is obvious if we look at Fig.5.3 that we could equally well represent any point in 1 0 the plane in terms of its coordinates along the and directions, or in terms of its 0 1

EE41 Physics of Electrical Engineering

Chapter 5: Modes and basis functions

1 1 coordinates along the or directions. Either one of these orthogonal basis sets 1 1 is equally usable. Another way of stating this same property is that the modes form a complete set for spanning the space 4 that describes the positions of the masses.
Hermitian adjoints and Hermitian operators If we look at the operator A in this problem, as given by Eq. (5.21), we can see that the matrix is symmetric if we reflect it along its leading diagonal (the diagonal with the elements of value 2). In the mathematics of modes, it is usual to allow for the possibility of complex quantities, and then, in an extension of the idea of a symmetric matrix, we can define a Hermitian matrix. First, we need formally to define the idea of a Hermitian adjoint of a matrix. The operation of taking a Hermitian adjoint involves reflecting a matrix along its leading diagonal (i.e., taking the transpose of the matrix), and taking the complex conjugate of the elements when we do so. The Hermitian adjoint is often denoted with the dagger symbol . For example, for a 2 x 2 matrix, we have
a b a c d b

c d

(5.33)

We can also have Hermitian adjoints of matrices that are not square, with the most common example being a vector. In the case of a 2-element vector, for example, we have
a b a

(5.34)

We can think of the Hermitian adjoint as being like the idea of a complex conjugate but now generalized to matrices or operators. A simple scalar number can be thought of as a 1 x 1 matrix, and the Hermitian adjoint of that matrix is simply the complex conjugate of the number. Given that we now have the possibility of complex numbers, we need to make one other minor change. Now the idea of a vector dot product is generalized to an inner product in which we use the Hermitian adjoint vector as the one on the left 5 , i.e., the inner c a c product of and is given by a b = ac + b d . d b d

Incidentally, the space we are speaking about here is not an actual geometrical space, but rather a function space, i.e., one in which we can represent entire functions. (Here the functions are rather simple, since they can be described entirely by specifying only two numbers, and so we can represent them in a two dimensional space.) Spaces of this class are known as Hilbert spaces. They are mathematically like ordinary geometrical spaces in many ways, but are different, first, in that they may have any number of dimensions, including an infinite number, because that many values may be required to represent a function, and, second, that the coordinates in such a space may be complex rather than only real.
5

The reader might be asking why do we need to use this Hermitian adjoint, with its complex conjugates, when taking this inner product? The reason is that we need to have a simple measure of length (for a

EE41 Physics of Electrical Engineering

Chapter 5: Modes and basis functions

A Hermitian matrix is one that is equal to its own Hermitian adjoint. In abstract operator function notation, a Hermitian operator is one for which

A = A
For example, the matrices

(5.35)

1 0 0 1 ,

1 i 0 1 i 1 i 2 , and 1 + i 3 exp i / 2

exp i / 2 1

are all Hermitian, but the matrices

1 0 i 1 1 0 , 1 i , and
are not.

1 i i 1

A linear Hermitian operator in general we can think of as being one that can be represented by a Hermitian matrix. Three very useful properties of Hermitian matrices are (i) that their eigenvectors are always orthogonal to each other, (ii) that the eigenvalues are always real, and (iii) that the eigenvectors form a complete set for the space in question. Because all the elements of the matrix A as given by (5.21) happen to be real, then this particular symmetric matrix is also Hermitian. Hermitian operators and physical problems In practice, for a large number of physical problems associated with linear systems, especially those that correspond to frictionless or loss-less systems, the key linear operators we deal with are Hermitian 6 , and can be represented to any degree of accuracy

vector) that gives a real number, even if the elements of the vector are not real. In the language of the mathematics of these Hilbert spaces, we need a norm that is real. It is very useful (and in fact a requirement for a Hilbert space) if that norm is constructed from the inner product of the vector with its own adjoint (in fact, the norm is defined as the square root of that inner product, just as it would be for vectors in an ordinary geometrical space). Taking the complex conjugate in constructing the adjoint guarantees the result is a real number, just as taking the product of a complex number with its complex conjugate always results in a real number. As the reader will have noticed, the operator d2/dt2 is Hermitian, and the rather trivial operation of multiplying a function by a real constant is also Hermitian. Somewhat surprisingly, the operator d/dt is not Hermitian, though the operator id/dt is. Simple frictional systems can have terms in the differential equation that involve the operator d/dt because frictional damping force is often proportional to velocity, at least in simple systems. Physical systems that have such frictional damping are therefore not represented by Hermitian operators. It is actually an axiom of quantum mechanics that all operators associated with measurable quantities are Hermitian.
6

EE41 Physics of Electrical Engineering

10

Chapter 5: Modes and basis functions

we wish by a sufficiently large Hermitian matrix. 7 The kinds of operator we are interested in here are all linear operators that can be represented by matrices to any desired degree of accuracy, and many of these operators are Hermitian. By extension from the matrix case, the eigenfunctions of Hermitian operators that can be represented by matrices are always orthogonal, with real eigenvalues, and form a complete set of basis functions. Modes as the eigenfunctions of operators Finally, we can write a definition that will encompass the concept of modes for our purposes. As the reader has been warned, this is rather mathematical, but we have now introduced enough terminology that we can make the statement. The broadest possible statement we can make is A mode is an eigenfunction of an operator that describes a physical system. (5.36)

The system need not necessarily be linear, and even if linear, it need not be represented by Hermitian operator. 8 The most useful sets of modes from a mathematical point of view are those that correspond to linear, Hermitian operators. We can also state 9 that For linear physical systems that can be described by a Hermitian operator that can be approximated to any degree of accuracy by a finite matrix, the modes are the eigenfunctions of the operator, they are orthogonal, they are complete, and they have real eigenvalues.
Standing waves on a string

(5.37)

So far we have been dealing with oscillating systems with only a discrete number of objects that can oscillate, namely the masses on the springs. That has enabled us to write simple vectors to represent the positions of the masses, and to show that the different modes have orthogonal vectors associated with them. We want to make the transition to continuous functions, to show how they also can be viewed as vectors, and to show that way in which continuous eigenfunctions can be orthogonal. A good example to show all these points is that of standing waves on a string. We will start by considering the modes

Operators that can be represented to any desired degree of accuracy by a sufficiently large matrix are called compact operators.

There are some operators that are not Hermitian that lead to real eigenvalues (in the problem of two unequal masses with three springs, the resulting matrix is not Hermitian, though the eigenvalues are real) This definition is slightly imprecise in that not necessarily all Hermitian operators can be represented to any degree of accuracy by finite matrices. In practice, however, we can typically reduce linear problems to ones that we can solve with matrices, and in that sense we can regard this definition as being broad enough for a discussion of such Hermitian problems.

EE41 Physics of Electrical Engineering

11

Chapter 5: Modes and basis functions

on a continuous string,. Then, to show the connection to functions as vectors, and to justify the wave equation we will use, we will consider discrete masses on a string. We presume we are considering a string of length L, tied to rigid walls at both ends. The direction along the string will be the z direction. The string itself is a continuous object. To describe the position of the string at each point between 0 and L, we need a continuous function, y(z), presuming for simplicity here that the string can only move in the y direction. This his function simply represents the position in the y direction of the string relative to the straight, undistorted string. We presume that waves on this string will obey a wave equation, of the form

2Y ( z, t ) 1 2Y ( z, t ) 2 =0 vo z 2 t 2

(5.38)

where vo is the wave propagation velocity on the string (which depends on the linear mass density and the tension in the string). (We will formally derive this equation later.) Here, z is the distance along the string, from 0 to L. Y(z,t) is the sideways displacement of the string from its equilibrium straight line at any particular point z along the string, and at any particular time t. The wave equation itself is not in the form of an eigen equation. Again, however, because we are looking for oscillating modes, we will ask that all the oscillations are at some frequency , so that Y is of the form

Y ( z, t ) = y ( z ) sin (t )
Substituting this form into Eq. (5.38), we obtain the simpler wave equation d 2 y ( z) = k 2 y ( z ) dz 2 where

(5.39)

(5.40)

k2 =

2
2 vo

(5.41)

A wave equation like Eq. (5.40) is known as a Helmholtz equation. We recognize that we have an eigenvalue equation, where the eigenvalue is k2, though this time it is a differential equation in z, not t. Just as before, this equation can be written in an abstract mathematical form as
Ay = k 2 y

(5.42)

where now y represents the entire function y(z), and A corresponds to the linear operator d2/dz2, which turns out also to be a Hermitian operator, just as d2/dt2 was before. The solutions of this equation are of the form

y ( z ) sin kz

(5.43)

EE41 Physics of Electrical Engineering

12

Chapter 5: Modes and basis functions

(or possibly cos kz or some combination of the two). To find the actual solutions, we need to put in boundary conditions, which are that

y ( z ) = 0 for z = 0, L

(5.44)

Our guess of sin kz was wise, because sin 0 = 0. Also sin n = 0 for any integer n. Hence we find that the allowed solutions occur for the possible eigenvalues
k n = n / L

(5.45)

with associated eigenfunctions n z yn ( z ) sin L (5.46)

The case of n = 0 is a trivial one, because the wave would then be zero everywhere on the line. Also, solutions for negative n are really the same solutions as for positive n, so we only need to consider n = 1, 2, 3, . These solutions are simple standing waves on the line. These different standing waves are the modes of this problem. These are illustrated in Fig.5.4. Of course what is drawn in this figure represents the amplitudes of the oscillating modes; from (5.39) we know that the string is oscillating at the appropriate frequency, from Eq. (5.41)

n = vo kn =

n vo L

(5.47)

Fig.5.4. Amplitudes of the first three standing wave modes on a string. Masses on a string Now we will consider a related problem, which is that of a string with equal masses m equally spaced by an amount z along its length. For the moment, we pretend that the string itself has no mass. This problem for the case of two masses is shown in Fig. 5.5. There is a tension T in the string. To analyze this problem, we look at the case of a single mass somewhere along the string. We presume the string is distorted because the masses
EE41 Physics of Electrical Engineering 13

Chapter 5: Modes and basis functions

are displaced from the original straight line along the z direction, and analyze the net force in the vertical (y) direction on the mass. This analysis is shown in Fig. 5.6, where here we are considering the jth mass on a string that may have many masses on it.

y z y1 T y2

z
L
Fig. 5.5. Two masses on a string, with tension T in the string. We see from this figure that, because of the tension T in the string, there is a force Tsinj pulling the center mass (mass j) upwards, and a force Tsinj-1 pulling the center mass downwards, so the net upwards force on mass j is
F j = T ( sin j sin j 1 )

(5.48)

For small angles,

sin j

y j +1 y j z

(5.49)

and similarly for sinj-1. Hence, for small amplitudes of oscillation,


Fj = T ( y j +1 2 y j + y j 1 ) z

(5.50)

yj+1 yj-1 yj T T

j-1 z z

Fig. 5.6. Forces on a mass with tension T in the string. For the case of the leftmost mass on the string, with j = 1, yj-1 = 0 because it is simply the point at which the string is tied to the wall on the left, and similarly for the rightmost mass on the string, yj+1 = 0 because the string is tied to the wall on the right. For the case of two masses, the equations of motion of the two masses become

d 2 y1 T T = 2 y1 + y2 2 dt z z

(5.51)

EE41 Physics of Electrical Engineering

14

Chapter 5: Modes and basis functions

d 2 y2 T T = 2 y2 + y1 2 dt z z

(5.52)

The reader will see that, not surprisingly, these two equations have exactly the same form as Eqs. (5.13) and (5.14), with T/z instead of k, and the solutions are therefore essentially identical. We will have two modes here, one with both masses oscillating up and down in phase with equal amplitudes, and the other with the masses moving equally in opposite directions. As we increase the number of masses, we can take the same approach to solving the problem. We will have three equations for three masses, and hence a 3 x 3 matrix, with three modes. Similarly we will have four equations for four masses, and hence a 4 x 4 matrix, with four modes, and so on. As we increase the number of masses, it becomes relatively easy to see that the problem settles into a simple form that grows with the number of masses, and the form of these modes starts to become clear. For example, for 5 masses, we would have
y1 2 1 0 0 0 y1 y 1 2 1 0 0 y 2 2 mz 2 y3 0 1 2 1 0 y3 = T y4 0 0 1 2 1 y4 y5 0 0 0 1 2 y5 which would give us eigenvectors

(5.53)

1/ 2 3 / 2 1 , 3 / 2 1/ 2

3/2 1 3/2 0 0 , 1 , 3 / 2 0 3 / 2 1

3/2 3 / 2 0 , and 3/2 3 / 2

1/ 2 3 / 2 1 3 / 2 1/ 2

(5.54)

We can plot the positions of the masses at some particular instant of time (or equivalently the amplitudes of the oscillations of each mass) at the positions of the masses along the string. For example, the positions or amplitudes corresponding to the first eigenvector are shown in Fig. 5.7. Note that these touch a bounding curve of the form sin z / L , which we recognize as the first standing wave mode on a vibrating string. If we were to examine each of the other eigenvectors, we would find that the corresponding positions of the masses would touch a bounding curve of the form sin n z / L , where n is the number of the mode (in the order of the eigenvectors above in (5.54)). Fig. 5.8 shows the third mode, with its bounding curve of sin 3 z / L . We can now see intuitively that the eigenvectors of problems with large numbers of equal and equally spaced masses on a string will, with progressively larger numbers of such masses, essentially give us the functions that we know to be the standing wave modes on a string. We can also see directly here how such a continuous function can be represented to any degree of accuracy we desire by a (large but finite) vector of values.
EE41 Physics of Electrical Engineering 15

Chapter 5: Modes and basis functions

y z sin z/L

y1

y2

y3

y4

y5

z
L
Fig. 5.7. Positions or oscillation amplitudes of the five masses on a string, in the first eigenmode. Note that these amplitudes touch a bounding curve of the form sin z / L .

y z sin 3 z/L

y1

y2

y3

y4

y5

L
Fig. 5.8. Positions or oscillation amplitudes of the five masses on a string, in the third eigenmode. Note that these amplitudes touch a bounding curve of the form sin 3 z / L . (The amplitudes y2 and y4 are zero.) For these finite vectors representing the functions, we could, as before, write an inner product between the two such vectors, for the case of q masses,

EE41 Physics of Electrical Engineering

16

Chapter 5: Modes and basis functions


yB1 y B2 yq yBq

A Inner product of vectors y A and y B = y y B = y 1 A

y 2 A

(5.55)

= y yBi = y ( zi ) yBi ( zi ) Ai A
i =1 i =1

where at the end we are writing the vector elements as being values of a function at the positions zi of the masses on the string. The definition of the inner product in Eq. (5.55) is alright, but as we keep increasing the number of masses on the string, it gets larger and larger for any specific overall scale of the displacements y. We can avoid that problem, and get an inner product that will converge to a specific result independent of how large a number of masses we have, if we redefine the inner product as Inner product of vectors y A and y B = y ( zi ) yBi ( zi ) z A
i =1 q

(5.56)

where we have multiplied by z. As the number of masses gets larger, and hence the overall size of the inner product would grow, z gets proportionately smaller, keeping the inner product the same size. Now we can propose a form of the inner product that will work well for continuous functions. We will now formally take the limit of Eq. (5.56) as z tends to zero (and q tends to infinity), in which case the sum becomes an integral, and we have

Inner product of functions y A ( z ) and yB ( z ) = y ( z ) yB ( z ) dz A


0

(5.57)

The lower limit on this integral is zero, because z1 = z 0 and the upper integral is L L L. because zq = qz = q q +1 Now we have a way of defining orthogonality between two continuous functions, which is to require that their inner product is zero, i.e., two functions y A ( z ) and yB ( z ) are orthogonal if

y ( z ) y ( z ) dz = 0
A

(5.58)

Of course, the limits to the integration will depend on the precise physical problem, and this concept can also be generalized to functions of multiple dimensions simply by performing the integral over all those dimensions (e.g., a volume integral for three dimensional functions.

EE41 Physics of Electrical Engineering

17

Chapter 5: Modes and basis functions

Formal derivation of the wave equation

The procedure we started above can also be used to derive a wave equation for the string. Instead of presuming the masses are discrete, and are spaced by z, we can instead presume the string has mass, with a uniform linear density (mass per unit length) of , and we can replace the masses of value m with masses of value z. We can consider the force that we formally deduced as being on a given mass m as being on the mass of this element of the string, so for that element of the string, we have, from Newtons second law
F = z

2Y t 2

(5.59)

We are now formally writing our position function in the y direction as Y ( z , t ) to allow explicitly for it to vary in both time and distance z, and we are taking the partial derivative with respect to time t because we are also about to introduce derivatives with respect to z. We note from Eq. (5.50) that we can write the net vertical force on this element of the string as
F (z) = T y ( z + z ) 2 y ( z ) + y ( z z ) z 1 y ( z + z ) y ( z ) y ( z ) y ( z z ) = T z z z z

(5.60)

The term inside the curly brackets is the difference of the gradients of the string at two points separated by z, divided by z, which is, in the limit of small z, simply the second derivative, i.e., in the limit of small z

2Y F ( z ) T z 2 z
and so, putting this together with Eq. (5.59), we have 2Y 2Y =0 z 2 T t 2

(5.61)

(5.62)

which is just the wave equation (5.38) we used earlier on for waves on string, with a wave propagation velocity

vo =
Second derivative operator

(5.63)

We note, incidentally, that the matrix

EE41 Physics of Electrical Engineering

18

Chapter 5: Modes and basis functions

2 1 0 0 0 1 2 1 0 0 1 0 1 2 1 0 M= 2 z 0 0 1 2 1 0 0 0 1 2

(5.64)

is, in the limit of small z, effectively a representation of the operator

2 , which z 2 illustrates by example how linear operators can be represented by matrices. Note that this 2 matrix is Hermitian, just as the operator 2 is Hermitian. z
Relation to Fourier series

As we mentioned above, the sets of modes that we get by solving a particular physical problem can also turn out to be complete sets for describing functions in the relevant function space. Here, our function space is one containing the one dimensional functions of position defined between 0 and L. If our sets of modes are complete, we should be able to describe any functions in terms of combinations of these modes. The reader may be aware of Fourier series. There are a few different forms of Fourier series, but essentially Fourier series can be used to represent arbitrary functions as sums of sine waves. To represent an arbitrary function f(z) between 0 and L, we could use a Fourier series of the form
n z f ( z ) = an sin L n =1

(5.65)

where the an are appropriate numbers (coefficients). But our modes here are actually identical to the sine waves used in the Fourier series. Hence, since we already know that Fourier series can represent arbitrary functions in this way, so also can our set of modes. In other words, from our knowledge of Fourier series, we already know that this particular set of functions forms a complete set, a property we have claimed that sets of modes possess. Alternatively, we could state that the Fourier basis functions (the sine waves) form a complete set because they happen to be the eigenfunctions of a Hermitian operator 10 .

Incidentally, just as we could represent a function to any degree of precision we wish by a vector of its values at equally spaced points in space, we can also represent the same function as a vector whose elements are the coefficients an of its Fourier series. One way of looking at this is to say that the change from representing the function as values at points in space to representing it with values of its Fourier coefficients is to say that we have not actually changed the vector in the function space it is still fundamentally the same function, and, as a vector it still has the same length and still points in the same direction but that we have rotated the coordinate axes that we use to represent it. In this (quite correct) view, the Fourier analysis does absolutely nothing to the function!

10

EE41 Physics of Electrical Engineering

19

Chapter 5: Modes and basis functions

Generality of basis sets of functions

Note, incidentally, that such basis sets of functions can be used to represent any function, regardless of whether the actual physical situation corresponds to the one used in deriving the set. For example, we could have a string whose density varied along its length. The modes of such a string would not be the simple sine waves we have here. Nevertheless, at any given time, we could if we wanted still represent the shape of the string as a sum of our sine wave modes. In the previous chapter, we looked at the two-dimensional functions that describe the modes that propagate in an optical fiber. Those two-dimensional functions form a complete basis set of functions, and can be used to describe any function of two dimensions, regardless of whether it even has anything to do with optical fibers. There is in fact an infinite number of possible basis sets of functions to describe functions in any given space. This mathematical fact can be very useful; in any given problem, there is often a very convenient basis to choose that makes the problem simple to solve. This notion of multiple different possible sets of basis functions is also basic to the mathematical basis of quantum mechanics.
Summary

Definitions of modes

Informal verbal definition specific distinct ways in which physical systems can oscillate, resonate, or have waves propagate with each way associated with a specific value of a numerical parameter, such as oscillation or resonance frequency, or propagation velocity or wavevector Mathematical definitions A mode is an eigenfunction of an operator that describes a physical system. For linear physical systems that can be described by a Hermitian operator that can be approximated to any degree of accuracy by a finite matrix, the modes are the eigenfunctions of the operator, they are orthogonal, they are complete, and they have real eigenvalues.
Mathematical terms

Operator An operator operating on a function turns that function into another function

EE41 Physics of Electrical Engineering

20

Chapter 5: Modes and basis functions

Abstract mathematical terminology for operators and functions If we denote an operator by A and functions by x and y respectively, then we can introduce a mathematical notation, and write

Ax = y
when we mean that operator A operating on function x gives rise to function y. Linear operator If an operator has the properties
Abx = bAx

and

A ( x1 + x 2 ) = Ax1 + Ax2
for any constant b and any functions x1 and x2 in the space of interest, then it is linear. Operators, matrices, and functions as vectors Matrices always represent linear operators. Functions can be represented as vectors, because we can always write out a function as simply a list of its values at all the possible values of the relevant variable or variables. Hermitian adjoint The Hermitian adjoint of a matrix or vector is formed by reflecting it about its leading diagonal (i.e., taking the transpose of the matrix or vector), and taking the complex conjugate of all of the elements. It can be viewed as a generalization to matrices of the concept of the complex conjugate of a number. Taking the Hermitian adjoint of a vector changes it from a column vector to a row vector (or vice versa), and takes the complex conjugate of all the elements. The Hermitian adjoint is denote by the symbol . Hermitian operator A Hermitian operator is one that is equal to its own Hermitian adjoint, i.e.,

A = A
Eigen equations, eigenfunctions, and eigenvalues An eigen equation is one that can be written in the form
Ax = x

where A is an operator, x represents a function, and is a constant. The values of for which there are solutions of this equation are called eigenvalues, and the functions that are solutions corresponding to those eigenvalues are called
EE41 Physics of Electrical Engineering 21

Chapter 5: Modes and basis functions

eigenfunctions. For the specific case of operators represented as matrices, the eigenfunctions are also called eigenvectors. Mathematical properties of mode, and complete sets of functions The following are properties of the eigenvalues and eigenfunctions of Hermitian operators, and hence are properties also of modes. All the modes of a given problem are orthogonal to one another. All the eigenvalues are real. The set of eigenfunctions (modes) is a complete set, which means that any function in the space of interest can be represented as a weighted sum of the eigenfunctions (modes). Basis sets Complete sets of orthogonal functions are called basis sets. Standing wave modes on a string As an example, on a string of length L, tied to walls at both ends, the standing wave modes on the string are n z yn ( z ) sin L with associated frequencies

n = vo kn =

n vo L

where vo is the wave propagation velocity on the string. Helmholtz wave equation The Helmholtz wave equation applies to waves of a specific frequency in a uniform medium, and is of the form d 2 y ( z) = k 2 y ( z ) 2 dz

EE41 Physics of Electrical Engineering

22

Chapter 6: Modern physics for electrical engineering Chapter 6: Modern physics for electrical engineering This section introduces some of the key concept in modern physics, especially quantum mechanics, that we need for understanding the modern devices that sense, communicate, process, and store information. How does a transistor work? How does a digital camera sense its picture? How do light emitting diodes (used in everything from indicator lights to traffic lights) work? How does the remote control for your television work? How do we get information on and off optical fibers? How do we read and write information on optical disks? How can we make any of these devices work better? Questions like these can only be answered if we introduce some physical ideas beyond the classical physics of electromagnetism and ordinary mechanics. The understanding of that modern physics allows us to engineer a remarkable and truly revolutionary collection of devices. Simply put, classical physics does not explain many of the things we see, nor does it give us enough tools to engineer everything we need. Many everyday aspects of the world around us cannot be understood without quantum mechanics. For example, all of chemistry essentially cannot be understood without introducing quantum mechanics. Almost no aspects of why materials have the mechanical properties they do (such as strength and hardness) can be understood without introducing quantum mechanics. We can understand almost no aspects of the color of objects without quantum mechanics. (We cannot even explain why an ordinary light bulb has the color it has without using quantum mechanics.) The electrical, magnetic, and optical properties of materials and devices can similarly only be understood through quantum mechanics. As we move beyond the everyday, macroscopic world into objects on size scales substantially below about 100 nm (about 1 thousandth of the thickness of a human hair), into the world of nanotechnology, we start to run directly into quantum mechanical effects that are related to the size of objects. Things stop working the same way when we make them so small. Quantum mechanics is therefore a great boon to engineers. It enables us, for example, to choose and modify materials to suit our purposes, it lets us join the worlds of optics and electronics, it enables us to predict the properties of devices, even as we make them very small, and it allows us to design and make completely new kinds of materials and devices. These abilities of quantum mechanics are particularly useful in the devices and technologies we need for handling information. The solid state physics that underlies most semiconductor electronic devices is firmly based on quantum mechanics. Successive generations of ever smaller transistors become increasingly influenced by additional quantum mechanical size effects (for example, electrons can tunnel quantum mechanically through the insulating oxide that separates a transistor gate from the conducting channel). Eventually, as we scale down devices, we may have transistors that operate on very different, quantum mechanical principles. Optical devices are quantum mechanical on many different levels. The generation and detection of light, especially in small, efficient and fast devices, is almost entirely done using quantum mechanical techniques; light is created and sensed by the creation and absorption of photons, the quantum mechanical particles of light. Additionally, many modern optoelectronic devices, the devices that connect the optical and electrical worlds, are engineered using quantum mechanical principles with very thin layered materials (quantum wells). Modern magnetic storage devices also involve very advanced quantum EE41 Physics of Electrical Engineering 1

Chapter 6: Modern physics for electrical engineering mechanical and solid state physics principles to get the ever-increasing storage densities we have come to expect. In addition to our engineering interest, of course, we might be motivated to understand quantum mechanics simple because it is a fascinating subject. At the philosophical level, quantum mechanics forces on us some quite bizarre notions about how the universe works. The direct consequences of these notions could influence systems we might engineer in the future, such as completely secure quantum encryption in data transmission, and quantum computers that can solve problems that are too hard for any known or conceivable classical computers. Certainly as we move forward in engineering, we can be sure that we will need more quantum mechanics, not less. Particles in quantum mechanics for engineering In quantum mechanics, there are many different elementary particles, with many of these having names ending in -on. These are grouped into families, which adds yet more names. Protons and neutrons are members of the baryon family, which in turn is a subset of the hadron family (which also contains another subsets, the mesons). Hadrons can be viewed as being composed of quarks. Electrons are members of the lepton family (which are not hadrons). Each of those families contains many members. There are also force carrier particles, of which the photon is the example we are most likely to encounter explicitly it can be thought of as carrying all the forces associated with the electromagnetic field. For the situations that we encounter in engineering, as far as fundamental particles are concerned, we will almost always need to consider only electrons, protons, and neutrons, which together effectively make up all the matter we work with, and with photons, the particles that make up light. Fortunately for us, the electron and proton are apparently stable particles, and the neutron is also stable in practice when inside the nuclei of the atoms we usually work with. 1 We will typically only have to think about electromagnetic fields and forces (and, of course, simple classical gravity if we dont want our creations to collapse under their own weight!).2 The larger family tree of the elementary particles and forces is summarized in the Appendix to this chapter. Fermions, bosons and spin For most of engineering, we do not have to consider the family attributes of the hadrons and leptons in any detail. There are two other sets of -ons that we will come across a lot, however, and that are very important from a practical point of view. Each of the elementary particles we have discussed above has a property called spin. We will denote the magnitude of the spin of the particle by some number s . Spin is an internal angular momentum that each particle possesses. We can think of it crudely as if each particle is like a spinning top, hence the name spin. Any such classical analogy cannot be taken very far, however. The magnitude of this spin is something that is intrinsic to each particle it cannot be changed without changing the particle to another
1

The neutron on its own is only stable for about 15 minutes or so, before undergoing beta decay into a proton, an electron and an electron antineutrino. Elementary particle physics also has to consider both the weak and the strong force, which are effective at distance scales much smaller than an atom.
2

EE41 Physics of Electrical Engineering

Chapter 6: Modern physics for electrical engineering kind of particle. It is quite distinct from the orbital angular momentum that an electron can have as it orbits a nucleus that angular momentum can change as the electron is moved from one orbital to another. The angular momentum of a mass m orbiting at a radius r around a point with a velocity of magnitude v is the product mvr , and so angular momentum has units of kg m2s-1, or, equivalently, J s (Jouleseconds). A constant that we will see again and again in quantum mechanics is Plancks constant, h
h

6.626 x 10-34 J s

It is also often found used in a slight different form, known as h bar

h / 2

1.054 x 10-34 J s

As we can see, Plancks constant happens to have the same dimensions as angular momentum. For the various elementary particles, it is found that each of them has a magnitude of angular momentum that is either a half integer number of s, or an integer number of s. It is convenient therefore to choose the value of spin s be a dimensionless number such that s is the magnitude of its spin angular momentum. Hence we divide all elementary particles up into two classes, bosons and fermions, as shown in Table 6.1. Bosons Spin Magnitude of angular momentum Common examples
s = 0, or 1, or 2, or

Fermions
s = 1/2, or 3/2, or 5/2, or

s = 0, or

, or 2 , or

s = / 2 , or 3 /2 , or 5 / 2 , or

photon, phonon (both with s = 1)

electron, proton, neutron (all with s = 1/2)

Table 6.1. Bosons and fermions. Here we have added one new particle in the list, the phonon. The phonon is a particle analogous to the photon, but for acoustic waves rather than electromagnetic waves. Its use is very common in solid state physics because the interaction of electrons with the vibrations going on all the time in the crystal lattice is one of the important scattering effects that limits how easily electrons can move through the crystal. Why do we go to the trouble of distinguishing these two new classes of particles? It turns out that fermions and bosons have dramatically different properties when we come to look at how many of them can be in a given state or mode. We can essentially only have one fermion in a give state or mode; this restriction is known as the Pauli exclusion principle, and is very important in the physics of materials, as we will discuss below. By contrast, we can have as many bosons in a given state or mode as we want. Essentially, in a slightly oversimplified statement 3 , we cannot have two electrons

We can actually have up to two electrons (or two protons, or two neutrons) in the same place, but they have to have opposite spin directions (one up and the other down). Technically, the two different spin

EE41 Physics of Electrical Engineering

Chapter 6: Modern physics for electrical engineering (or two protons, or two neutrons) in exactly the same place, but we can have as many photons in the same light beam as we want. This difference between fermions and bosons has profound consequences in particular for the statistics of these particles. We will return to this point later; it is extremely important in the quantum mechanical picture of matter and waves. Note that all the elementary particles of which ordinary matter is made (electrons, protons, and neutrons) are fermions, and the particles associated with the electromagnetic and acoustic waves we ordinarily encounter are bosons. Though the electron, proton and neutron are each fermions, combinations of even numbers of them behave like bosons because the magnitude of the combined spins is then an integer. Such even numbered combinations are sometimes called composite bosons. They do have the same statistics as other bosons. Such composite bosons are quite common, though their boson character is usually only important at high densities. Such high-density boson effects are of substantial interest in physics, giving rise to phenomena such as Bose-Einstein condensation, a special state of matter that can be seen at very low temperatures. Quasi-particles The reader might complain that, in introducing the phonon above, we are postulating some quantum mechanical particles that are not truly fundamental or elementary, and are merely confusing the issue of quantum mechanical particles. It is often the case, though, that, when we look at specific collections of particles, or when we look at the collective motion of a lot of particles, we find we have to look at that collective behavior quantum mechanically as well. When we do, we often find, very conveniently and very usefully, that this collective behavior can be described as if they we had new quantum mechanical particles. Such particles are called quasi-particles. Indeed, we could argue that protons and neutrons are not fundamental particles either, since they are made up out of collections of quarks, though, for historical reasons and by convention, we do not usually refer to protons and neutrons as quasi-particles. It is not even entirely clear if the particles we currently think of as elementary are not also made up out of other particles maybe all the particles we currently know are really quasiparticles, or at least composite particles of some kind. The quantum mechanics of quasi-particles is often very similar to that of the elementary particles such as the electron or the photon. This quasi-particle quantum mechanics is particularly important for engineering applications, where we are seldom concerned with many actual elementary particles other than electrons or photons. The bottom line, though, is that quasi-particles are very useful constructs, and occur very often in engineering applications of quantum mechanics. An acoustic wave is a behavior of a collection of atoms all linked to one another. When we look at the case of such waves in crystalline solids, like, for example, the semiconductors we use to make many devices, we find that the behavior of those waves can be described quantum mechanically in a way very similar to the quantum mechanical behavior of light. Hence, because we had the photon as part of the quantum mechanical description of light, we end up with a phonon as part of the description of acoustic waves
directions correspond to different states. The more precise statement is that we cannot have two fermions in the same state.

EE41 Physics of Electrical Engineering

Chapter 6: Modern physics for electrical engineering in solids. We can if we wish call this phonon a quasi-particle. Quasi-particles like this occur whenever we have some kind of mode of vibration. In the case of the phonon, this mode is like one of the possible notes that you could get if you hit a solid material. 4 Frequencies of phonons can be very high (e.g., 1013 Hz ) 5 when they correspond to modes where adjacent atoms are oscillating in opposite directions, and the quantum mechanical description is particularly important then. There are other examples like the phonon one that sometimes come up in solids. For example, think about the collective movement of a large number of free electrons in a solid. If we pull all of the electrons a little bit sideways, the electrostatic attraction of the positive charges on the nuclei that have now been left behind will tend to pull the electrons back. Rather like a mass on a spring, the whole body of electrons could then oscillate back and forth. This is known as a plasma resonance, and when we look at it quantum mechanically, we end up introducing another quasi-particle, the plasmon. Quasi-particles are important not only for describing oscillations. We encounter them also when thinking about particles associated with matter. A good classical example or analogy is a bubble in a glass of liquid. The bubble behaves as a particle, though there is essentially nothing inside it in that case it is essentially an the absence of material that makes it up. We will use a model very like this when we consider semiconductors. In semiconductors it is often very useful (and quite correct mathematically) to consider the absence of an electron as a quasi-particle, which we call a hole. In fact, even the electron itself as we see it in a crystalline material is really a quasi-particle. We can think of it like an electron, with charge e, but because of the interaction with the crystalline lattice of nuclei and other electrons, it is really more correctly a quasi-particle it does not behave as if it has the same mass as an ordinary free electron, for example. So, with the additional possibilities of quasi-particles, the bad news is that we have not finished naming quantum mechanical particles even if we restrict ourselves to areas that might touch engineering there are a lot more -ons that we might come across. The good news is that to deal with those we can very often use models and concepts that we have already figured out in other problems. What is so different about quantum mechanics - wave-particle duality So far in discussing quantum mechanics, we have introduced many particles, have given them some somewhat unusual properties. But we have not yet dealt with the way in which quantum mechanics is really so different from the classical view of the world. A hint at a key difference has been that we have said that forces (such as electromagnetism) can also be considered as being made up of particles (photons). Perhaps the single most dominant difference from classical physics that quantum mechanics introduces is the idea of wave-particle duality. We believed, based on many

For a simple example of modes of oscillation in a solid structure, try (gently) hitting your coffee mug or tea cup on the rim above the handle with a pen or pencil. You should here a fairly clear note. Now move round the rim about 1/8th of a turn (45) and hit again with the pen or pencil. You should be able to find a spot here that also gives a different but quite clear note (usually about a semitone higher, for those with a musical ear). In each case, you are exciting a different mode of oscillation of the cup or mug. You should be able to guess roughly what the vibration would look like in each case if you could see it.
5

About 35 octaves above middle C on the piano, for the musically inclined!

EE41 Physics of Electrical Engineering

Chapter 6: Modern physics for electrical engineering successes of 19th century physics (such as the diffraction of light and Maxwells equations of electromagnetism, published in 1873), that light was best represented as an electromagnetic wave. But since then we have found that light also behaves in some ways as if it were made up of particles (photons). More radically still, we find that objects like electrons, which we think of as particles, also behave in some ways as if they are waves. Quantum mechanics operates in this strange world in which all waves can behave as particles and all particles can behave as waves. We will now examine this idea and how we resolve some of the paradoxes that arise as a result. Early history of wave-particle duality Development of wave-particle duality for light The idea that light might be made of particles is an old one. In the 17th century, Newton, for example, advanced the corpuscular theory of light, in which light was made of particles that traveled in ray-like paths, in disagreement with Huygens, who advanced a wave theory of light. The wave theory began to dominate in the 19th century, following Fresnels success in explaining diffraction effects. Youngs slits Diffraction effects are basically phenomena we can see with waves that cannot be explained through the use of a ray picture. There are several well-known examples, but one particularly important one involves the passage of light through narrow slits. We will discuss the case of single slit later below. Here we will look at what happens if we have two very narrow slits. This is known as Youngs double slit experiment, after Thomas Young, who discovered this effect in the early 1800s. This is illustrated in Fig. 6.1. In this experiment, we shine a monochromatic (single color) light beam at the slits from the left, and look at the pattern formed on another screen, far away on the right. If the slits are very narrow, we can think of each of them as a point source of light, with circularly expanding waves coming out of each one.

slits

screen

d sin

ds

L
Fig. 6.1 Illustration of Youngs slits.

brightness on screen

The idea that each point on the surface of a wave can be considered as a source of circularly (or, in three dimensions, spherically) expanding waves is known as Huygens

EE41 Physics of Electrical Engineering

Chapter 6: Modern physics for electrical engineering principle after Christiaan Huygens, who published it in 1690. It is a good first model of diffraction, 6 and is enough for us to describe quite well what happens here. The slits are separated by a distance d. The screen at the right is very far away, so the mathematical lines drawn from the two slits to the screen very far away on the right are approximately parallel, and at angle to the horizontal. The difference in lengths of the two lines from the slits to the screen on the right is d sin . When this difference is an integer number of wavelengths, i.e.,
d sin = m

(6.1)

where m is an integer, we will have constructive interference, and a peak in the intensity. When the distance is

1 d sin = m + 2

(6.2)

we will have destructive interference, and a minimum in the intensity. Hence, as a function of angle , the intensity will oscillate, giving an interference pattern in the intensity as shown in the figure. For small angles, sin , and so the interference peaks occur at m / d . Also for small angles, x L , so for small angles, the separation s of the interference peaks is, approximately,
s L / d

(6.3)

One of the useful aspects of this experiment is that, for large distances L, the interference peaks can be spaced quite far apart, much larger than a wavelength, and hence enable us to see interference effects even though we cannot see distances as small as a wavelength itself. (Note, incidentally, that, for smaller separations of the slits, the separation between the interference peaks gets larger.) The existence of this interference pattern for light is one of the key experiments that demonstrates that light is a wave. Maxwells equations of electromagnetism were very successful in giving a detailed quantitative wave theory of light, and can accurately explain all the detailed behavior of the Youngs slits interference effects, for example. Black body radiation, the photoelectric effect, and the photon The particle theory became important again in the early 20th century. In 1900, Planck proposed some particle-like properties of light emission to explain something that mystified classical physics - the spectrum of light given off by hot bodies (so-called black body radiation), such as the red glow of molten glass or metals, the color of an ordinary tungsten light bulb, or the color of the sun. Note, incidentally, that quantum mechanics was started here by an attempt to explain an everyday phenomenon why do hot objects have the colors they have not for some obscure reason to do with very small or bizarre objects. Plancks proposal was formally the beginning of modern quantum mechanics.

It is not quite right, because in this simple form it would also predict backwards traveling waves, but if we only use it for forward waves it is actually quite a good model for diffraction.

EE41 Physics of Electrical Engineering

Chapter 6: Modern physics for electrical engineering

In 1905, Einstein proposed that light was actually made from particles, photons, with properties that agreed with Plancks particle-like emission, to explain something called the photo-electric effect. The photoelectric effect was first discovered in 1887 by Heinrich Hertz. When we shine light at a metal surface, electric charge can flow out of the surface. If we do this in a vacuum, and connect the metal surface and another electrode in a simple circuit, we can collect this current. A simple view of the experiment is shown in Fig. 6.2. There is nothing particularly surprising classically about the fact that we can liberate charge from a metal by shining light on it. After all, we are putting energy into the metal, and we could imagine that we were somehow boiling off charge, or burning it off, or something similar. In this particular experiment, however, we have two important variables we can adjust so that we can try to understand the process in more detail, and when we vary these, we see something that is very surprising and inexplicable from a classical viewpoint. The variables we can adjust are the frequency, , of the light, and the voltage, V, between the plates.

Metal plate

Light

Collecting plate

Flow of negative charge

V +

Fig. 6.2 Apparatus for the photoelectric effect. The first surprising result is that, as we progressively reduce the frequency of the light, below some particular cut-off frequency, o, we suddenly see no more current in the circuit, regardless of the voltage V, and regardless of the power of the light beam. We have no classical model to explain this phenomenon. Why would the light frequency matter much at all? Would it not just be the power in the light beam that would determine how much charge was emitted? Furthermore, when we are at a frequency above this cutoff, we continue to see charge emitted even as we decrease the power of the light beam, in an amount that remains proportional to the light beam power. This is very unlike a boiling or burning phenomenon, where we would expect that we have to be above some critical power density to get the phenomenon to happen, e.g., we would have to heat the material up to some critical temperature perhaps. When we are above the cut-off frequency, and are emitting charge, we can change the voltage V to see what happens to the charge collected. When we do so, for any given frequency above the cut-off frequency, we can stop the current entirely once we get to some particular stopping voltage, Vs. (Note that the voltage shown in Fig. 6.2 is set up
EE41 Physics of Electrical Engineering 8

Chapter 6: Modern physics for electrical engineering

with a polarity in a direction opposite to the one that would attract the electrons to the collecting plate.) Strangely enough, this stopping voltage depends linearly on the frequency of the light. Again, there is no classical explanation for this. By the time Planck and Einstein were thinking of these problems, the idea of electrons had been established, based on experiments by J. J. Thomson in 1897. Assuming now that the charge is being carried by electrons, we can interpret the stopping voltage in terms of the maximum kinetic energy of the emitted electrons. The potential energy of an electron that has gone uphill against a voltage of magnitude Vs is
P.E. = eVs

(6.4)

which we now interpret as the maximum kinetic energy, K.E.max, the electron could have had when it left the metal. When we do this experiment quantitatively, we find that
K .E.max = h

(6.5)

where is a quantity, known as the work function, that is different for different metals. We can understand this effect, therefore, if we propose that light comes in packets of energy, photons, of size h, and that there is some specific minimum energy required to extract an electron from a specific metal. This model works if we can think of the various energies involved as being like those in Fig. 6.3. 7

electron energy

photon energy, h

kinetic energy of emitted electron work function,

pool of electrons in metal

metal

vacuum

Fig. 6.3 Energy diagram for the photoelectric effect.

Possibly electrons could be pulled from deeper in the pool of available electrons, with correspondingly less kinetic energy. It is also true that the electrons need not be emitted with all of their kinetic energy corresponding to motion perpendicular to the surface. The photoelectric effect, however, by looking for the voltage at which the electron current ceases altogether, looks for the maximum possible kinetic energy in the direction perpendicular to the surface.

EE41 Physics of Electrical Engineering

Chapter 6: Modern physics for electrical engineering

This explanation in terms of quanta of light with energy h has stood the test of time. Aspects of this explanation of the photoelectric effect occur routinely with light. Nearly every photodetector we have today relies on the absorption of photons to release electrons into states where they can move and be collected as current, and nearly all such processes have a threshold frequency below which they do not take place. In fact, nearly all absorption of light has to be described by processes involving photons raising electrons from one energy level to another, with the photon energy corresponding to the separation of those energy levels. Hence we find we are forced to use both wave and particle pictures together to understand how electromagnetic radiation propagates and interacts with matter. The resulting modern picture of quantum optics, building on Maxwells equations and the concept of the photon, apparently successfully explains all the phenomena we see with light, and has wave-particle duality implicit in it from the beginning. Development of wave-particle duality for matter Just by our ordinary everyday experience of matter, at least of solids and possibly liquids, we are used to the idea that matter is made of particles of some kind. The idea that there might be essentially indivisible basic particles (the word atom comes from the Greek for indivisible) was postulated in the 5th century BC by Greek philosophers, most notably Democritus. At that time, of course, it was a philosophical notion only with no direct experimental evidence, but was advanced to explain the observable unchanging and apparently unchangeable nature of matter (chemical reactions such as burning notwithstanding). Such underlying unchangeability could be explained if matter were made of particles that themselves were unchangeable. The essential correctness that there was an underlying constancy to matter, at least under normal chemical reactions, was bolstered, for example, by Lavoisiers discovery in 1782 that weight was conserved in chemical reactions. The modern notion of atoms as the building blocks of chemistry was advanced by Dalton in ~ 1803, who also advanced the notion that chemical reactions proceeded with strict ratios between the constituents, consistent with the idea of atoms. The idea that we could describe phenomena such as gas pressure in terms of billiard-ball like atoms or molecules was put on a firm theoretical footing through work of people such Clausius and Maxwell in the 1850s (Bernoulli had apparently published such a theory in 1738, but it had been forgotten until then). The ideas of matter as particles was therefore firmly established, though not directly observed, in the 19th century. Thomsons discovery of the electron in 1897 was the first identified subatomic particle. Rutherfords work on scattering of alpha particles (helium atoms without their electrons) by atoms in 1909 exposed the first real structure of the atom, showing that the nucleus was very small. The investigation of subatomic particles by collisions in accelerators has continued ever since, and has been the main experimental tool for exposing the families of elementary particles discussed above. Certainly, therefore, by the early 20th century, the particle nature of matter was clear and quite well understood, even if detailed models were not available as to why atoms, molecules, liquids and solids had their specific forms and behavior. Ironically, to understand those forms and behavior, it was necessary to make the transition to viewing matter also as waves, because that enabled the construction of modern quantum mechanics, which in turn gives us the microscopic models necessary to understand atoms.
EE41 Physics of Electrical Engineering 10

Chapter 6: Modern physics for electrical engineering

The idea that matter might also be described as waves dates from 1923, with a proposal by de Broglie. He proposed that particles with mass (known as massive particles, even though they can be very light!), such as electrons or protons, behave as if they have a wavelength

h p

(6.6)

where p is the particles momentum, and h is Plancks constant. Note that he chose the momentum of the particle as determining the wavelength, not the velocity. For particles moving at some specific speed, heavier particles therefore have shorter wavelengths. This postulation is used by Schrdinger in his famous wave equation, which we will describe later. That equation led to one of the first major quantitative successes of quantum mechanics the explanation of the hydrogen atom. This postulate and Schrdingers equation also explain the phenomenon of electron diffraction. Electron diffraction was first seen around 1927 by Davisson and Germer, who shone beams of electrons at crystals of nickel. The result of such an experiment is that electrons are strongly reflected at specific angles, which can be understood quite easily from models of diffraction from periodic structures. This experiment was quite convincing that electrons did indeed have to be considered as waves. We will also look at a Youngs slit experiment for electrons below. Now let us look at just how matter and electromagnetism behave both as waves and particles.
Electrons, protons and neutrons as particles

When we think of a particle, at least in the classical sense, we mean some object, such as a particular grain of sand, a billiard ball or even a brick, that has particular and definite properties. Obvious examples of properties for a classical particle are weight, size and shape, and possibly other properties, such as a specific color, or a particular magnetization (if we are talking about some ferromagnetic material such as a piece of iron), or even possibly a particular amount of electric charge on it. Electrons (and protons and neutrons) are like such classical particles in several important ways. They each have a well-defined mass (~ 9.11 x 10-31 kg for an electron, ~ 1.67 x 1027 kg for protons and neutrons), a well-defined charge (~ -1.602 x 10-19 C for electrons, ~ +1.602 x 10-19 C for protons, and zero for neutrons), a well-defined spin, and a welldefined "magnetic moment" (~ 1.86 x 10-23 J/T for electrons, ~ 1.41 x 10-26 J/T for protons, ~ 9.65 x 10-27 J/T for neutrons). We can see the particle-like nature quite clearly if we look, for example, at electric currents. All currents in semiconductor electronic and optoelectronic devices are carried by electrons. Protons (usually in the form of ionized atoms) carry some of the current in plasmas (as in fluorescent light tubes) or in liquid solutions. In either case, if we look carefully enough, we can see that current is made up of these discrete units of charge. The fact that current does only flow in discrete units of charge gives rise to one of the fundamental noise processes in all electronic devices, a phenomenon called shot noise. We see the particle nature also very clearly if we examine chemistry. As we discussed above, it is quite clear to us now that ordinary matter is made of quite discrete
EE41 Physics of Electrical Engineering 11

Chapter 6: Modern physics for electrical engineering

atoms, which we know to be combinations of integer numbers of electrons, protons and neutrons. Magnetic effects are also very quantum mechanical in nature, and are connected also with the particle nature of electrons, protons and neutrons. Magnetic effects in materials occur both because of the magnetic fields that result from the currents of electrons in orbits around atoms, as well as from the intrinsic magnetic moments of the electrons, protons, and neutrons. (The strongest effects, such as the "ferromagnetism" of permanent magnets, come predominantly from the "spin" magnetic moments intrinsic to the particles.) So far, then, our idea of an electron, a proton, or a neutron is not so different from our classical idea of a particle. The specific properties may be somewhat different from those of macroscopic classical particles, but we have not yet had to give up any of our classical ideas of what a particle is. One important way in which such quantum mechanical particles differ from the classical particles we know is that the quantum particles of a given type are all absolutely identical to one another. When we look at sand, we know it is made up out of many different grains that are very similar in broad terms, but we also know that, if we looked at them under a microscope, we would find they were all slightly different. Similarly, the bricks that we use to build a house all look quite similar and have very similar sizes, but again a close inspection would show they were all quite different in detail, and we could, if we really wanted, tell them apart. But with quantum mechanical particles such as electrons, there is no way of telling "which" electron we are looking at - in fact, the question is quite meaningless. We might as well ask which dollar is which in our bank account that is a meaningless question. In this money analogy, classical particles are like dollar bills all quite distinguishable (in fact, each one has its own unique serial number), and quantum mechanical particles of a given type are like the dollars in our bank account. This "indistinguishability" is very important when we come to look at the details of the statistics of electrons (or protons, neutrons, photons, or any other such elementary particle).
Electrons, protons and neutrons as waves

As we mentioned above, we can see that electrons and other massive particles can behave as waves. We can take a beam of electrons, for example, and bounce it off of a crystal surface. If we let the reflected electrons land on a phosphorescent screen (i.e., just like the inside of a television cathode-ray tube), we will see a diffraction pattern that is just like what we would see if we bounced monochromatic (i.e., single wavelength) light off of a periodic surface, but with a wavelength that agrees with de Broglies hypothesis above (Eq. (6.6)). The brightness of the pattern on the screen would represent the square of the electron wave amplitude (actually, the modulus squared, since quantum mechanical waves are usually represented as complex numbers). The Youngs slits experiment discussed above for optics can also be performed for electrons. It was technically somewhat harder to do that specific experiment because it involves fabricating two slits very close together, or something closely equivalent to that,

EE41 Physics of Electrical Engineering

12

Chapter 6: Modern physics for electrical engineering

but that experiment has also since been done. An excellent example is the work by Akira Tonomura. The interference patterns from his work 8 are shown in Fig. 6.4. 9

Fig. 6.4 Interference patterns from a two slit interference experiment performed with electrons (A. Tonomura). Pictures (a) through (d) are taken at progressively later times on a detector screen that shows one white dot at each position where an electron landed. These show the progressive build up of an interference pattern.
Schrdingers wave equation

Given that we see wave effects, we can postulate an equation that describes them. For a monochromatic wave, we will have only one wavelength. Considering only one spatial direction for simplicity, we can postulate a simple wave equation, considering only the spatial parts of the wavefunction, of the Helmholtz equation form
d = k 2 dz 2

(6.7)

where we use the wavevector magnitude


k = 2 /

(6.8)

8 9

http://www.hqrd.hitachi.co.jp/em/doubleslit.cfm

One other remarkable aspect about this demonstration is that, deliberately, the current in the electron beam is so low that there is almost never more than one electron in the apparatus at once. Yet still the interference pattern builds up. This raises the question of how does an electron interfere with itself? Does it go through both slits at once? How would that be possible? This kind of question is a very basic one in quantum mechanics. The essential resolution of it is that it is meaningless to ask the question of which slit the electron went through! In a certain sense the electron does go through both slits. Any attempt to find out which slit the electron went through will actually destroy the interference pattern! The same bizarre behavior is also true for photons in a Youngs slit experiment. We can do such an experiment under conditions where there is only one photon in the apparatus at once. We will see a build-up of the interference pattern just as we did for electrons. The resolution of the paradox is the same for photons. It is meaningless to ask which slit the photon went through.

EE41 Physics of Electrical Engineering

13

Chapter 6: Modern physics for electrical engineering

This equation has solutions such as sin(kz), cos(kz), and exp(ikz) (and sin(-kz), cos(-kz), and exp(-ikz)), that all describe the spatial variation in a simple wave. Equivalently, given the empirical wavelength exhibited by the electrons

k = p/
and we can rewrite the wave equation (Eq. (6.7)) as

(6.9)

d = p 2 2 dz
2

(6.10)

where k = 2 / . If we now choose to divide both sides by 2mo , where, for the case of the electron, mo is the free electron rest mass
mo 9.11 10 31 kg

we obtain

d 2 p2 = 2mo dz 2 2mo
2

(6.11)

In classical mechanics, we know that the momentum p = mo v (with v as the velocity), and so we expect, as in classical mechanics

p2 1 mo v 2 kinetic energy of electron 2mo 2


and, in general, 10

(6.12)

Total energy (E )=Kinetic energy + Potential energy (V ( z ))


Hence, we can postulate that we can rewrite our wave equation (Eq. (6.11)) as

(6.13)

d 2 = ( E V ( z ) ) 2mo dz 2
2

(6.14)

or, in a slightly more standard way of writing this,


2 d2 + V ( z ) = E 2 2mo dz

(6.15)

Note that this potential energy V(z) is the energy that results from the physical position z of the particle. Though the symbol V is used, it does not refer to a voltage, though the potential energy can be (and often is) an electrostatic potential. It (and other energies in quantum mechanical problems) is often expressed in energy units of electron-volts, this being a convenient unit of energy, but it is always an energy, not a voltage.

10

EE41 Physics of Electrical Engineering

14

Chapter 6: Modern physics for electrical engineering

which is the time-independent Schrdinger equation for a particle of mass mo for onedimensional problems. It is worth emphasizing here that we have not derived Schrdingers equation we have merely postulated it. There is no way of deriving Schrdingers equation because there is nothing to derive it from; the whole of quantum mechanics is simply postulated, and the only justification for it (or any other scientific theory) is that it works! We can see that, if the potential V is constant, then we have solutions for any energy E we choose. We would have an equation

d 2 = ( E V ) 2mo dz 2
2

(6.16)

which has solutions of the form sin(kz), cos(kz), and exp(ikz) (and sin(-kz), cos(-kz), and exp(-ikz)), with

k=
i.e., the simple waves we would expect.
Energy levels

2mo ( E V )
2

(6.17)

Though we do see the wave effects with electrons and other particles, the most important consequences of this wave equation are not so much the wave diffraction phenomena. Perhaps much more important is that the Schrdinger wave equation leads to energy levels. These energy levels can be spaced by discrete amounts, and it is these discrete amounts or quanta that give quantum mechanics its name. Let us look at where these come from by looking at a very simple example that is easy to solve mathematically the particle in a box. Suppose that we have a box that has infinitely high walls, separated by a distance L. This potential is sketched in Fig. 6.5. Formally to solve this problem, for convenience we can choose the value of V in the well to be zero for simplicity (this is only a choice of energy origin, and makes no difference to the physical meaning of the final results). Because the potentials are infinitely high, but the particle's energy E is presumably finite, there can be no possibility of finding the particle in these regions outside the well; if there were, we would have a finite probability of the particle being in a region with infinite energy, which would mean the particle would have to have infinite average energy. Hence the wavefunction must be zero inside the walls of the well, and, to avoid a discontinuity in the wavefunction, we therefore reasonably ask that the wavefunction must also go to zero inside the well at the walls 11 . Formally putting this "infinite well" potential into the Schrdinger equation, Eq. (6.15), we are therefore now solving the equation

If the reader is bothered by the arguments here to justify these boundary conditions based on infinities (and is perhaps mathematically troubled by the discontinuities we are introducing in the wavefunction derivative), the reader can be assured that, if we take a well of finite depth, and solve that problem with boundary conditions that are more mathematically reasonable, the present infinite well results are recovered in the limit as the walls of the well are made arbitrarily high.

11

EE41 Physics of Electrical Engineering

15

Chapter 6: Modern physics for electrical engineering

d 2 ( z ) = E ( z ) 2m dz 2
2

(6.18)

within the well, subject to the boundary conditions

= 0;

z = 0, L

(6.19)

E1 Energy z
Fig. 6.5 Sketch of the potential energy for a well of thickness L, with infinitely high potential energy barriers on either side. The solution to Eq. (6.18) is very simple. The reader will likely recognize the form of the equation (6.18) as being the same as that for standing waves on a string. The general solution to this equation can be written

( z ) = A sin ( kz ) + B cos ( kz )

(6.20)

where A and B are constants, and k = 2mE / 2 . The requirement that the wavefunction goes to zero at z = 0 means that B = 0 . Because we are now left only with the sine part of (6.20), the requirement that the wavefunction goes to zero also at z = L then means that k = n / L , where n is an integer. The boundary conditions have forced us only to allow specific values of k this is the source of the quantization in this problem. Hence, we find that the solutions to this equation are, for the wave,

n ( z ) sin
with associated energies

n z L

(6.21)

n En = 2m L
2

(6.22)

We can restrict n to being a positive integer, i.e.,

n = 1, 2,
EE41 Physics of Electrical Engineering

(6.23)
16

Chapter 6: Modern physics for electrical engineering

for the following reasons. Since sin(a) = sin(a) for any real number a, the solutions with negative n are actually the same solutions as those with positive n (the sign of the wavefunction does not matter, and the energy is the same for positive and negative n). The solution with n = 0 is a trivial case; the wavefunction would be zero everywhere. If the wavefunction is zero everywhere, the particle is simply not anywhere, so the n = 0 solution can be discarded. The resulting energy levels and wavefunctions are sketched in Fig. 6.6. Again, to emphasize, the reader will notice that these are the same kind of wavefunctions we get from standing waves on a string or a transmission line 12 the mathematics of the spatial form of the standing waves is identical. As we discussed in the preceding Chapter, solutions such as these, with a specific set of allowed values of a parameter (here energy) and with a particular function solution associated with each such value, are called eigen solutions; the parameter value is called the eigenvalue and the function is called the eigenfunction. 13 Here, since the parameter is an energy, we can also call these eigenvalues the eigenenergies, and can refer to the eigenfunctions as the energy eigenfunctions.

energy n=3

wavefunction

n=2 n=1

L
Fig. 6.6 Sketch of the energy levels in an infinitely deep potential well and the associated wavefunctions. Now we have mathematically solved this problem. The question is, what does this solution mean physically? We started out here by considering the known fact that electrons behave in some ways like propagating waves, as shown by electron diffraction effects. We constructed a simple wave equation that could describe such effects for monochromatic (and hence monoenergetic) electrons. What we have now found is that, if we continue with this equation that assumes the particle has a well-defined energy and put that particle in a box, there are only discrete values of that energy possible, with specific wave functions associated with each such value of energy. We are now going beyond the wave-particle duality we discussed before. This problem is showing us our

12 13

To be precise, standing voltage waves on an electrical transmission line with short circuited ends.

Again, it is possible to have more than one eigenfunction associated with a given eigenvalue, a phenomenon known as degeneracy.

EE41 Physics of Electrical Engineering

17

Chapter 6: Modern physics for electrical engineering

first truly quantum behavior in the sense of the discreteness (or quantization) of the solutions and the quantum steps in energy between the different allowed states. Note that there is a minimum possible energy for the particle, which is above the energy of the classical "bottom" of the box. In this problem, the lowest energy corresponds to 2 n = 1, with the corresponding energy being E1 = ( 2 / 2m ) ( / L ) . This kind of minimum energy is sometimes called a "zero point" energy. We can use this simple example to get some sense of orders of magnitude in quantum mechanics. Suppose we confine an electron in a box that is 5 (0.5 nm) thick, a characteristic size for an atom or a unit cell in a crystal. Then the first allowed level for the electron is found at E1 = (
2

/ 2mo )( / 5 1010 ) 2.4 1019 J .


2

In practice in quantum mechanics, it is usually inconvenient to work with energies in Joules. A more useful practical unit is the electron-Volt (eV). An electron-Volt is the amount of energy acquired by an electron in moving through an electric potential change of 1 V. Since the magnitude of the electronic charge is

e 1.602 1019 C
and the energy associated with moving such a charge through an electrostatic potential change of U is e U , then one electron-volt (1 eV) corresponds to an energy 1.602 1019 J. With this practical choice of energy units, the first allowed level in our 5 wide well is 2.4 10-19 J 1.5 eV . The separation between the first and second allowed energies ( E2 E1 = 3E1 ) is ~ 4.5 eV, which is of the same magnitude as major energy separations between levels in an atom. (Of course, this one dimensional particlein-a-box model is hardly a good one for an atom, but it does give a sense of energy and size scales.) We could take the same approach here to a more realistic problem. A classic one in quantum mechanics is the hydrogen atom, which consists of an electron and a proton. The potential energy in that case is not a simple square one like we used above, but is the Coulomb potential energy between the negative electron charge and the positive proton charge. We would also have to solve the problem with a Schrdinger equation in three dimensions, and the mathematics is somewhat harder. The qualitative results are rather similar though for the bound states of the hydrogen atom. For the bound states of this system (i.e., where the electron does not have enough energy to get away from the proton, just like the moon cannot get away from the earth), we get only a discrete set of possible energies, which we can call the energy levels of the hydrogen atom. Associated with these energy levels are specific eigenfunctions, which in this case are the orbitals of the hydrogen atom, the electron clouds often familiar from chemistry. These eigenfunctions or orbitals are the states or modes for the electron waves in the hydrogen atom. As we go to more complicated atoms and collections of atoms, we still get only discrete possible values of energy for all of the bound states of the atom or atoms, and we get other, possibly more complicated, orbitals or modes.

EE41 Physics of Electrical Engineering

18

Chapter 6: Modern physics for electrical engineering

Size of an electron? A classical particle, such as our billiard ball, grain of sand, or brick, has a well-defined size and shape that is pretty much the same under all conditions. An important consequence of this wave nature of an electron is that it does not have a size. We can say that we are very likely to find an electron within a specific volume if we go and look for it, and quantum mechanics will allow us to calculate such probabilities for any given volume. The size of volume we have to include if we want to be fairly sure of finding the electron depends entirely on the situation, and is not an intrinsic property of an electron. The volume occupied by an electron in a hydrogen atom is quite different from that for an electron in a sodium atom, or an electron inside a cathode ray tube (e.g., television picture tube), or an electron inside a transistor. The electron does not have a hard core as far as we understand it, at least not on any size scale we will encounter. It is quite meaningless to ask how big an electron is. Thus the wave nature of the electron in quantum mechanics is forcing us to give up one of the attributes of a classical particle in this quantum world. Heisenbergs uncertainty principle Diffraction of waves from apertures One well-known phenomenon with waves is the diffraction of waves coming through an aperture. Waves coming through a hole that has a size much smaller than the wavelength are spread out over a very large angle. Waves coming through a hole whose size is comparable to or larger than the wavelength are more directional, and can become increasingly very directional as the hole is made larger. We know this quite well just from listening to speech, for example. When someone talking to you is facing you, you can here them clearly, including all the high frequency components, but when they face in the opposite direction, not only is the sound quieter overall, but the high frequency components are lost, and the speech becomes much more difficult to hear and to understand. The velocity of ordinary acoustic waves is about 340 m/s at sea level at room temperature, which means that a frequency of about 6.9 kHz has a wavelength roughly the size of an open mouth (~ 5 cm). Certainly only a small fraction of the sound at frequencies above this 6.9 kHz makes its way round behind the person speaking, and likely a large fraction of sounds in the kilohertz frequency range is relatively very weak to the listener who is behind the speaker. Anyone with a multi-loudspeaker audio system is also likely aware of this phenomenon of the greater directionality of high frequencies. Most multi-speaker systems have just one woofer of sub-woofer that handles all of the low frequencies 14 ; it is not necessary to have more than one because the low frequency sounds carry little or no directional information. Two or more small loudspeakers (sometimes known as tweeters), which handle only high frequencies, are placed at specific points in space to convey all the directional information in stereo or surround-sound systems.

To get much volume at low frequencies also requires relatively very large movements of the loudspeaker surface, and results in large and heavy loudspeakers for the low frequencies, hence the desire not to have to make all of the loudspeakers handle the low frequencies.

14

EE41 Physics of Electrical Engineering

19

Chapter 6: Modern physics for electrical engineering

From an approximate quantitative viewpoint, if we have a plane wave hitting an aperture of size d, most of the energy emitted on the other side of the aperture is emitted into a cone of angle

D / d

(6.24)

at least for d . This angle is known as the diffraction angle.15 (For smaller apertures, the wave spreads over even larger angles, though this simple formula for estimating the angular spread will break down.) Hence smaller apertures mean larger angles of spread (larger diffraction angles). This spread is illustrated in Fig. 6.7. Any wave that obeys a Helmholtz wave equation like Eq. (6.7), even when extended to cover 2 or 3 dimensions, will show this same diffraction phenomenon, and so we also expect it for our quantum mechanical waves for electrons (or protons or neutrons).
Screen Incoming plane waves Approximate cone angle of waves far from aperture

D / d D / d

Fig. 6.7. Illustration of diffraction of a wave though an aperture of size d. At large distances from the aperture, the wave approximately spreads out so that most of the wave is found within a cone of diffraction angle D. In the case of the electrons (or protons or neutrons), it will be more useful to think in terms of momentum. Now after the aperture, instead of having just a single direction of wave propagation, we have a range of angles. Consequently we have a range of possible directions for the electron momentum (we have to remember that momentum is a vector quantity, with direction as well as magnitude). The situation for a wave whose momentum direction lies along the edge of the diffraction cone is illustrated in Fig. 6.8. Now the component of p in the x-direction, px, is, at this characteristic angle for the cone, using the de Broglie wavelength from Eq. (6.6),
px = p sin D p D = p / d = ph / pd = h / d

(6.25)

We can understand where this diffraction angle comes from by a relatively simple argument. Consider the interference of the waves from the two extreme edges of the aperture. At an angle , one path will be longer than the other by an amount d sin , just as in the analysis of the Youngs slits experiment above. For small angles, we can approximate this by d. When this difference is half a wave, we must have destructive interference, so we will have destructive interference between at least these two extreme waves at an angle = /2d. Of course, these are the extreme waves, and technically we should add all the sources in the aperture. If we considered that a more typical pair of waves to interfere were those of the distance from the center of the slit on either side, i.e., spaced by d/2, we would recover our specific answer above.

15

EE41 Physics of Electrical Engineering

20

Chapter 6: Modern physics for electrical engineering

px p = h / d x z

D / d = h / pd
d

D / d = h / pd

Fig. 6.8. Illustration of momentum p corresponding to the direction at the edge of the diffraction cone, and the corresponding component, px, in the x-direction. Since the possible directions span the range from approximately -D to D, we can say that, approximately, px is the uncertainty in the momentum just to the right of the aperture. Correspondingly, the uncertainty in the position just to the right of the aperture, relative to the center of the aperture, is x = d/2. Hence, for this simple example here, we are finding a relation between these uncertainties
px x h / 2

(6.26)

This is an uncertainty relation. If we were to follow a more rigorous derivation, and formally consider px and x to be the standard deviations (in the statistical sense) of the momentum and position in the x-direction respectively, then it is possible to prove quite rigorously that
px x / 2

(6.27)

This relation is the best known example of the Heisenberg uncertainty principle. In this form, it states that a particle cannot simultaneously have both a well-defined position and a well-defined momentum (in the same direction). This follows quite naturally from the wave description, but it is a remarkable conclusion from a classical point of view, where we are used to thinking that a particle can have both a specific position and a specific momentum at the same time. 16

Students who have studied Fourier transforms will be familiar with the notion, for example, that a signal of a finite duration must also have a finite spread of frequencies. The mathematics of that relationship between frequency spread and time spread is exactly identical with the underlying mathematics of the uncertainty principle discussed here. In that case, the relationship is t 1/ 2 . In both cases the equality is reached for the case of Gaussian shaped pulses; the Fourier transform of a Gaussian shaped pulse is a Gaussian shaped distribution of frequencies, i.e., a pulse of the form t (t ) exp{[(t t ) / 2t ]2 } in time, round about some central time t and modulating a carrier of (angular) frequency , can equivalently be expressed as a set of amplitudes of frequency components of the form ( ) exp{[( ) / 2 ]2 } , where t = 1/ 2 .

16

EE41 Physics of Electrical Engineering

21

Chapter 6: Modern physics for electrical engineering Electromagnetic radiation as waves

We are already very familiar with electromagnetic radiation as waves. Since the discovery of Maxwells equations, which completely describe classical electromagnetics, we have understood the wave equations for electromagnetic waves. Those equations are technically vector equations because electric and magnetic fields are vectors, but in broad terms, for a monochromatic wave, the wave equations for the propagation of electromagnetic waves are similar to the Helmholtz equation we have been discussing. All the phenomena of diffraction discussed above, such as diffraction from a single slit for example, exist also for electromagnetic waves. The work on radio waves has given us a very deep understanding of electromagnetic waves. Optical waves obey all the same physics of propagation and diffraction, just with a shorter wavelength (e.g., 500 nm) and a higher frequency (e.g., 6 x 1014 Hz). In almost no situations do we ever have to use quantum mechanics to describe the propagation of optical waves the classical description can handle nearly all of that very well. Of course, there is a full quantum mechanical description of electromagnetism, and there are some non-classical states of light that are possible in the laboratory, but we seldom have to use the full quantum mechanical description just to discuss electromagnetic wave propagation.
Electromagnetic radiation as particles

The major area where the quantum mechanical nature of electromagnetism is important is in the generation and detection of light. Radio frequency electromagnetic radiation is typically generated by running currents through antennas, and is detected by the currents induced in antennas. But for frequencies in the THz (tera hertz - 1012 Hz) range and beyond, we almost always use quantum mechanical generation and detection. Relation between energy and frequency To understand these processes of generation and detection of light, we need first to understand a second very important relation in quantum mechanics, one that complements the de Broglie relation between momentum and wavelength. This is the relation between energy and frequency. Quite generally, the frequency (the Greek letter nu) corresponding to an energy difference E is

=
or equivalently (and more commonly)

E h

(6.28)

E = h

(6.29)

We already saw an example of this kind of relation between frequency and energy when we considered the energy of a photon above. Now we are generalizing this to any energy in quantum mechanics. Very often in quantum mechanics, we prefer to work with the angular frequency (the Greek letter omega) defined as

= 2
which leads to the alternative statement
EE41 Physics of Electrical Engineering

(6.30)

22

Chapter 6: Modern physics for electrical engineering


E =

(6.31)

Photons and their interaction with electron systems The idea of a photon is basically that the energy in a light beam of (angular) frequency is divided into discrete chunks of size , and these chunks are called photons. 17 We postulated a simple model for photons being absorbed by electrons in a metal when we considered the photoelectric effect above. Now that we have formally introduced the idea of energy levels for electrons, we can look at this more deeply. To understand the major way in which photons interact with matter, consider a simple system with an electron, in which there are two energy levels of interest to us, with energies E1 and E2, with a separation E2 E1 = E . Absorption Suppose now a photon comes along, with energy = E . 18 Then there is a probability that the photon will be absorbed, raising an electron from the first level to the second level, as illustrated in the absorption process in Fig. 6.9. We can often use this absorption process to make a detector; for example, we could arrange that, when we put the electron in the upper state, we are able to extract it and get a current of one electron out of the system, thereby getting an electrical signal corresponding to the detection of the photon. For example, we could arrange that, if the electron is in the upper state, it then has enough energy to get over some potential barrier, and fall out to be collected. This is exactly the kind of process we discussed above for the photoelectric effect. We can also conveniently use it inside semiconductor photodiodes, which are much more practical than vacuum devices. We will discuss later how photodiodes work based on this kind of absorption. Spontaneous and stimulated emission The other kind of process we need to understand is how light is emitted. Basically, an electron in the upper state can fall to the lower state, emitting a photon of energy corresponding to the separation of the states. There are actually two flavors of this emission process. The most common process in everyday life is spontaneous emission (see Fig. 6.9). In this case, if the system is in the upper state, it will spontaneously decay at random, with a characteristic lifetime, but without any external stimulus. Ordinary light bulbs and light emitting diodes work predominantly this way. The resulting light is also generally emitted in random directions. The second process is stimulated emission (see Fig. 6.9). In this case, if there is already a photon present of the right frequency, and the electron is in the upper state, then it can stimulate the emission of another photon, leading to two photons coming out of the system. The important difference here is that the radiation associated with this second photon has exactly the same frequency and

17 18

The photon was invented by Einstein.

The energy matching does not have to be absolutely exact because there is always some width or uncertainty in the exact value of the energies associated with the levels for a variety of reasons.

EE41 Physics of Electrical Engineering

23

Chapter 6: Modern physics for electrical engineering

direction as the first photon, or, equivalently, that the second photon is emitted into the same mode as the first photon.
Absorption before E2 Incident photon after

= E
E1

Spontaneous emission before E2 Emitted photon + after

E = E2 E1

= E

E1

Stimulated emission before E2 Incident photon after Incident + emitted photons +

= E
E1

= E

Fig. 6.9. Illustration of the three processes of absorption, spontaneous emission, and stimulated emission, with photons and a system with two energy levels. Lasers This stimulated emission process does go on to a very small extent all the time in any light emitter. In the laser, 19 however, we do our best to make this stimulated emission process dominate completely. A laser consists of two things a gain medium, which at

19

Laser in a acronym for Light Amplification by Stimulated Emission of Radiation.

EE41 Physics of Electrical Engineering

24

Chapter 6: Modern physics for electrical engineering

its simplest would consist of many different electron systems, each with energy levels separated by approximately the same energy, E, and a resonator that bounces some of the light back through the gain medium, often multiple times. We have to arrange for two things to get a laser to work. 1. We have to get stimulated emission to be a more likely process than absorption. This means that we have to have more of the electron systems in their upper state than in their lower state, a condition known as an inverted population. In practice, this means we have to have some way of continually clearing electrons out of the lower state and pumping the electrons back into the upper state. Usually this is done electrically, though sometimes we use another light source to put the electrons in the upper state again. 2. We have to get stimulated emission to dominate over spontaneous emission. This will happen if we have a large number of identical photons (i.e., ones in the same mode) always present with the right energy. This is the function of the resonator to keep building up the number of identical photons by reflecting them back through the gain medium. The result of both of these processes together is a laser beam, with a very well defined frequency and direction. A simple view of a laser is shown in Fig. 6.10.
gain medium laser beam

mirror

mirror (partially reflecting)

Fig. 6.10. Simple view of a laser, showing a gain medium with an inverted population, and mirrors to reflect many of the photons back through the gain medium. A partially reflecting mirror at one end allows some of the photons to escape, giving the laser beam. Incidentally, laser beams can be what is known as diffraction limited, which means essentially that the laser beam diverges as little as possible given the lateral size of the beam, and this is a consequence of the photons all being in the same mode. This actually is identical to saying that the momentum and position of the photons in the lateral direction are known to the limit imposed by the uncertainty principle. 20 A key consequence of the beam being diffraction limited is that it can be focused to the smallest possible point, allowing very high intensities with laser beams, which in turn allows them to be used for welding and cutting metal, for example. This diffraction-limited property

Many lasers actually produce beams with a Gaussian cross-sectional intensity distribution, which leads to the optimum diffraction limit possible because the Gaussian is the form that reaches the limit of the uncertainty principle.

20

EE41 Physics of Electrical Engineering

25

Chapter 6: Modern physics for electrical engineering

also allows laser light to be coupled very efficiently into optical fibers (especially single mode optical fibers). By contrast, other light sources are generally not diffraction limited, having a much larger angle of light beam than required by diffraction for their size. As a result, they cannot be efficiently focused to very small spots, and their light cannot be efficiently coupled into the single mode fibers we use for long distance communications. Size of a photon? Just like the electron, the photon also does not have an intrinsic size. It is possible to arrange to confine photons within some volume of space, just as we can arrange to confine electrons within some volume of space, but that volume is something defined by whatever is doing the confining it is not an intrinsic property of a photon.
Consequences of fermion and boson character

Bosons The discussion of stimulated emission and lasers above points out a key point about photons, which is associated with the fact that photons are bosons. We can have as many photons as we want in a given light beam, even one that is diffraction limited. Unlike fermions, there is no Pauli exclusion effect. In fact, stimulated emission shows that in some circumstances photons even prefer to be in exactly the same light beam. A more rigorous notation for this is to say that we can have as many bosons as we want in a given mode. As we discussed in the preceding Chapter, modes are mathematical functions that are quite distinct 21 from one another, and that, in combination, can be used to describe any function we want. For example, we can describe any wave as a combination of plane waves going in different directions. There are many other possible sets of functions or modes that we could use, including those that arise naturally from the solutions of certain kinds of problems, such as the modes of an optical fiber or of a laser resonator. Regardless of which specific modes we are talking about, we can have arbitrary numbers of bosons in a mode. Fermions and the Pauli exclusion principle We mentioned briefly above the Pauli exclusion principle, that we can only have one fermion in a give state or mode. The constituents of ordinary matter, such as electrons, protons, and neutrons, are all fermions, as we discussed above. The Pauli exclusion principle for these has major consequences without it, objects as we know them simply would not exist! The major specific consequence of the Pauli exclusion principle that matters to us here is its effect on electrons and the states they can occupy. The fact that we cannot have two electrons in the same state means that, as we add electrons, they have to go into different states, usually ones with successively higher energies. This principle allows us to understand the electronic structure of atoms and molecules; electrons go into different orbitals (the specific name given to the modes or states when we are talking about atoms or molecules) as we add more to the atom or molecule. The structure of the periodic table of the elements is largely determined by this principle. The noble gasses

Technically, as discussed above, functions that we can use as sets of modes must all be orthogonal to one another, in the sense that, for two different such functions, f1(x) and f2(x), f1* ( x) f 2 ( x)dx = 0

21

EE41 Physics of Electrical Engineering

26

Chapter 6: Modern physics for electrical engineering

(e.g., argon, neon, krypton) have essentially no chemical reactivity because the outer set of orbitals are all filled with electrons, for example. This principle is also very important as we look at the behavior of electrons in solids. We will look at solid state physics in the next section. That physics is a major enabler for most of the electronic and many of the optical devices that we use today.
Summary Elementary particles

Elementary particles divide into families, with protons and neutrons being in the baryon subset of the hadron family, electrons being in the lepton family, and photons being in the force carrier family.
Plancks constant
h

6.626 x 10-34 J s

It is also often found used in a slight different form, known as h bar

h / 2
Spin

1.054 x 10-34 J s

All elementary particles possess a spin s, which has either a half integer magnitude or an integer magnitude. The spin is associated with an angular momentum s . The magnitude of the spin of a particle is an intrinsic property of the particle, and cannot be changed without changing the particle into another kind of particle.
Fermions and bosons

Bosons have integer spin. Fermions have half integer spin. Electrons, protons and neutrons are all fermions. Photons are bosons.
Quasi-particles

Quasi-particles behave like particles, though they are associated with the collective motion of other particles. An example is a phonon, which is the analog for acoustic vibrations of the photon for electromagnetic vibrations. The phonon describes the collective motion of many atoms in a solid.
Wave-particle duality

In quantum mechanics, particles can behave as waves, and waves can be described as particles, with both the particle and wave aspects coexisting.
Youngs slits

Youngs slits are two narrow slits in a screen, with a wave being incident on the slits from one side. An interference pattern is formed on a screen placed at a large distance from the other side of the screen. The spacing of the interference peaks is approximately
s L / d

(6.32)

where L is the distance from the screen to the slits, is the wavelength of the waves, and d is the separation of the slits.
EE41 Physics of Electrical Engineering 27

Chapter 6: Modern physics for electrical engineering Photon

A photon is a particle of light, and it has energy


E = h =

where is the frequency, and is the angular frequency ( = / 2 )


De Broglies hypothesis

For particles with mass, they have a corresponding wavelength

=
where p is the particles momentum.

h p

(6.33)

Schrdingers time-independent wave equation

For problems in one dimension


2 d2 + V ( z ) = E 2 2mo dz

(6.34)

where E is the total energy, and V(z) is the potential energy as a function of position z.
Particle in a box

For a particle in a one-dimensional box of length L, the allowed energy eigenvalues (or energy levels), relative to the energy of the bottom of the box are

n En = 2m L
2

(6.35)

with

n = 1, 2,
with associated eigenfunctions for the waves

(6.36)

n ( z ) sin
Diffraction angle for waves from a slit

n z L

(6.37)

D / d
for a slit of width d and a wavelength .
Heisenbergs uncertainty principle

(6.38)

The principle states, mathematically, that the uncertainty in position x and momentum px are related by
px x / 2

(6.39)
28

EE41 Physics of Electrical Engineering

Chapter 6: Modern physics for electrical engineering

which can be taken to mean that a particle cannot simultaneously have both a welldefined position and a well-defined momentum. Diffraction is an example of the consequences of the Heisenberg uncertainty principle.
Relation between energy and frequency in quantum mechanics

In general, an energy difference is associated with a frequency through


E = h

(6.40)

Quantum mechanical absorption of photons

For two energy levels with energies E1 and E2, with a separation E2 E1 = E , absorption of a photon of energy h = E2 E1 can raise the system from level 1 to level 2.
Spontaneous emission of photons

For a system in level 2, it can spontaneously emit a photon of energy h = E2 E1 , and fall to level 1. This photon will generally be emitted in a random direction, and at a random time. This is the dominant process in ordinary light bulbs and sunlight.
Stimulated emission of photons

For a specific electromagnetic mode of frequency , when there is one or more photons already present in that mode, and there is a system in level 2, with an energy separation of h = E2 E1 between levels 1 and 2, a photon can be stimulated to be emitted into this specific mode, with the system falling to level 1. This is the dominant process in lasers.
Laser

A laser consists of a gain medium and some kind of cavity. The gain medium needs to have an inverted population, e.g., more electron systems in their upper state than in their lower state. The cavity reflects the light in the cavity to increase the number of photons in a given mode (or small collection of modes). Light is emitted as a result of stimulated emission into the given mode (or small collection of modes).
Diffraction limited light

Light is diffraction limited if it can be focused to the smallest spot allowed by the laws of diffraction. Laser beams can have this property.
Pauli exclusion principle

There can be only one fermion in a given quantum mechanical state (or mode). This leads, among other things, to the progressive filling of orbitals in atoms, which in turn gives rise to chemical properties as we know them.

EE41 Physics of Electrical Engineering

29

Chapter 6 Appendix: Elementary particles and forces Chapter 6 Appendix: Elementary particles and forces

Quantum mechanics is a broad and fascinating subject, and, if we head off into the world of fundamental physics, there is a large set of elementary particles and forces. For completeness here, we will briefly introduce some of the terminology of elementary particles and forces (often described in terms of force fields or, simply, fields). The names of many of these particles and families of particles end in -on. Most of these we will not in practice have to deal with in engineering, but there are a few that are very important to us, and a brief discussion of all of them will help clarify the situation. The model we will briefly discuss here is known as the standard model in elementary particle physics. This has been the accepted basic model of elementary particles since about the late 1970s. It still leaves many open questions, but seems at least to describe the chemistry of elementary particles in the same way that electrons, protons and neutrons describe ordinary chemistry.

Quarks up
e.g., charge + 2e/3

down
e.g., charge - e/3

charm

strange

top

bottom

plus antiparticle versions of each quark with, e.g., opposite charge

Hadrons Antibaryons
made from three antiquarks

Baryons Proton
made from two up quarks and one down quark made from three quarks

Mesons Pion
made from an up quark and a down antiquark made from a quark and an antiquark

Neutron
made from one up quark and two down quarks

Fig. 6A.1. Family tree of particles in the standard model. There are many other properties of the various quarks. The charge of the up and down quarks is shown as an example property. The quarks form into hadrons (isolated quarks are not apparently found), and all hadrons fit within the three families of baryons, antibaryons and mesons. There are many particles in each of these families, nearly all of which are unstable. The proton and the neutron of ordinary matter are in the baryon family.

EE41 Physics of Electrical Engineering

30

Chapter 6 Appendix: Elementary particles and forces Hadrons, baryons, mesons and quarks

All protons and neutrons are thought to be made up of three quarks each. Quarks come in several different types (up, down, charm, strange, top (or truth), and bottom (or beauty)), though quarks apparently cannot be separately isolated. The family of particles made from three quarks are called baryons, so protons and neutrons are baryons. There are also particles called mesons, made from two quarks, of which the so-called pions are the most common example, though none of the mesons are stable. Taken together, the particles made from quarks, i.e., the baryons and mesons, form the family of hadrons. Fig. 6A.1Fig. 6A.1 shows a family tree of particles in the family of matter made from quarks.
Leptons electron electron neutrino muon muon neutrino tau tau neutrino

plus antiparticle versions of each lepton, with opposite charge

Fig. 6A.2. Lepton family. The electron, muon, and tau each have e charge, and their antiparticles have +e charge. The antiparticle of the electron is conventionally called the positron (rather than the anti-electron). The neutrinos have no charge. Whether they have any mass at all is an open question at the time of writing.
Leptons

Electrons are the best known member of the family of leptons. Leptons are not made from quarks. The lepton family also includes the neutrinos. There is one neutrino corresponding to each of the electron, the muon and the tau. When a neutron decays by radioactive beta decay 22 , it turns into an electron, a proton, and a neutrino 23 . The other members of the lepton family are the muon, and the tau, both of which are much heavier than the electron and are unstable. Each of the muon and the tau has its own neutrino. Fig. 6A.2 shows the family of lepton particles.
Forces and force carriers

There are four known forces between elementary particles, the gravitational force, the electromagnetic force, the strong force and the weak force. The gravitational force acts between all bodies with mass. The electromagnetic force acts between all bodies with charge or magnetic moments. The weak force acts on all particles, and the strong force acts between matter made from quarks (hadrons). Both the strong and weak forces fall off much faster with distance than the gravitational or electromagnetic forces, so are typically only important inside nuclei or in the scattering or decay of elementary particles. The weak force is apparently the one involved in all interactions involving

22

Note that, although the neutron turns into an electron, a proton, and a neutrino under beta decay, it is not correct to say that a neutron is made up from an electron, a proton and a neutrino. It is made up from one up quark and two down quarks. This can be verified in certain kinds of scattering experiments with particle accelerators, where direct evidence for the quarks in the neutron can be seen.

Modern terminology actually requires that we refer to the neutrino in this particular decay as an electron antineutrino if we want to be more precise.

23

EE41 Physics of Electrical Engineering

31

Chapter 6 Appendix: Elementary particles and forces

neutrinos. Each of these forces has a particle of particles that carry the force in the particle view of such forces. For electromagnetism, the force carrier is the photon, which has no mass. For gravity, the force carrier is thought to be the graviton, though it has yet to be seen. For the strong force, the force carriers are thought to be gluons, the particles that carry the forces between quarks. (There are thought to be eight such gluons.) For the weak force, there are three carriers, the W+, the W-, and the Zo.
Particles and antiparticles

Many of the elementary particles have antiparticles. The antiparticles are in many ways like a mirror image of the particles. For example, the positron (or antielectron) is just like the electron, but has a positive charge rather than a negative charge. The antiparticle of an antiparticle is the particle itself; e.g., the antiparticle of the positron is the electron. Similar lepton antiparticles exist for the muon and the tau. There are also antiparticles for the proton (the antiproton, with negative charge), and the neutron (the antineutron). There are antiparticle versions of each of the quarks, so for any hadron, there is an antihadron composed of the complementary antiquarks. A particle and its antiparticle have a remarkable property, which is that, when they meet as a pair, they can completely annihilate one another, turning into energy (or into sets of completely different particles). There is similarly an opposite possibility that a particleantiparticle pair can be created from energy. These particle-antiparticle interactions might seem rather exotic, and in everyday life, with such fundamental particles, they are relatively unusual. There is, however, an important analogy that we do come across very often in optics, especially with semiconductors. We will discuss below that the absence of an electron (with its usual negative charge) in a semiconductor can often be usefully considered as the presence of a positively charged hole. This hole behaves very much like the antiparticle of the electron. Absorbing a photon in a semiconductor can create an electron-hole pair, and an electron and hole meeting one another can annihilate (the electron fills up the hole), creating a photon in the process. This is in fact the most correct way to look at optical absorption and emission in semiconductors.
Elementary particles in the world around us

Nearly all elementary particles other than the proton, neutron and electron, are not found naturally in any significant numbers at the earths surface. Many are created inside stars, and cosmic rays (fragments of nuclei, such as protons, going at very high speeds in space) do generate some as they hit the atoms of the earths atmosphere. A few such elementary particles reach the earths surface; for example, muons, which are like heavy versions of the electron, arrive at the earths surface at a rate of about 1 per sq. cm. per minute. Otherwise, particle accelerators are required to generate most elementary particles on earth. Nearly all such elementary particles are not stable for any significant period of time that is, they turn into other particles. We certainly therefore cannot make anything out of them from a practical point of view, and, with a few exceptions, they are not of any practical relevance in devices, though the byproducts of cosmic rays (mostly very energetic neutrons) can cause errors in semiconductor memory, for example, likely by freeing protons and electrons in collisions with the atoms in the devices. One other elementary particle, the neutrino, does arrive at the earth from the sun in very large numbers, at a rate of roughly 100 billion per sq. cm. per second! It interacts extremely weakly with matter, however a typical solar neutrino might have to travel an average
EE41 Physics of Electrical Engineering 32

Chapter 6 Appendix: Elementary particles and forces

distance of ~ 3 x 1016 m through ordinary matter to be absorbed and so it is difficult to think of any direct use of neutrinos for engineering. Neutrinos may be very important for cosmology, however, and the issue of whether neutrinos have mass is an extremely important topic for understanding the universe.

EE41 Physics of Electrical Engineering

33

Chapter 7 Semiconductor physics Chapter 7 Semiconductor physics A very large fraction of the devices we use to process and communicate information, such as diodes, transistors, photodetectors, and semiconductor lasers, use so-called semiconductor materials. The vast majority of these semiconductor materials are crystals. Here we will introduce some of the key features and processes in these materials that we will need for a basic understanding of how such devices work. Crystalline materials Many simple materials will form themselves in crystalline structures of atoms. Crystals have several advantages as materials. If we are careful about the environment and manner in which we grow crystals, they come out essentially exactly the same every time, so they are very predictable materials. The regularity of their structure means that, even though there may be a vast number of atoms in a piece of crystal, we can have relatively simple models by which we can understand their properties, such as their electronic or optical behavior. With semiconductors in particular, it is also often the case in practice that introducing small amounts of controlled impurities in the structure enables us to get some very precise and important kinds of control over the detailed electronic properties of the material.

Fig. 7.1. Example of a two-dimensional rectangular crystal lattice. A crystal in a mathematical sense is a structure that is made up by stacking identical bricks together in a regular fashion in such a way that they fill up all space. Essentially any repeating pattern, such as a wall paper pattern, is crystalline in this way. A simple two-dimensional lattice is shown in Fig. 7.1. We can think of this as a collection of rectangular bricks, or in the terminology of crystals, rectangular unit cells. As we can see, stacked together, these unit cells will fill all space (in two dimensions), and so this structure fits the definition of a crystal. We can also view the crystal as a lattice of points (the black dots), connected by lines to guide the eye, and, once we start thinking about a crystal being made up of atoms that are chemically bonded to one another, this is the view we tend to take, with the lines representing, in a loose way, the chemical bonds between the atoms.

EE41 Physics of Electrical Engineering

Chapter 7 Semiconductor physics The three-dimensional crystals are slightly more difficult to visualize, but they follow the same rules. In semiconductor electronic and optoelectronic devices, nearly all of them are based on one of the closely related lattices shown in Fig. 7.2. The diamond lattice is the crystal lattice for silicon and for germanium. The zinc blende 1 lattice is the lattice for the most of the so-called compound semiconductor materials based on III-V materials. 2 III-V compound semiconductors are materials made from a combination of elements from Group III of the periodic table (e.g., gallium, aluminum, indium) and Group V of the periodic table (e.g., arsenic, phosphorus, antimony, nitrogen). Typical IIIV semiconductors include gallium arsenide and indium phosphide. Both the diamond and zinc-blende lattices are based on atoms forming four chemical bonds, equally spaced in angle, with their neighbors. Silicon and germanium are both Group IV elements, which means four of their outer eight electron orbitals are filled. This fact makes them particularly want to bond with four other atoms. Carbon is also a group IV element, and when it crystallizes in this form, it is called diamond (hence the name of the lattice structure). 3 In these lattices, the four atoms connected to a given atom form the corners of what is called a (regular) tetragon if we joined these four atoms with lines, they form a structure with four equilateral triangle sides, all of equal size.

(a)

(b)

Fig. 7.2. (a) Diamond lattice. (b) Zinc blende lattice. A close look at the diamond lattice shows that, somewhat surprisingly, a cubic structure emerges from these sets of bonds. The cube is indicated by the dashed lines (which are not representing chemical bonds). If we look carefully, we see that there is an atom on each corner of the cube, and one in the middle of each face. A structure consisting of sets of atoms arranged like that is called a face-centered cubic lattice. There are also other

1 2

Zinc blende is zinc sulphide in a cubic crystalline form.

One exception is gallium nitride, and some of its alloys, which can crystallize in a so-called wurtzite structure. The wurtzite lattice is like the zinc-blende lattice, but in one direction the alternate planes are rotated by 60 degrees compared to those of the zinc-blende lattice. Gallium nitride is useful in short wavelength (e.g., blue) light emitting diodes, for example. Carbon can also crystallize in another completely different form, graphite, which is most common as the lead in a pencil. The mechanical properties of graphite are completely different from diamond; graphite is quite soft, and is a good lubricant, whereas diamond is extremely hard, and is a good abrasive!
3

EE41 Physics of Electrical Engineering

Chapter 7 Semiconductor physics atoms that do not lie on the corners or the faces. If we were to continue this lattice by adding more atoms in their crystalline positions in adjacent unit cells, we would see that these other atoms also lie on a face-centered cubic lattice that is simply displaced by one chemical bond in space. Thus the diamond lattice is actually made from two interlocking face-centered cubic lattices (or sometimes described as interlocking sublattices). In the zinc blende form of lattice, one of the kinds of atoms (e.g., arsenic, represented by the larger balls in the figure) lies on one of these face-centered cubic sublattices, and the other (e.g., gallium, represented by the smaller balls in the figure) lies on the other facecentered cubic sublattice. Though the Group III element only has 3 electrons in its outer shell, and the Group V has 5, they share these between the atoms to make four bonds for each atom. An important additional set of materials, especially for optoelectronic devices, is the set of III-V alloy materials. These are materials such as indium gallium arsenide, or aluminum gallium arsenide. Materials with three elements in them are known as ternary alloys. In these example cases, the Group III face-centered cubic sublattice has a usually random distribution of, e.g., indium and gallium atoms on it. It is also possible to extend this to choices of two atoms for each of the sublattices, as in indium gallium arsenide phosphide (which, because it has four components, is called a quaternary alloy), or three different choices for one lattice, as in indium gallium aluminum arsenide (also a quaternary alloy). These alloys give a great deal of flexibility in the design of materials with exactly the desired properties, and they are used extensively in optoelectronic devices. Changing the alloy composition can change the wavelength or color of light emitted by a light-emitting diode or semiconductor laser, for example. Of course, not all the materials we use in making devices are crystalline. Usually, we take care to make the semiconductors crystalline because we need the best, most predictable, and precisely controlled performance from them. Other materials also used in making devices, such as metals for conductors and oxides for insulators, do not have to have precise crystalline forms to operate successfully, and we can be more relaxed in the way we make them. Materials where we are not too concerned about a precise crystalline structure can often be made by simple, flexible techniques, such as evaporation onto the surface of interest. Growth of crystals usually requires carefully controlled temperatures, and also crystals can typically only be grown uniformly and precisely if they are being grown on some substrate or seed that itself is crystalline with the same separation between the atoms; such growth is called epitaxial growth, meaning that the growing surface takes on the order of what lies beneath it (epi meaning surface, and the tax.. part being as in, for example, taxonomy, meaning an ordering). Formation of bands in solid materials We discussed above that electrons in atoms have specific energy levels they can occupy. For every atom of a give type, those levels are, of course the same. But what happens to the energy levels when we start to bring two or more atoms closer together, as must happen as we form a solid? It is quite general that, when we bring two identical systems together, and start to allow them to interact, the energy levels split. That is, where in the separate systems we had two identical energy levels, in the interacting system, we have two slightly different energy levels of the joint system. This kind of coupling and splitting is common in a EE41 Physics of Electrical Engineering 3

Chapter 7 Semiconductor physics broad range of classical and quantum systems. We saw this behavior above when we considered coupled oscillators, such as two masses and three springs, with the center spring coupling the motion of the two masses. There are two stable modes of oscillating system. In one of these modes, the two masses move exactly in synchronism and in phase with identical amplitudes. In the other of these modes, the two masses move in exactly opposite directions, with equal and opposite motions. This second mode is at a higher frequency because there is more restoring force pulling each mass back towards the center (the force from the spring being stretched between the masses). If the center spring is very weak or is not there at all, we have two identical oscillators with identical frequencies. If we add the coupling spring between them, we get one lower frequency and one higher frequency. This kind of coupled oscillator or coupled resonator problem is common throughout physics. 4 If we increase the number of identical systems we couple, the splitting of the levels increases. In fact, for N identical coupled systems, each of the original levels will be split into up to N distinct ones 5 , just as we saw with the case of many masses coupled together by a tensioned string. As we bring the atoms together to form a solid, therefore, the isolated atomic energy levels turn into bands of allowed energies, each of which has N distinct possible electron states. The detail of the band structure can be quite complicated, but the simple principle of the creation of a band of N states discussed here still holds. It is also possible that the bands created from different atomic levels will overlap. Conduction and valence bands We can imagine conceptually (though not actually) that we construct the band structure mathematically by solving for the energies of all the electron states, and then we start pouring electrons into it. The Pauli exclusion principle tells us that we can only have no more than one electron in each electron state. Hence, the bands will progressively fill up as we pour the electrons in. The most interesting bands to us are the highest band that is full or partially so, and the first band that is empty or partially so. In practice, for most optical or electrical properties that we are going to engineer, the other full bands do not matter much, and neither do the other, higher-energy empty bands. 6

A specific example practical example is a well-known acoustic problem in string instruments, especially the cello (violoncello). When the cellist tries to play a specific note, usually around F below middle C, the cello may instead play only a note a bit above or a bit below the desired pitch, and may oscillate between the two of them, producing what is called the wolf note. The underlying problem here is that there are two coupled oscillators at approximately the same frequency, one being the resonance of the string, and the second being a resonance of the cello body. The coupling of these resonances leads to the splitting of the desired note into two possibilities, neither of which is the desired note. (The oscillation back and forward between the two is a consequence of the rather complicated, and nonlinear, interaction of the bow with the string.) It is not necessary that all the energy levels are different, because sometimes, as a result of symmetry for example, there are two or more different states with the same energy, just like the 3 P orbitals in a hydrogen atom all have the same energy. There is always only a finite number of states with the exactly the same energy, however.

These other bands are quite important, however, for giving rise to some physical properties, including for example, major contributions to the refractive index.

EE41 Physics of Electrical Engineering

Chapter 7 Semiconductor physics The highest completely or very nearly full band is called the valence band. The lowest band that is empty or only fractionally full is called the conduction band. Sometimes the various bands actually overlap, but we show a simple case in Fig. 7.3 in which there is a clear separation or band gap between the valence band and the conduction band. Metals, insulators and semiconductors In solids, the positions of the protons are all fixed, so all the conduction has to take place by movement of electrons. Obviously, an empty band conducts no electricity there are no mobile charge carriers available to do so. One surprising fact is that a full band does not conduct electricity either. Conduction involves moving electrons, and that in practice means changing the electron from one state to another. If all the states in a given band are full, then by Pauli exclusion no electron can change state if it is to stay within a band. Hence a full band conducts no electricity.
Metal Insulator, or semiconductor at low temperature

Conduction band Band gap, energy EG Valence band

Fig. 7.3. Illustration of a band structure for a metal, showing a full valence band and a partially full conduction band, and an insulator, or a semiconductor at low temperature, showing a full valence band and an empty conduction band. In pure semiconductors at low temperature and in most insulators (at least pure crystalline ones), the valence band is essentially full, and the conduction band is essentially empty. Hence in both the pure, low-temperature semiconductor and the insulator, there is very little conduction of electricity. A typical metal, by contrast, may have a conduction band in which a specific fraction of the states is full; for example, the conduction band might be half full, as shown in Fig. 7.3. Because there are many electrons in the band and also many empty states for electrons to move into, conduction can be very strong, and most metals are in fact very good conductors of electricity. The distinction between an insulator and a semiconductor is mostly that an insulator has a much larger band gap energy. As a result it is very unlikely that thermal energy will be sufficient to allow many of the electrons in the valence band to be thermally excited into the conduction band, at least at moderate temperature, and the insulator remains insulating as it is heated up. In semiconductors, the band gap energy is smaller, typically somewhere in the range of about 0.5 eV to 3 eV, and even at room temperature there are EE41 Physics of Electrical Engineering 5

Chapter 7 Semiconductor physics some electrons that can be excited across the band gap energy just because of the available thermal energy. As a result, at finite temperatures even pure semiconductors are partially conducting. They are much better conductors than insulators, but they are very poor conductors compared to metals, hence the name semiconductors. If it were only the case that semiconductors were just poor conductors of electricity, they would be of little use to us. The real usefulness of semiconductors comes from the fact that they can be doped, which leads to a broad set of engineering tools and resulting devices. Doping in semiconductors One of the key points about a semiconductor is that, though in its pure or so-called intrinsic state it is quite insulating, its electrical conductivity can be altered by the introduction of very small amounts of so-called doping materials. This is easiest to understand if we think of a Group IV semiconductor material, such as silicon. As we discussed above, the silicon atom has four electrons in its outer shell, and hence can make four bonds, each with one electron. In the pure material, the resulting fully bonded structure is a tetrahedral one, as shown above in Fig. 7.2. An atom of a Group V material, such as arsenic, has five electrons in its outer shell. If we substitute such a Group V atom for one of the silicon atoms in the lattice, we are left with an extra mobile electron. This electron usually appears in the conduction band, making the semiconductor now conducting. This addition of Group V atoms to silicon is called ndoping. Here, n stands for negative. The Group V atom is then known as a donor atom, because it donates an electron to the material. If, instead, we substitute an atom of a Group III material, such as boron, then we are missing an electron. In practice, this absence of an electron appears in the valence band. From an electrical point of view, adding an absence of an electron is in some ways like adding a positive charge to the mobile charges in the material, and so this type of doping is called p-doping. Here p stands for positive. The Group III atom is then known as an acceptor, because it has accepted an electron from the pool of electrons in the valence band of the material. Essentially, we can often view the crystal as still having each individual bond complete, but with a floating deficit of one electron in the valence band. Note, incidentally, that adding donor or acceptor atoms to the crystal still leaves the crystal electrically neutral overall. The extra electron from the donor is matched by an extra positive charge on the donor nucleus. The missing electron from the acceptor is matched by its nucleus having one less positive charge than the other atoms. We can see that this ability to dope the semiconductor allows us to control the conductivity of the material. Increasing n-doping gives us more electrons, and hence more conduction, in the conduction band. Increasing p-doping takes electrons out of the valence band. That leaves a band that is no longer completely full, and that fact allows the valence band to conduct electrical current. To understand that valence band current better, it is useful to introduce the idea of holes. Electrons in n-type semiconductors and holes in p-type semiconductors are illustrated in Fig. 7.4.

EE41 Physics of Electrical Engineering

Chapter 7 Semiconductor physics Holes It is inconvenient to keep referring to the absence of an electron, and instead we can talk about a positively charged hole. The hole then is simply the absence of an electron, in a somewhat similar way that a bubble in water is an absence of a small sphere of water. This hole description works much better than we might imagine in describing semiconductor behavior. It turns out we can consider the hole as behaving like a positively charged particle that moves in the direction we would expect a positively charged particle to move if we applied an electric field to it. The electric current in p-type semiconductors can then be considered as being from the movement of these positively charged holes in the valence band. We can think of holes, as suggested above, as being like bubbles in water. Drops of water in an empty bucket will fall to the bottom of the bucket; similarly, electrons in an empty conduction band will fall to the bottom of the band. Bubbles in a full bucket of water will rise to the top of the bucket; similarly, holes in a full valence band will rise to the top of the band. To view holes in the same way as we view electrons, we should stand on our heads and view the picture upside down. 7 If we also take a negative image of the picture at the same time, the p-type semiconductor now looks just like the n-type semiconductor looked before, with black dots in a light background. In fact, we could view the valence band as initially being empty of holes, and then the doping has added a few holes to it.

n-type semiconductor Conduction band

p-type semiconductor

Valence band

Fig. 7.4 Illustration of additional electrons in the conduction band on an n-type semiconductor, and absences of electrons (or, equivalently, presences of holes) in the valence band of a p-type semiconductor. In both the n-type and p-type cases, it is the added particles, electrons in the n-type and holes in the p-type, that allow the material to conduct well. Still, we have not yet made it clear what is so useful about semiconductors. One major class of devices is based on the special properties that results when we bring together n-

It might be easier just to turn the picture upside down!

EE41 Physics of Electrical Engineering

Chapter 7 Semiconductor physics type and p-type materials, giving diodes and also bipolar junction transistors. Another major class of devices, the metal-oxide-semiconductor field effect transistors used in the great majority of integrated circuits result because we are also able electrically to vary the carefully controlled amount of electrons or holes we can introduce in semiconductors. These various diode and transistor ideas are behind nearly all the electronic and optical semiconductor devices in use today, and we will return to discuss these later. Before looking at those devices, we need to understand how electrons and holes are distributed thermally in the bands. Thermal distributions of electrons and holes Thus far we have just considered that there can be greater than zero electrons in the conduction band or greater than zero holes in the valence band as a result of doping. To understand many effects that are important in devices, we have to know the energy distribution of the electrons or holes within the bands, and how temperature affects them. This matters for some aspects of electrical conduction, and also for understanding the spectra of light emitted by light emitting diodes, for example. It is particularly important that we understand these thermal distributions if we want to understand semiconductor junctions, such as the p-n junction diode, and if we want to understand the effects of applying voltages to semiconductors. Boltzmann factor When we heat things up, we are giving them energy. Thermal energy always has a certain randomness to it, but we can draw some conclusions about the probabilities of a given system, such as an atom or the collection of electrons in a solid, being in a specific energy state at a given temperature. One well-known relation for the relative probability P (E ) of a system being in a state of energy E is the so-called Boltzmann factor
E P (E ) exp kBT

(7.1)

Here T is the temperature (in degrees Kelvin), and kB 1.38 x 10-23 J/K is the Boltzmann constant. The Boltzmann constant as a number may not seem a particularly meaningful one it is just a very small number but a more meaningful way to understand the energy scale here is to note that, at room temperature, if we express all the energies in electron-Volts (by dividing each energy by the electronic charge, e 1.602 x 10-19)

kBT 25 meV at 20 C ~ 293 K

(7.2)

This Boltzmann factor is exactly correct for the specific situation of a particle that can be in one of several different states, such as an atom. The relative probability that we would find the atom in a specific state of energy E 2 as opposed to a specific state of energy E1 is
(E E1 ) P (E 2 ) exp (E2 / kBT ) = = exp 2 P (E1 ) exp (E1 / kBT ) kBT

(7.3)

EE41 Physics of Electrical Engineering

Chapter 7 Semiconductor physics The branch of physics that enables us to deduce this Boltzmann factor and other thermal distributions of particles is called statistical mechanics, though unfortunately we do not have space here to discuss the relevant derivations. Note that this probability distribution has several things we might expect given the notions of available thermal energy. (i) As we increase the energy E 2 , the probability that the particle is in that state progressively decreases we are less likely to find particles in states with higher energies. (ii) As we increase the temperature T , we are progressively more likely to find particles in higher energy states. Fermi-Dirac distribution In a semiconductor, the statistical problem we are interested in is usually somewhat different from the one we discussed above for the Boltzmann factor, though it is related. We want to know whether a specific electron state of some energy E is occupied or not it can have zero or one particles in it which is different from asking what is the relative occupation probability of two specific energy states of the entire system. The result of that statistical mechanics analysis for electrons (or any particle that obeys Pauli exclusion, i.e., any fermion) is the so-called Fermi-Dirac distribution

PFD (E ) =

1 (E E F ) 1 + exp kBT

(7.4)

1.5

Number of particles per state

Maxwell-Boltzmann

1
Fermi-Dirac

0.5

10

0 E EF k BT

10

Fig. 7.5 Graph of the Fermi-Dirac distribution, and of the Maxwell-Boltzmann distribution to which it asymptotes. This distribution is illustrated in Fig. 7.5. The parameter EF is known as the Fermi energy. The Fermi energy depends strongly on the amount of doping in the material, and, EE41 Physics of Electrical Engineering 9

Chapter 7 Semiconductor physics usually to a lesser extent, on the temperature. For the moment, we can regard it as a normalizing parameter. Obviously if we adjust it, then for every possible state of the system, we would be adjusting the probability of occupation of that state. Hence, if we knew that there were altogether, say, 1017 electrons per cubic centimeter in the system, there would be some specific value of EF we could choose that, when we added up all the occupation probabilities for all the electron states, we would get a total result equal to 1017 electrons in that cubic centimeter. One can regard it as the parameter that has to be set correctly to get the right total number of particles in the system. The Fermi energy has a more specific physical meaning also, and we will discuss that in more detail below. Note the following characteristics of the Fermi-Dirac distribution. (i) The occupation probability does fall off at higher energies as we would expect. (ii) The occupation probability does also increase at higher energies as we increase the temperature (at least if we presume the Fermi energy does not change much with temperature). (iii) At an energy E = EF , the occupation probability is . (iv) The occupation probability, which can also be regarded as the effective average number of electrons in the state of interest, never rises above 1, as required for particles obeying Pauli exclusion. Maxwell-Boltzmann distribution as an approximation to the Fermi-Dirac distribution In classical physics, we view each particle as being quite distinct, and we do not have a Pauli exclusion principle preventing us from putting two particles in the same state. In that classical case, the distribution that tells us the number of particles in a given state of energy E is the Maxwell-Boltzmann distribution
E N ( E ) = N o exp k BT

(7.5)

where No is a normalizing constant chosen also to give us the correct total number of particles in the system. 8 For energies E that are larger than EF by many kBT, the FermiDirac distribution becomes asymptotically equal to the Maxwell-Boltzmann distribution with
B

E N o = exp F k BT

(7.6)

The mathematics of working with a simple exponential are much easier in many cases than working with the full Fermi-Dirac distribution. This approximation turns out to be a good one for many situations, including most situations with silicon electronic devices. It is, however, necessary to use the full Fermi-Dirac distribution for handling many

Note that there is a distinction between the Boltzmann factor and the Maxwell-Boltzmann distribution. The Boltzmann factor tells us the relative probability that a given entire system could be in one of two different states separated by an energy E. It is actually always correct for a system in thermal equilibrium, including systems of identical particles, including ones obeying the Pauli exclusion principle. The Maxwell-Boltzmann distribution tells us, in a system of many (non-identical) particles, with no Pauli exclusion, what is the average number of particles that are in a state of a given energy. It does not refer to the energy of the entire system. Other particles can be in states of other energies.

EE41 Physics of Electrical Engineering

10

Chapter 7 Semiconductor physics situations involving III-V semiconductors, and it is essential for analysis of most lightemitting devices, such as semiconductor light-emitting diodes and lasers. Meaning of the Fermi energy There is another important aspect about the Fermi energy, which is that it is the same everywhere in a system in equilibrium. To understand why that might be, we need to take a step back to a broader concept. We are already used to the notion that for two bodies in contact, when they are at thermal equilibrium, they have the same temperature. They achieve the same temperature by exchanging energy with one another. Another way in which systems can be in equilibrium with one another is with respect to exchange of particles. Imagine, for example, that we had a box divided in two by some barrier, and that we had two different gasses, one in the left hand side of the box and one in the right hand side of the box. We can even presume if we wish that the temperature and pressure are equal on the two sides of the barrier, so these two sides of the box are at least in thermal equilibrium, and that there is no force on the average pushing on the barrier in one direction or another. But we know if we remove the barrier, the two gasses will mix, and they will eventually settle down, at least in this simple example, to equal fractions of each gas in each side of the box. This situation is illustrated in Fig. 7.6.

barrier

(a) with barrier

(b) no barrier

Fig. 7.6 Illustration of a box, initially, (a), with a barrier between the two halves, with different gasses in the two halves, and (b), after the barrier has been removed and the gasses have mixed. We can see that the situation in Fig. 7.6 (b) is an equilibrium situation. Aside from moderate statistical fluctuations in the exact numbers of atoms of each kind of gas on each side, these gasses are not going to unmix. But what quantity is it that is being equalized between the two sides of the box in such a particle mixing experiment? The answer is something called the chemical potential. There are, in fact, separate chemical potentials for each type of gas or species of particle, and when we allow all the particles to move across the barrier, each of these chemical potentials is separately equalized on EE41 Physics of Electrical Engineering 11

Chapter 7 Semiconductor physics both sides of the box. We could, for example, additionally have had equal fractions of a third gas in both sides of the box to start with, in which case the chemical potentials of this third gas on both sides might already have been equal. The Fermi energy in a material is, in fact, simply the chemical potential for electrons. We can see, from our definition of the Fermi-Dirac distribution, Eq. (7.4), if the Fermi level is equal on both sides of some junction of two pieces of semiconductor, for example, the occupation probability of electron states of the same energy on each side of the junction will be equal, so it is reasonable to presume then that there should be no net flow of electrons from one side of the junction to the other once we have reached this equilibrium with equal Fermi energies (electron chemical potentials) on each side of the junction. The position of the Fermi level is very important in setting the electrical and optical properties of material. Now we will look at where the Fermi energy is in the various materials of interest to us. Fermi levels in different materials Metals The Fermi level in metals is, rather obviously, at the top of the pool of carriers in the conduction band. States above this level are mostly empty; states below it are mostly full.
n-type semiconductor conduction band Fermi energy, EF EF EF valence band p-type semiconductor undoped semiconductor, or insulator

Fig. 7.7 Position of the Fermi level in doped and undoped semiconductors, and in insulators. Semiconductors The approximate positions of Fermi levels in doped and undoped semiconductors are shown in Fig. 7.7. Pure semiconductors and intrinsic carrier concentration In undoped, pure semiconductors, the Fermi level is somewhere approximately in the middle of the band gap. Because semiconductors have relatively smaller bandgaps, e.g., 0.1 to 3 eV, however, the occupation probabilities for electrons in the conduction band and holes in the valence band are not quite negligible. In pure semiconductor materials, there is therefore a so-called intrinsic carrier concentration. This concentration is the number of electrons per unit volume in the conduction band in this undoped material, and EE41 Physics of Electrical Engineering 12

Chapter 7 Semiconductor physics it is necessarily equal to the number of holes per unit volume in the valence band since there are no added dopants in this ideal case. At 20 C, in silicon, with a bandgap energy of ~ 1.1 eV, the intrinsic carrier concentration is ~ 6.6 x 109 cm-3, and for GaAs, with its larger bandgap energy of ~ 1.4 eV, it is much smaller, at about 1.1 x 106 cm-3. These intrinsic concentrations are also strongly dependent on temperature. Because of this intrinsic carrier concentration, pure semiconductor materials, though not highly conducting, are also not good insulators, hence the name semi conductors. It is also common to refer to such undoped semiconductors as intrinsic semiconductors, since they have their intrinsic properties rather than those caused by added dopants. Doped semiconductors In doped semiconductors, for many purposes we can simply assume that the Fermi energy is at the edge of the appropriate band, i.e., for n-doped semiconductors, the Fermi energy is at the bottom edge of the conduction band, and for p-doped semiconductors, the Fermi level is at the top edge of the valence band. This is not exactly correct, because the precise position of the Fermi level does vary with dopant concentration, but this variation is of the scale of 10s of meV as we change the doping concentrations over a large range. Such Fermi energy variations are usually small compared to the bandgap energy. For many situations with silicon, the Fermi energy is just below the conduction band in ntype materials, and just above the valence band in p-type materials. For III-V materials, the Fermi energy is often slightly above the bottom of the conduction band in n-type materials. These statements about the position of the Fermi level in doped semiconductors only apply in the bulk of the semiconductor. In a small region (usually ~ 0.1 to 1 micron) near the junction of semiconductors of different dopings (as in the p-n junction diode we will discuss in the next section), the Fermi level goes through a transition from one band-edge to the other, though, in a diode without applied bias, it is still constant throughout the structure. We will look at such junctions in the next section. Fermi level and Fermi-Dirac distribution for holes We have discussed so far the Fermi level for electrons. What about holes? This turns out to be quite simple. If the probability of occupation of a state by an electron is some number PFD, then the probability that there is no electron in that state, or equivalently the probability that there is a hole in that state, is PFdhole = 1 - PFD, i.e.,

PFDhole (E ) = 1

1 (E EF ) 1 + exp kBT (E E F ) exp kBT = (E E F ) 1 + exp kBT

(E E F ) 1 1 + exp kBT = (E EF ) 1 + exp kBT

EE41 Physics of Electrical Engineering

13

Chapter 7 Semiconductor physics i.e.,

PFDhole (E ) =

1 (E E F ) 1 + exp kBT

(7.7)

In other words, again for holes we simply have to stand on our heads and look at the energy upside down. The holes have a Fermi-Dirac distribution also, but with the sense of the energy reversed. So, as we move below the Fermi energy, the probability of finding a hole in a state decreases, or, as we move above the Fermi energy, the probability of finding a hole in a state increases. In a complete equilibrium situation, as we are discussing here at the moment, the hole Fermi energy is exactly the same as the electron Fermi energy. Holes and electrons are simply two different ways of looking at the same thermal distribution. Insulators In insulators, the Fermi level is usually somewhere in the middle of the bandgap. Given that insulator bandgaps are large, e.g., 4 eV, the resulting thermal occupation probability for any given state in the conduction band is very low. A separation of 2 eV from the middle of the insulator bandgap to the conduction or valence band edge corresponds to ~ 80 x kBT at room temperature, so the occupation probability of a given electron state in the conduction band is ~ exp (80) 1.8 x 10-35, which is a very small number. The probability that there is an electron missing from any specific state in the valence band is similarly very low. Hence insulators insulate! It should appear now that there is little qualitative difference between insulators and undoped semiconductors other than that insulators have larger band gap energies. One additional point about insulators, which does make them qualitatively different from semiconductors in practice, is that it is very difficult or impossible to dope insulators to make them conducting. Hence in practice, insulators do not need to be very pure to be insulating. By contrast, a broad range of impurity materials can make semiconductors conducting, even at low densities of the impurities. Dopant levels in semiconductors and insulators Why is it so difficult to dope insulators? The reasons for this tend to apply to all materials with wide band gaps, and these reasons come from some of the physics of dopants. In fact, when we add dopant atoms, they do not necessarily give up their additional electron or additional hole to the conduction or valence band. For the dopants we normally use in semiconductors, a small amount of energy is required to get them to give up their additional particle to the relevant band, an energy called a dopant ionization energy. We can think of the dopants as adding some additional energy levels in the semiconductors, energy levels that are just below the conduction band for n-type dopants and just above the valence band for p-type dopants. The electrons or holes in these levels are easily ionized by the available thermal energy to put the corresponding electron or hole in the conduction or valence band. For a complex set of reasons, dopants added to wide band gap materials often have large dopant ionization energies, too large for the thermal energy available at normal temperatures to enable the necessary ionization. In insulators added materials tend to give rise to additional isolated energy levels in the band-gap region, well away from the EE41 Physics of Electrical Engineering 14

Chapter 7 Semiconductor physics conduction or valence bands. Because these levels are so far from the conduction or valence bands, the thermal energy available is not enough to take the electrons or holes out of these states and put them in the conduction or valence bands where they can give conduction. Conduction in doped semiconductors From the discussion above, we understand how we get electrons and holes into semiconductors by doping, and we expect that, as a result, they will conduct electricity and will in general be able to move about. Now we will try to understand just how they do move about in semiconductors. There are two mechanisms for their movement that are particularly important in electronic devices. One is called drift, and the other is called diffusion. Drift We might imagine that the movement of an electron in a semiconductor is just a simple matter of applying a force to such a charged particle and accelerating it. Because we have accelerated it, it is now moving with some velocity, and hence there is a current. That process is part of what is going on, but it is not the whole story, and, indeed, could not explain the kind of behavior associated with the idea of resistance (Ohms law). For example, suppose we had an electron at one end of a piece of conducting material, and we applied a voltage V to the material, pulling the electron to the other end. Assuming we simply accelerated the electron, at the other end the electron would have a kinetic energy

K .E . =

1 2 mv = e (charge) V (voltage) 2

(7.8)

where m is the electron mass and v is its velocity. Hence the velocity at the other end would be v = 2eV / m (7.9)

Since the current from this electron is simply the charge times the velocity, we find that, at the other end of the material, we would have a current that is proportional to V , whereas Ohms law must correspond to a current proportional to the voltage V. Since we know that materials such as metals and semiconductors do behave like a resistor for a broad range of applied voltages or electric fields, this simple accelerating electron model cannot explain the movement of electrons inside a semiconductor or metal. 9 A model that does seem to correspond well to the movement or transport of electrons in many conducting materials is the drift transport model. In this model, we presume that the electron is continually bumping into things inside the material. So, it is accelerated for a time, and then hits something. We presume in this model that, when it hits something, that collision is so strong that, effectively, the electrons velocity after the collision has been completely randomized in direction. Hence, on the average, the electrons velocity after such a collision is zero. We also presume in this model that there is some characteristic or average time, , between collisions, a time that does not depend on how
9

This kind of simple accelerating electron model does, however, sometimes work over very short distances and times, in which case it is called ballistic transport.

EE41 Physics of Electrical Engineering

15

Chapter 7 Semiconductor physics

fast the electron is going. Hence the movement of the electrons behaves on the average as if they are each accelerated by the applied electric field for some time , then their velocity is reset to zero, then they are accelerated again for some time , then their velocity is again set to zero, and so on. The something that the electron hits could be, for example, another electron or hole, or an impurity atom. It could also be that it encounters a temporary distortion of the lattice; if an atom is slightly out of place in the lattice, that would seem like an imperfection to the electron, and the electron could scatter off of it. Actually, having just a single atom out of place would be relatively unusual. What is more likely is that the whole crystal is vibrating slightly, just because of the thermal energy, and those vibrations will cause atoms to be temporarily very slightly out of their strict crystalline positions. The electron can scatter off of those vibrations. Because such vibrations are quantum mechanically described in terms of both waves and the quantum mechanical particles, phonons, associated with those waves, this kind of scattering is known as phonon scattering. Given that the electrons are scattering off of something, with a characteristic time between such scattering events, we can now attempt to calculate the current in such a situation. To calculate that current, we need to calculate the average or drift velocity, vd, of the electron. If we accelerate an electron with an electric field of magnitude E, then the force is of magnitude eE. From elementary classical mechanics (force = mass times acceleration), the acceleration is of magnitude a= eE m (7.10)

where m is the electron mass. The peak velocity of the electron, just before its average collision, is a. The average velocity is half of this, because the velocity is linearly increasing through this time , and so we have, for the magnitude of our average drift velocity

vd =

e E 2m
area A

(7.11)

Fig. 7.8 Bar of semiconductor with electrons moving under drift transport. Now we see that we have a model in which the average electron velocity is proportional to the applied electric field, which is what we require for behavior consistent with Ohms
EE41 Physics of Electrical Engineering 16

Chapter 7 Semiconductor physics

law. In practice with a drift model, it is not at all easy to measure the scattering time in any direct way, and we simply bundle the various terms in Eq. (7.11) to rewrite it as vd = eE where the quantity e is called the electron mobility, and (7.12)

e =

e 2m

(7.13)

To calculate the current from such a drift model, imagine that we have a bar of semiconductor, of length L and cross-sectional area A, with a density of N mobile electrons per unit volume, as shown in Fig. 7.8. We are applying a voltage V from one end to the other, resulting in a field E = V/L. In this model, we presume that the electrons are traveling at the corresponding drift velocity vd. Imagine we have a surface mathematically cutting through the bar perpendicular to the direction of the electric field; this surface also will have an area A. The number of electrons crossing unit area per second is simply the number of mobile electrons per unit volume, N, times the velocity with which they are moving, vd. The total number of electrons crossing that area A per unit time is therefore NvdA, so the magnitude of the electrical current is I e = Nvd Ae or, using the field E = V/L and the relation (7.12) for the drift velocity I e = Nee Since Ohms law is
V = IR

(7.14)

A V L

(7.15)

(7.16)

where R is the resistance, we have a resistance for this bar of semiconductor, assuming only electron drift current, of
R= L 1 A Nee (7.17)

To describe the resistive properties of the material itself, independent of the particular size of bar, we can define a resistivity e

e =

1 Nee

(7.18)

which would be the resistance of a bar of unit length and unit cross-sectional area. It is actually more common to use the conductivity

= Nee

(7.19)

EE41 Physics of Electrical Engineering

17

Chapter 7 Semiconductor physics

Drift transport works equally well for holes. The electrical current will flow in the same direction as for the electron case, though the holes are flowing in the opposite direction. We obtain an electrical current from holes of I h = Peh A V L (7.20)

where P is the number of holes per unit volume, and h is the hole mobility. The total current will be the sum of the electron and hole drift currents
A I drift = ( Nee + Peh ) V L A = ( e + h ) V L

(7.21)

where h is the hole conductivity

= Peh

(7.22)

and we have also defined a resistivity h that would correspond to hole current only. The fact that we can just add the hole and electron conductivities to get the total conductivity is one reason why we more often work with conductivities than resistivities when looking at such transport. The fact that the electrons and holes keep bumping into things in the crystal and, on the average, losing all their kinetic energy in the process, of course dumps all that energy into the crystal, giving the energy dissipation that we associate with ordinary resistors.
Diffusion

The other major mechanism for carrier (i.e., electron or hole) movement in semiconductors is diffusion. Diffusion is driven not by electric field or voltage, but by differences in concentrations of particles. Diffusion is the process behind the simple mixing of particles, as in the gas box example above in Fig. 7.6. Diffusion will result if particles move in random directions, changing that direction randomly in time, a process sometimes known as a random walk. If we have a high concentration of particles near one position, then, over time, such a random walk process will cause the particles progressively to spread out, for example. In general in a diffusion process, particles on the average move from regions of high concentration to regions of low concentration. This movement results not from any specific force in a given direction, but just from the sum of all of the random motions. For example, consider again a box, as in Fig. 7.9, with two regions separated initially by a barrier. We are in thermal equilibrium between the two halves they are at the same temperature so the particles in each half each have the same average thermal energy available to allow them to move randomly. In the left half there is a concentration N1 of a given kind of particle and in the right half there is a concentration N2 of the same kind of particle. (We can arrange so that there is no pressure driving the particles from one side to the other, for example by adding another species to balance the overall pressure as in
EE41 Physics of Electrical Engineering 18

Chapter 7 Semiconductor physics

Fig. 7.6.) Now, as before, we remove the barrier. Because of the particles random walk motion, we expect that, after we remove the barrier, in some particular short time, some small fraction B of the particles on the left will cross the barrier into the right half, and similarly the same fraction of the particles from the right half will cross into the left half. The net number of particles that move from left to right is therefore BN1-BN2. Note that the net number of particles that move in a given time is proportional to the difference in concentrations on the two sides, not the actual concentrations.
N1 N2

BN1 BN2

Fig. 7.9 Box just after a barrier has been removed (in the position of the dashed line). Initially, there are N1 particles per unit volume in the left half of the box, and N2 of the same type of particles per unit volume in the right half of the box. In some short time after the barrier is removed, some fraction BN1 crosses from the left half into the right half by a diffusion or random walk process, and similarly a fraction BN2 crosses from the right half to the left half. In actual diffusion, we do not usually have abrupt changes in concentration, but rather we have smooth variation of concentration. It is still the case that the net flow is given by the change in concentration from one side to another. The net flow in the x-direction then tends to be proportional to the derivative of the concentration in the x-direction, dN/dx. Larger values of this derivative correspond to more abrupt changes in concentration, and less balanced flow from one side to another. Note that the net flow is from high to low concentration, so the flow is in the opposite direction to the direction of increase of concentration. Hence the net particle flow (per unit area per unit time) across some surface perpendicular to the x-direction can be written as Net particles per second per unit area = D
dN dx

(7.23)

for some (positive) proportionality constant D. The minus sign here is because the particles on the average flow from higher concentration to lower concentration. This constant, D, is called the diffusion coefficient, and is in general different for different materials, temperatures, and particles. For the case of electrons, which have charge e, we would also see a net electrical current from such a diffusion. I.e., for a cross-sectional area A, we would see a net electron electrical current
I ediff = eDe A dN dx

(7.24)

EE41 Physics of Electrical Engineering

19

Chapter 7 Semiconductor physics

where De is the electron diffusion coefficient. Similarly, for holes with concentration P, we would see a net hole electrical current
I hdiff = eDh A dP dx

(7.25)

where Dh is the hole diffusion coefficient. Of course, if we have both electrons and holes present, we would get the sum of these two currents. Now that we have understood the basics of semiconductor bands, thermal distributions, and drift and diffusive transport, we have the basic tools to understand how semiconductor devices work, and that will be the subject of the remaining sections.
Recombination and diffusion length

Two more basic concepts are needed to discuss devices such as diodes, namely recombination and diffusion length.
Recombination

In general, if we have electrons in the conduction band, and holes in the valence band, there will always be some probability that the electron could somehow fall into the valence band and fill in the hole. Such a process is called recombination. In discussing populations of electrons and holes in semiconductors, we have so far considered essentially only thermal equilibrium situations, and have presumed the populations to be steady. In fact, the processes of recombination are going on all the time, even in thermal equilibrium, and then they are balanced by continual processes whereby an electron in the valence band somehow acquires enough energy from the surrounding environment to be excited into the conduction band. 10 Recombination becomes particularly important to understand once, for some reason or another, we have excess numbers of electrons and/or holes in a piece of semiconductor, i.e., the numbers differ from their thermal equilibrium values. Recombination then acts to try to restore the thermal equilibrium balance of carriers. This process turns out to be very important in many electronic and optoelectronic devices. In a piece of doped semiconductor, to a first approximation there is essentially only one kind of carrier the "majority" carrier (holes in p-type material, electrons in n-type material), the density of which determines the conductivity; with essentially only one kind of carrier present, there is little recombination going on anyway. Recombination processes are, however, very important in understanding the operation of diodes because "minority" carriers (i.e., electrons in p-type material, holes in n-type material) are injected from one side of the diode into the other. Majority-carrier electrons in the n-material are injected into the p-material, where these electrons are minority carriers. These minority-carrier electrons then recombine with the majority-carrier holes in the p-material. A similar process occurs in the opposite sense for holes from the p-material. This recombination of minority carriers with the majority carriers within the p

In fact, for every recombination process there is an exact inverse physical generation process, and in thermal equilibrium a process and its inverse are always running at equal rates so that the equilibrium is maintained, an equality known as detailed balance.

10

EE41 Physics of Electrical Engineering

20

Chapter 7 Semiconductor physics

and n regions is an important part of the normal functioning of the diode in fact, without it, there would be little current in the diode, as we will see below. For light emitters, recombination is a crucial process, since this is typically how photons are generated. Recombination can also be an important but undesirable parasitic process. It can lead to "leakage" currents in diodes, and non-radiative recombination reduces the efficiency of light-emitting diodes and lasers. There are two main broad classes of processes by which electrons in the conduction band may recombine with holes in the valence band: radiative recombination, and nonradiative recombination.
Radiative recombination

The radiative recombination processes are usually the spontaneous and stimulated emission processes discussed before. They are intrinsic processes in the semiconductor, being present even in a perfect material with no defects of any kind. We can calculate radiative recombination rates from first principles if we wish, given particular carrier densities in the material, though we will not do that here. Typical radiative lifetimes (i.e., average times before an electron and hole would recombine, emitting a photon) can vary from ~ 100 ps to microseconds. They are strong in the class of materials known as direct gap semiconductors, which includes essentially all of the III-V semiconductors used for light emitting diodes and laser diodes. They are weak in the class of materials called indirect gap semiconductors, which includes silicon and germanium 11 ; this unfortunately means that it is difficult to make light emitters out of these materials. There is no single answer to the value of the radiative lifetime in a given material, because it depends on the number of electrons and holes present, and, in the case of stimulated emission, also on the number of photons present.
Non-radiative recombination

One of the most common mechanisms of non-radiative recombination is recombination through traps. In imperfect semiconductors, there can be some states, arising from impurities or crystal defects, at energies deep inside the bandgap. These states can lead to recombination of electrons and holes by non-radiative processes. One mechanism is the so-called Shockley-Read-Hall mechanism, in which these states may first capture, for example, an electron, and then subsequently capture a hole that neutralizes the electron to complete the recombination process (or capture a hole first, then an electron). With large electron or hole densities, it is possible to saturate the traps, but otherwise this recombination through traps often behaves like a simple constant lifetime for the carriers (i.e., a recombination rate simply proportional to the number of electrons or the number of holes as appropriate), i.e., for electrons
rtrap = n

trape

(7.26)

The problem in indirect semiconductors is that the electrons and holes tend to have quite substantially different momenta. A photon has relatively small momentum, and momentum cannot be conserved if all we do in recombining the electron and hole in the indirect semiconductor is emit a photon. We therefore have to have something like a phonon, which can have quite large momentum, also participate in the process, but that makes the whole process much less likely.

11

EE41 Physics of Electrical Engineering

21

Chapter 7 Semiconductor physics

where n is the excess density of electrons in the material relative to the thermal equilibrium density, and trape is the associated electron lifetime; similarly for an excess density, p of holes,
rtrap = p

traph

(7.27)

with an associated hole lifetime of traph. Because recombination through traps is generally not an intrinsic effect, and because it is caused by imperfections that are often not well understood, recombination through traps is generally not calculated from first principles, and the resulting lifetime can vary widely over ranges as wide as ~ 1 ps (in materials with intentionally large numbers of defects) to microseconds or possibly longer, depending on the imperfections in the material. One important kind of recombination through traps, which is almost always an undesirable parasitic effect, is so-called surface recombination, in which the traps are at surfaces or interfaces. Sometimes these traps are empty bonds at the crystal surface, or come from impurities at the surface. The surfaces can be the external surface of the semiconductor, or can be interfaces between two materials, such as between a semiconductor and an oxide, or a semiconductor and another semiconductor. Some such interfaces have very few so-called "surface states" or "interface states, with a classic (and very important) example being the interface between silicon and silicon oxide as used in integrated circuit transistors. The existence of such an interface with very few surface states or recombination centers is one of the main reasons that silicon is used for integrated circuits, allowing very large numbers of stable and reliable transistors to be made. Interfaces between III-V semiconductors and other materials are typically not as good from the point of view of surface recombination, and it can be difficult to find ways of treating the surface to get rid of surface states permanently. Interfaces between different (lattice-matched) III-V semiconductors, such as the GaAs-AlGaAs interface, generally have relatively little surface recombination however, which is very fortunate for many optoelectronic devices.
Diffusion length

One important situation in semiconductor devices is where there is some excess carrier density (e.g., electrons), say at one end of a piece of semiconductor material. Holding such a density above the equilibrium value for that material leads to a current, which is in fact the main forward current in ordinary diodes, as we will see in the next section. We are interested here in the steady state, where the net number of electrons entering a volume is balanced by the number of electrons recombining in the volume. We will presume the recombination is characterized by a simple lifetime, e. We will presume that there is negligibly small electric field present for simplicity, which is anyway a reasonable assumption if we are talking about a doped semiconductor that is therefore quite conducting. With no field, there is no drift transport of electrons, and the only transport is then from diffusion. Consider a thin slice of material, thickness x. We know from Eq. (7.23) that the net number coming in to this slice from the left is De (d n( x) / dx) per unit area. (Here we

EE41 Physics of Electrical Engineering

22

Chapter 7 Semiconductor physics

are using n , the difference compared to the thermal equilibrium concentration, rather than N, the total concentration of electrons, but that choice makes no difference to the derivative.) Similarly, the total number leaving the slice to the right is De (d n( x + x) / dx) per unit area. Hence
Net number of carriers entering per unit volume = d n ( x ) d n ( x + x ) 1 +D D dx dx x d 2 n as x 0 dx 2

(7.28)

(Note that the volume of this slice of unit area is simply x, and we have divided by this volume to get a result per unit volume in Eq. (7.28) above.) In the steady state, this number must equal the number recombining per unit volume, i.e.,
d 2 n n De = dx 2 e

(7.29)

The solution to this problem is


n ( x ) = n ( 0 ) exp ( x / Le )

(7.30)

where n(0) is the excess electron density at the end where the electrons are being injected, and the parameter
Le = De e

(7.31)

is referred to as the electron diffusion length. We can obviously define a similar quantity
Lh = Dh h

(7.32)

as the hole diffusion length, with h as a corresponding hole lifetime. The diffusion length is the characteristic length a particle will diffuse, in the absence of an electric field, before recombining. Hence we find that the excess electron population will decay progressively into the ptype material (and similarly for excess holes in an n-type material). That decay means we have a population gradient, and there is diffusion current as a result. That current, as we will see below, gives rise to the forward current in a diode.
Summary Crystal

A crystal is regular structure formed by repeating the same unit (the unit cell) in such a way as to fill all space.

EE41 Physics of Electrical Engineering

23

Chapter 7 Semiconductor physics III-V compounds

III-V compounds are made from elements from Group III of the periodic table (e.g., gallium, indium, aluminum) and Group V (e.g., arsenic, phosphorus, antimony, nitrogen).
Diamond and Zinc-Blende lattices

The diamond lattice is formed of two interlocking face-centered cubic lattices. On the zinc-blende lattice, one of the face-centered cubic lattices has one kind of atom (e.g., Ga) and the other has anther kind of atom (e.g., As). Silicon and germanium have the diamond lattice. Most III-V compounds in practical use have the zinc-blende lattice.
Ternary and quaternary alloys

A ternary alloy as three components, e.g., InGaAs, and a quaternary has four, e.g., InGaAsP.
Epitaxial growth

Growth in which the new material takes on and extends the crystalline form of the substrate on which it is grown.
Bands in solid materials

When atoms are joined in crystalline materials, the individual energy levels of the atoms form into bands of energy levels in the material.
Conduction and valence bands

The conduction band is the lowest energy empty or partially filled band. The valence band is the highest energy filled or partially filled band.
Band gap

The energy separation between the edges of the conduction and valence bands.
Metals, insulators and semiconductors

Metals typically have a partially filled band. Insulators and semiconductors have both a full valence band and an empty conduction band (at least in the pure material at low temperatures). Semiconductors can be made to conduct by adding dopants. Insulators and pure semiconductors do not conduct electricity well (note that a full band does not conduct electricity).
Holes

A hole is an absence of an electron, and is a convenient way to describe conduction of charge in bands that are nearly full of electrons. It can be thought of behaving as a positively charged particle that can be accelerated by electric fields in the way a positively charged particle would be.
Donor and acceptor dopants

A donor atom (n-dopant) can add one extra electron to the conduction band. An acceptor atom (p-dopant) can add one extra hole to the valence band.

EE41 Physics of Electrical Engineering

24

Chapter 7 Semiconductor physics Boltzmann factor

The relative probability P (E ) of a system being in a state of energy E is the Boltzmann factor
E P (E ) exp k T B

(7.33)

where

kBT 25 meV at 20 C ~ 293 K


is Boltzmanns constant.
Fermi-Dirac distribution and Fermi level

(7.34)

The Fermi-Dirac distribution


PFD (E ) =

1 (E E F ) 1 + exp kBT

(7.35)

gives the probability that a give state of energy E is occupied by a fermion (usually an electron).
EF is the Fermi energy. For a specific total number of fermions present, EF will take on a specific value, i.e., it is a parameter that can be adjusted to get the correct total number of particles in the system. It is also formally the chemical potential for the distribution of fermions.

In metals, the Fermi level is somewhere in the middle of a band. In insulators and in pure semiconductors, it is somewhere in the middle of a band gap between the valence and conduction bands. In n-doped semiconductors, it is near the conduction band edge. In pdoped semiconductors, it is near to the valence band edge.
Chemical potential

Just as temperature is the quantity that comes to equilibrium if we allow two systems to exchange energy, chemical potential is the quantity that comes to equilibrium if we allow two systems to exchange particles. There is one chemical potential for each species of particle.
Maxwell-Boltzmann distribution

The Maxwell-Boltzmann distribution


E N ( E ) = N o exp k BT

(7.36)

gives the number of particles in a state of energy E, if those particles are classical, i.e., are distinguishable and have no limits on the number of particles in the state. No is the parameter that can be adjusted to get the correct total number of particles in the system.

EE41 Physics of Electrical Engineering

25

Chapter 7 Semiconductor physics

The Maxwell-Boltzmann distribution is also an approximation to the Fermi-Dirac distribution for energies well above the Fermi energy, i.e., where the probability of occupation of any state is much less than 1.
Drift conduction

Drift conduction corresponds to electrons (or holes) moving at an average velocity proportional to the applied electric field, in which the electrons (or holes) are continually being accelerated and having collisions. On the average, the collisions, occurring on the average every seconds, reset the velocity to zero. The drift velocity for an electric field E is of the form
vd = e E 2m

(7.37)

leading to the definition of a mobility, e, such that


vd = eE

(7.38)

and similarly for a hole mobility h. For an electron density N per unit volume, this leads to a conductivity, e, e.g., for electrons, of

= Nee

(7.39)

where e is the resistivity, with similar quantities for holes.


Diffusion

Diffusion is the process by which particles move on the average from regions of high concentration to regions of low concentration, based solely on a random walk process. In such a process, the net particle flow (per unit area per unit time) across some surface perpendicular to the x-direction can be written as Net particles per second per unit area = D
dN dx

(7.40)

where D is the diffusion coefficient and N is the concentration of particles per unit volume. For electrons, such a process corresponds to a current
I ediff = eDe A
Recombination

dN dx

(7.41)

Recombination is the process whereby excess electrons in the conduction band recombine with excess holes in the valence band. One recombination will remove one electron from the conduction band and one hole from the valence band. One typical recombination process is recombination through traps, which often can be characterized by a life time, e.g., trape for electrons, leading to a recombination rate of
EE41 Physics of Electrical Engineering 26

Chapter 7 Semiconductor physics


rtrap = n

trape

(7.42)

for an excess number n of electrons. Another important recombination process is radiative recombination, in which an electron and hole recombine to emit a photon.
Diffusion length

When excess particles are being injected into a semiconductor from one end, and the particles are recombining with a characteristic lifetime, e.g., for electrons, of e, the density of excess particles falls off as
n ( x ) = n ( 0 ) exp ( x / Le )

(7.43)

where the diffusion length is


Le = De e

(7.44)

EE41 Physics of Electrical Engineering

27

Chapter 8 Semiconductor diodes Chapter 8 Semiconductor diodes Diodes are used extensively in electronics for rectifying (turning a.c. into d.c.), and perform various other functions in electronic circuits (e.g., level shifting, by effectively subtracting an approximately constant voltage (the diode forward voltage) from a signal). The vast majority of semiconductor optoelectronic devices are also based on diodes. This includes most photodetectors, light-emitters (both light-emitting diodes and lasers), and modulators. For optoelectronic devices in particular, we need to understand diodes at a level deeper than might be required for simple electronic applications. It is also the case that, once we have understood such diodes and the various physical phenomena that go on within them, we are close to understanding the operation and design of many of the various practical semiconductor diode optoelectronic devices. In optoelectronics, diodes are used primarily for three purposes. (i) To collect generated photocurrent in photodetectors The diode structure gives a good way to collect electrons and holes generated by optical absorption and turn the result of the absorption into an electrical current (photocurrent). A reverse-biased photodiode, for example, has relatively little current in the absence of light, but very efficiently generates the photocurrent from the absorbed photons, giving a detector with low background noise. Photodiodes can also be used to generate electrical power from light, as in the solar cell. (ii) To apply strong electric fields High fields can be generated, without substantial conduction current, by reverse biasing the diode. Such fields are useful in optical modulators (so-called FranzKeldysh and quantum well modulators) and so-called avalanche photodetectors, in which one absorbed photon can give rise to many electrons of current, by avalanche multiplication. (iii) to inject large densities of electrons and holes into the device In light emitters, the diode provides a convenient structure for controllably generating electrons and holes in a particular volume so that they can recombine to emit light. Under forward bias, holes are injected from the p-region and electrons from the n-region into the volume between or near to the edges of the p and n regions. p-n junction The semiconductor pn junction diode in its simplest form is made (conceptually if not actually) by joining two semiconducting materials, one p-doped, with its chemical potential (Fermi energy) near or in the valence band, and the other n-doped, with its chemical potential near or in the conduction band. To understand what happens when we do this, we start out by analyzing the case of thermal equilibrium. Equality of chemical potentials in thermal and diffusive equilibrium From basic thermodynamics, the chemical potentials must be equal throughout a system if it is in thermal and diffusive equilibrium, because otherwise particles would move so as to equalize the chemical potentials. Equality of chemical potentials ensures that the occupation probability of states of the same energy in different parts of the system is the EE41 Physics of Electrical Engineering 1

Chapter 8 Semiconductor diodes same, so there is no net transition rate between states of the same energy. In common semiconductor terminology, we say the Fermi levels must be equal (we continue to use the convention that the terms chemical potential and Fermi level are exactly equivalent in a semiconductor). This equality in equilibrium must hold even if the materials are different, and no matter how many materials of different kinds we join together. Hence it holds for any kind of structure including heterostructures (structures made from two or more different materials), no matter, for example, what the bandgap energies or dopants are. Well away from the junction between the two materials, we expect that the materials each should look just like they did before we joined them. Hence we can draw the valence and conduction bands far away from the junction, and we can draw the chemical potential throughout the whole structure, though we do not yet know what happens to any other properties in the region near the junction (Fig. 8.1)
conduction band

chemical potential valence band

? conduction band

? valence band p n

Fig. 8.1. n and p materials far on either side of the junction, and chemical potential (Fermi energy) throughout the whole structure, at thermal equilibrium. Formation of the depletion region When we initially join the materials, because of the imbalance of the number of electrons between the n and p materials, electrons tend to move (for example, by diffusion) from the n material into the p material, leaving behind a net positive charge in the n material (in the form of fixed ionized n-dopant atoms). In the p material, the electrons will tend to recombine with holes, tending to reduce the number of holes near the edge of the p material and creating a net negative charge just inside the p material (in the form of fixed ionized p-dopant atoms). Equivalently, we could say holes tend to move from the p material, where they are plentiful, into the n material, where they are scarce, leaving behind a net negative charge in the p material (ionized p-dopant atoms). By recombining with electrons just inside the n material, the holes also create a net positive charge just inside the n material. This transfer of particles is illustrated in Fig. 8.2. Because we have neither added nor removed charge overall from the material, the net positive charge built up in the n material is exactly balanced by the net negative charge built up in the p material. We therefore form a dipole in the region near the junction EE41 Physics of Electrical Engineering 2

Chapter 8 Semiconductor diodes between the p and n materials, which leads to an electric field in this region. The appearance of this electric field is, from a mechanistic point of view, the process that eventually stops the continued transfer of electrons and holes between the materials. The field is of such a sense that it tends to push electrons back into the n material and holes back into the p material. To put it another way, the presence of the field increases the energy required to put an electron in the p material or a hole in the n material because we have to push the particles against the field to get them to transfer from one material to the other. The transfer continues until states with the same energies in the two materials have equal occupation probabilities. We see that such equal occupation probabilities are consistent with the equivalent statement that the Fermi levels for electrons are the same now throughout the structure. Note that in equilibrium, with no external applied forces such as voltages or light beams, the Fermi levels for electrons and holes are equal; in that case there is no distinction between these two quantities saying that the occupation probabilities for electrons in states of given energy are the same throughout the structure is the same as saying the probabilities that electrons are not in a state of given energy are the same throughout the structure.
a) + + + + - + b) junction + + +

+ +

+ +

- + + + + + +

- - -

+ + +

- -

- + + + + + +

p-type

n-type

depletion region

ionized acceptor ionized donor

+ free hole

free electron

electric field

Fig. 8.2. (a) n-type and p-type semiconductor materials before joining. (b) Materials after movement of electrons and holes has occurred. Thus far, we have made no approximations. To proceed further, we need to make some simplifications. A very useful simplification is the depletion approximation. In its simplest form, the depletion approximation is that there is a region of the material lying between the p and n regions, and encroaching into them, which is totally depleted of mobile carriers. (This simplifying assumption of total depletion of mobile carriers in this region is the first element of this approximation.) In that region, we find ionized dopant (and possibly other impurity) atoms. At the edges of the depletion region (in the simplest version of the approximation), we immediately return to the approximately uniform n or p carrier concentrations in the n or p materials. (This immediate, abrupt return to the

EE41 Physics of Electrical Engineering

Chapter 8 Semiconductor diodes original doped values of the majority carrier densities is the second part of the approximation.) 1 The requirement of uniform chemical potential throughout the system tells us what voltage has to be dropped across the depletion region (the so-called "built-in" voltage, Vbi). The depths of the depleted regions on either side can then be deduced from simple electrostatics given the (presumably known) dopant concentrations, with the requirement that we achieve a voltage difference of Vbi from one side of the depletion region to the other. We will derive the details of this behavior below.

Fig. 8.3. (a) Depletion region in a diode with p doping slightly larger than n doping. (b) Resulting electric field (with no applied bias). (c) Resulting voltage distribution (with no applied bias). Fig. 8.3 shows the kind of results that arise if we make the depletion approximation for a simple pn junction at thermal equilibrium (with no applied bias). The electric field, which is zero in the conducting p and n regions because there is no current flowing at equilibrium, increases in magnitude (though it is negative in sign) as we move from the p conducting material into the depletion region on the p side because of the presence of the ionized acceptor charge density (negative in the case of acceptors); we can have a large electric field in this region precisely because it is depleted of mobile charge carriers and hence does not conduct. The electric field reaches a peak at the junction between the materials, and rises linearly back to zero as we progress through the ionized donors (positive charge density). Note incidentally that we presume the electric field is continuous. The electric field cannot make jumps since a jump at some point would require the presence of a sheet of charge inside the material at that point, and there generally is no such sheet of charge.

Note the depletion approximation is an approximation. The actual transition between essentially no mobile carrier density in the depletion region and a finite carrier density in the remaining doped regions is not exactly abrupt, but the thickness of this transition region is usually small compared to the size of the depletion region, and so it can be neglected as a first approximation.

EE41 Physics of Electrical Engineering

Chapter 8 Semiconductor diodes We already know that there is charge neutrality overall even without making the depletion approximation. In the depletion approximation, charge neutrality requires equal total numbers of ionized donors and acceptors in the depletion region. Those equal total numbers ensure the electric field comes back to zero in the n conducting material. The resulting voltage is the integral of the field through the depletion region, leading to a voltage rising in quadratic segments. We need to make one other simplifying assumption, which is that, at least approximately, we can view the band structure of a semiconductor (including, e.g., the so-called effective masses that electrons and holes have in such materials) as being unaffected by the presence of an electric field, except that we add the electrostatic potential to the whole band structure at each point in space. Hence we presume that we can draw the conduction and valence bands in the presence of an electric field, E, as shown in Fig. 8.4, where the magnitude of the slope of the conduction and valence bands in such energy diagrams is E electron-volts per unit distance. The increase in energy in the direction of the field comes from the fact that we conventionally plot electron energies in band diagrams; the electron has negative charge, and hence gains potential energy as it is pushed against the electrostatic force in the direction of the field (the choice of electron-volts for expressing energies is obviously very convenient in drawing such diagrams).
conduction band conduction band

electron energy position valence band with electric field no electric field E valence band

Fig. 8.4. Illustration of assumed tilting of the bands in the presence of electric field, simply adding the electrostatic potential to the electron energies. In examining the depletion approximation, let us ask, at least away from the very edges of the depletion region, is it reasonable to say there is negligible mobile carrier density in the depletion region? The answer is yes, at least in thermal equilibrium (and even also in practice when we are biasing the device, as long as we are not injecting substantial charge densities by forward current or, conceivably, with strong optical absorption). The main reason why there is very little mobile charge density in the depletion region is that the voltage dropped (which is of the order of the bandgap energy in eV, e.g., ~ 1.4 eV in GaAs) is much larger than the thermal energy (kBT/e in eV, e.g., ~25 meV at room temperature). Hence throughout the bulk of the depletion region, the carrier concentration is very low because the chemical potential is many kBT inside the bandgap region, "below" the "bottom" of the bands. For example, by the point at which the band edge has
B B

EE41 Physics of Electrical Engineering

Chapter 8 Semiconductor diodes risen 100 meV as a result of the potential rise in the depletion region as we move away from the conducting n region into the depletion region, the electron concentration will have dropped approximately by a factor e-4 (~1/50), making the electron charge density negligible compared to the ionized donor density (i.e., about 1/50 of the ionized donor density), and hence negligible for calculation of the electrostatics. Hence, the vast majority of the voltage drop between the conducting n and p regions occurs in material with essentially negligible free carrier density, thus justifying the depletion approximation, except possibly very close to the conducting n and p regions. Hence we have justified the reasonableness of the depletion approximation, at least as long as there is not strong injection of carriers (as there may well be under strong forward bias). Now we will review the simple and useful model of diode behavior that results from the depletion approximation. pn junction in the depletion approximation The form of the conduction and valence band edges for a junction between the p and n regions in the same material in the depletion approximation is shown in Fig. 8.5.
Conduction band Electron energy Depletion region eVbi

Valence band

Fermi energy

Fig. 8.5. Illustration of the form of the bands in a pn junction without bias. It is obvious from Fig. 8.1 and Fig. 8.5 that the value of the built-in voltage, Vbi, is the separation of the Fermi levels (chemical potentials) between the n and p materials. Vbi is also essentially the difference in energy (in electron-volts) between the conduction band edges (or the valence band edges) on either side. In practice, the Fermi levels (chemical potentials) in the n and p materials are almost always relatively close to (e.g., within several kBT of) the "bottom" of their respective conduction and valence bands), so Vbi is almost always relatively close numerically to the bandgap energy, Eg (in electron-volts).
B

Depletion layer width at zero bias To analyze the depletion layer width, we first define the junction to be at a point x = 0; the edge of the depletion layer on the p-side is at a point -xp, and the edge of the depletion layer on the n-side is at a point xn. The density of acceptors on the p-side is taken to be NA, and the density of donors on the n-side is ND. Because all of the mobile carriers are removed in the depletion region, the net charge densities are simply given by the charge densities from the ionized dopant atoms, and hence from Maxwell's equations we have in the depletion region on the p-side (-xp < x < 0) EE41 Physics of Electrical Engineering 6

Chapter 8 Semiconductor diodes


dEx e = N dx r o A

(8.1)

and on the n-side (0 < x < xn)


dEx e = ND dx r o

(8.2)

where Ex is the electric field (which is in the x direction in this one-dimensional problem, presuming the structure to be uniform in the y and z directions), and r is the relative dielectric constant of the material. The field must be zero just inside the conducting p and n materials, so the boundary conditions at x = -xp and x = xn are Ex = 0 at both of these points. We define the field at x = 0 to be some value Eo, so formally integrating Eq. (8.1) gives

Eo

dEx =

r o

eN A

xp

dx

(8.3)

i.e.,
Eo =

r o

eN A

xp

(8.4)

and similarly for the n-side of the junction


Eo =

r o

eN D

xn

(8.5)

From the known values of the fields, Eqs. (8.4) and (8.5), we can integrate the field to obtain the voltage (since dV/dx = -Ex). The voltage from the edge of the depletion region on the p-side, up to the junction is

Vp =

eN A 2 xp 2 r o

(8.6)

and similarly, the voltage from the junction to the edge of the depletion layer on the n-side is
Vn = eN D 2 xn 2 r o (8.7)

Now we know that, in the zero bias case we are considering,


V p + Vn = Vbi

(8.8)

and we also know from the overall charge neutrality


N A x p = N D xn

(8.9)

EE41 Physics of Electrical Engineering

Chapter 8 Semiconductor diodes

(a fact we could also deduce from equating (8.4) and (8.5)). Defining the total depletion layer width, wd,
wd = xn + x p ,

(8.10)

we have, substituting from Eq. (8.9) xp = and xn = wd N A N A + ND (8.12) wd N D N A + ND (8.11)

Substituting Eq. (8.11) into Eq. (8.6), and Eq. (8.12) into (8.7), then substituting those results into Eq. (8.8), after some rearrangement, gives Vbi = or
wd =

N A ND 2 wd 2 r o N A + N D
2Vbi r o N A + N D eN A N D

(8.13)

(8.14)

We see that the width, wd, of the depletion region grows as the square root of the built-in voltage, Vbi, and also tends to have a square root dependence on the dopant densities. For example, for NA = ND = 1017 cm-3 in a material with bandgap energy Eg(~ Vbi) of 1.5 eV, and r = 13, we would have wd ~ 0.2 m. The same material with NA = ND = 1016 cm-3 would have wd ~ 0.7 m. If we apply a reverse bias voltage, Vreverse, to the diode, the boundary conditions are still that Ex = 0 at the (new) edges of the depletion region, x = -xp and x = xn. We can follow through the same argument, using instead of Eq. (8.8)

Vp + Vn = Vbi + Vreverse
and hence we obtain, instead of Eq. (8.14)
wd = 2(Vbi + Vreverse ) r 0 ( N A + N D ) eN A N D

(8.15)

(8.16)

The depletion width now increases as the square root of Vbi + Vreverse.

EE41 Physics of Electrical Engineering

Chapter 8 Semiconductor diodes

conduction

eV valence p

EFe EFh

Fig. 8.6. Illustration of the band edges of a diode under forward bias, showing the electron and hole quasi-Fermi levels, EFe and EFh, approximately continuing into the depletion region, and decreasing into the conducting p and n regions. Current under bias As we add a bias voltage to the diode, we move away from a thermal equilibrium situation. We are now driving the system with some source of power. Fortunately, we do not have to get rid of all of our thermodynamic concepts as we do this, and we can retain the idea of a Fermi level. Now, however, it does not have to be the same throughout the structure. Additionally, again since we are no longer in thermal equilibrium, we can have hole and electron populations that are not the same as they would be in thermal equilibrium, and we can actually define different Fermi levels for electrons and holes. In this case, the electron and hole Fermi levels are sometimes called quasi-Fermi levels. The Fermi level is still a useful concept because, even though we are not in thermal equilibrium, the electrons can still behave as if they were in a thermal distribution; this is because there is very efficient scattering within the gas of electrons that means they do still have an effective temperature. There is similarly good scattering among the holes that allows them to behave as if they still have an effective temperature. But, quite importantly, the electron and hole gasses can behave as separate species, each with its own Fermi level (chemical potential). Though electrons and holes can recombine (and it is very important in diodes that they do), a process that tends to equalize the quasi-Fermi levels again, that process happens much less often than the scattering within the electron gas or within the hole gas. The net result of this is that, when we bias a diode, we basically pull apart the electron (quasi-)Fermi level in the n-type material and the hole (quasi-)Fermi level in the p-type material by an amount that is, in electron-volts, numerically equal to the voltage we apply. What then happens inside the depletion region is particularly interesting, and gives much of the important behavior of the diode. As we forward bias the diode, there will tend to be a higher value of electron density, n(-xp) just at the edge of the p conducting region (and similarly, a higher value of hole density, p(xn) just at the edge of the n conducting region). To know the current that is going to flow across the junction, we need to know the values of these minority carrier densities at the edges of the depletion region. The diffusion of these excess minority carriers into the conducting p and n regions is one of the major components of diode current (and in the simplest "ideal" diode model it is the only contribution to the current). The simplest useful approximation we can make is to assert that the electron quasi-Fermi level, EFe, remains approximately the same in the depletion region as it is in the n-doped

EE41 Physics of Electrical Engineering

Chapter 8 Semiconductor diodes

material, and, similarly, the hole quasi-Fermi level, EFh, in the depletion region is approximately the same as that in the p-doped material. This assertion is illustrated in Fig. 8.6, for the case of some (forward) bias voltage, V. This is a reasonable assertion, at least if we assume that the current flowing across the junction (from all possible causes) is small 2 . With very small current flowing, the occupation probabilities of electrons in states of given energies in the depletion region must be approximately the same as the occupation probabilities of electrons in states of the same energies in the n-doped conducting material (otherwise a large current would flow). However, farther inside the p conducting region, the electron density becomes lower because the minority carrier electrons will progressively recombine with the majority carrier holes in the p material, and at some large distance away from the junction the electron density becomes the very low equilibrium value of electron density found in a p-doped semiconductor. This minority carrier electron concentration tends to falls exponentially, with a characteristic distance of the electron diffusion length, Le, as we move further into the p-doped material. As a result, the quasi-Fermi level drops, eventually tending to the hole quasi-Fermi energy in the p conducting material. Similar processes happen for holes injected as minority carriers into the n conducting material. The electron and hole concentrations inside the forward biased diode are illustrated in Fig. 8.7.

Fig. 8.7. Illustration of the electron and hole carrier concentrations in a diode under forward bias. We are now in a position to calculate the forward diode current due to diffusion, at least for the case where we can approximate the electron and hole distributions as MaxwellBoltzmann distributions. Based on our assertion that the quasi-Fermi level for electrons is constant throughout the depletion region at its value in the n material, we therefore know the electron concentration just at the edge of the depletion region on the p side. Explicitly, if the electron concentration in the conducting n material is nno, then the electron concentration just at the edge of the p conducting region is

We also have to presume that recombination within the depletion region is small, a presumption that is likely not valid for light-emitting devices, and that generation of electrons and holes within the depletion region is also small, a presumption that is likely not valid for photodetector diodes, but this presumption will give us a good first model for electrical diodes at least.

EE41 Physics of Electrical Engineering

10

Chapter 8 Semiconductor diodes


n p x p = nno exp

d i

FG ebV V gIJ H kT K
bi B

(8.17)

since the electron quasi-Fermi level at that point is an energy distance e(Vbi - V) further away from the conduction band edge than it was in the conducting n material. The thermal equilibrium concentration of electrons in the p material is,
n po = nno exp

FG eV IJ H kT K
bi b

(8.18)

again since (deep into the p material) the Fermi energy in the p material is an energy distance eVbi further away from the conduction band edge than it is in the n material. Hence the excess number of electrons just inside the p material (compared to thermal equilibrium or to the concentration well away from the junction) is
n p x p = n p x p n po

d i d i

(8.19)

i.e.,

n p x p = n po exp

d i

LM F eV I 1OP N GH k T JK Q
B

(8.20)

Now we have stated above that the excess electron concentration falls off exponentially because the excess electrons progressively recombine with holes as they diffuse into the p-doped material, i.e., noting here that the exponential fall-off occurs for increasingly negative x because of the choice of coordinates in this particular case,

n p x = n p x p exp

af

c h FGH x + x IJK L
p e

(8.21)

where Le is the electron diffusion length in the p-type material. Hence the gradient of the electron concentration just inside the conducting p material becomes

dn p dx
xn

n po Le

LMexpF eV I 1OP N GH k T JK Q
B

(8.22)

From this gradient, we may now calculate the current per unit area (current density) from electrons diffusing into the p material, Jn, from our diffusion equation from the previous section, to obtain
Jn = eDe n po Le

LMexpF eV I 1OP N GH k T JK Q
B

(8.23)

where De is the diffusion coefficient for electrons in the p material. We can follow an exactly similar argument for the current density Jp, from holes diffusing into the n material, to obtain
EE41 Physics of Electrical Engineering 11

Chapter 8 Semiconductor diodes


Jp = eDh pno eV exp 1 Lh kBT

LM F I OP N GH JK Q

(8.24)

with analogous definitions of terms. Adding the electron and hole diffusion currents together gives
J = J s exp

LM F eV I 1OP N GH k T JK Q
B

(8.25)

where the so-called "saturation" current, Js, is Js = eDh eD pno + e n po Lh Le (8.26)

The expression Eq. (8.25) gives the classic exponential forward characteristic of an "ideal" diode (the "-1" becomes negligible for large forward currents). The forward current-voltage characteristic is sketched in Fig. 8.8.

Current Density, J

Js

Voltage

Fig. 8.8. Sketch of the current density, J, for an ideal diode. Though Eq. (8.25) was derived as if we were discussing forward bias, the basic result is also valid for the case of reverse bias voltages (in which case V is negative) when there is negligible recombination or generation of electrons and holes in the depletion region. In this case, we see an essentially constant reverse leakage current of magnitude Js at large reverse biases. This is a fundamental contribution to reverse leakage that cannot be avoided. Incidentally, the source of this reverse leakage can also be understood directly physically. Under strong reverse bias, the electron quasi-Fermi level in the depletion region is very far below the conduction band edge of the conducting p material. Hence there is approximately zero electron density in the depletion region near the conducting p material. np(-xp) is then -npo, independent of the value of the reverse bias. Instead of
EE41 Physics of Electrical Engineering 12

Chapter 8 Semiconductor diodes

falling exponentially into the p conducting material, the concentration now rises exponentially into the p region. Instead of electrons diffusing into the p material, they are now diffusing out of it. Instead of electrons recombining within the p region, they are now actually being generated within it by the inverse of the process that previously caused the recombination. The forms of the electron and hole densities are sketched in Fig. 8.9.

Fig. 8.9. Electron and hole densities for a diode under reverse bias.
Other currents in diodes

There are several other possible contributions to currents in diodes. For discussion of optoelectronic devices, two are particularly important, recombination current and photocurrent. Recombination current and light emission If there is any recombination of electrons and holes within the depletion region of the diode, there is a resulting forward current of one electronic charge for every electron-hole pair recombined. Note that such recombination is quite possible, because under forward bias in the diode, there can be quite substantial populations of both electrons and holes in the depletion region. Recombination constitutes a forward current, because an electron has been taken from the n material, and a hole from the p material, which is the same sense of current as forward current in the diode. Such recombination current simply adds (in the forward sense) to the diode current, and is a different mechanism from the diffusion current discussed above for the "ideal" diode. The diffusion current was associated with recombination in the p and n conducting regions explicitly outside of the depletion region. In a light-emitting diode or laser diode, ideally the forward current would all be recombination current from the radiative recombination process. Electrons will still be found near to the bottom of the conduction band, and holes will still be found near to the top of the valence band, so such radiative recombination, with an electron falling down from the conduction band to the empty electron state (i.e., the hole state) in the valence band, will tend to generate light with a photon energy approximately equal to the band gap energy of the material. Photocurrent and optical absorption Optical absorption in the depletion region generates extra pairs of electrons and holes. Optical absorption can take place for any photon energy above the band gap energy, essentially. Such absorption will take an electron from the valence band, and put it in the (essentially empty) conduction band, leaving behind a hole in the valence band. Because the net result is to create both an electron in the conduction band and a hole in the valence

EE41 Physics of Electrical Engineering

13

Chapter 8 Semiconductor diodes

band, we can also refer to this optical absorption as a process that creates an electron-hole pair. Under reverse bias, the electron and hole are swept out by the electric field to give a reverse current of one electron for every generated electron-hole pair (note that the electron going in one direction and the hole going in the other are in the end the same current; there is only one particle passing any given point, so there is only one electronic charge traveling around the circuit, not two). Usually, the time taken to sweep the carriers out under reverse bias is quite short. The electric field in the depletion region will likely be >104 V/cm (i.e., > 1 V/m) for a diode under reverse bias. That is typically enough to accelerate the carriers to the fastest speed they can go inside a semiconductor material, a number known as the saturated drift velocity, which is usually ~107 cm/s (~ 10ps for 1 m). Hence the sweep-out times are generally much faster for a reverse biased photodiode than any recombination times inside the depletion region (which are usually nanoseconds or greater), so the carrier collection efficiency for photocarriers generated inside the depletion region is generally very good, and in many situations we can very reliably collect one electron per photon absorbed in the depletion region. In forward bias, as long as the bias is small (e.g., significantly less than the built-in voltage), the photocurrent may also be close to one electron per photon absorbed in the depletion region, and the photocurrent still behaves as a reverse current (because the slope of the conduction and valence band edges is still in the opposite direction to the applied electric field), though the forward diffusion current needs to be added to get the total diode current. Hence, the current-voltage characteristic of a diode with light shining on it looks approximately as in Fig. 8.10. Normally, in passing a positive current through a diode, with a forward bias voltage (to the right in the figure), we are dissipating power in the diode, just as we dissipate power in a resistor, in an amount V times I (where V is the forward voltage and I is the forward current). Such power dissipation corresponds to being in the top right quadrant of the figure. But note that there is a range of forward voltages for which the current is negative, and we are in the bottom right quadrant. In this situation, the diode is generating power. This is called photovoltaic operation, and is the way in which semiconductor photocells generate power. The power is coming from the photon energy.

Current

Voc (open circuit voltage)

Reverse bias photocurrent (plus reverse leakage current)

Voltage

Isc (short-circuit current)

EE41 Physics of Electrical Engineering

14

Chapter 8 Semiconductor diodes

Fig. 8.10. Sketch of the current-voltage characteristic of a diode with light shining on it and being absorbed in the depletion region (forward bias is to the right, and forward current is towards the top) The voltage at which the curve crosses the axis is the open-circuit voltage, Voc, i.e., the voltage we would get in this photovoltaic diode if we conduct no current (basically, the voltage we would see if we simply connected a volt meter, and nothing else, across the diode). Similarly, the current at zero voltage is called the short-circuit current, Isc, i.e., the current we would get in this photovoltaic diode if we short-circuited it (basically, the current we would see if we put a current meter across the diode).
Summary Uses of diodes

In electronics, for rectifying signals, and for other circuit functions. In optoelectronics, to collect generated photocurrent, to apply strong electric fields, and to inject large densities of electrons and holes.
p-n junction diode

A diode formed by joining p-doped and n-doped semiconductor regions.


Equality of Fermi levels at equilibrium

With no external bias voltages, or any other external drive, and for a system in equilibrium, the Fermi level (chemical potential) for electrons is constant throughout the system, and the Fermi levels for electrons and holes are the same.
Depletion region

A region, between the p and n regions in a diode, that is essentially completely empty of mobile charge carriers (electrons or holes).
Depletion approximation

The approximation that there is a depletion region of finite thickness, extending into the p and n regions, that is exactly completely empty (depleted) of mobile charge carriers (electrons or holes).
Built-in voltage

The built-in voltage Vbi is numerically equal to the difference in the Fermi level positions of electrons and holes if we were somehow to line up the conduction bands (or valence bands) of the p and n materials of a junction diode. It is also the potential barrier height (in electron-Volts) seen by electrons in the n-materials trying to move to the p-material, and similarly it is the potential barrier height seen by holes in the p-material trying to move to the n-material.
Depletion layer width in a pn junction
wd = 2(Vbi + Vreverse ) r 0 ( N A + N D ) eN A N D

(8.16)

EE41 Physics of Electrical Engineering

15

Chapter 8 Semiconductor diodes Quasi-Fermi levels

The different Fermi levels that can be used in practice for electrons and holes even when they are not in thermal equilibrium (as, for example, when the diode is biased by an external voltage).
Current density in an ideal diode

The current, resulting from diffusion, in an ideal diode is


eV J = J s exp k BT 1

(8.25)

where Js =
Recombination current

eDh eD pno + e n po Lh Le

(8.26)

Current in a diode resulting from the recombination of electrons and holes in the depletion region. This behaves as a forward current (current in the same direction as the normal forward current of a diode).
Photocurrent

Current in a diode resulting from the generation of electrons and holes by optical absorption, usually inside the depletion region. This behaves as a reverse current (opposite in direction to the normal forward current of a diode). In the simplest model, such photocurrent shifts the diode I-V curve bodily down.
Open circuit voltage

Under illumination, the voltage Voc appearing across a diode when there is no current flowing.
Short circuit current

Under illumination, the current Isc flowing through a diode when there is no voltage across it.

EE41 Physics of Electrical Engineering

16

Chapter 9 Light sources and optoelectronic devices Chapter 9 Light sources and optoelectronic devices Human visual response Perhaps the largest use of light sources is for illumination so we humans can see. The human visual system has two kinds receptors on the retina, rods and cones. Rods are more sensitive, they are used for night vision, motion detection, and peripheral vision, and give no color information. Cones come in three types, which we can think of loosely as red, green and blue, and are responsible for our color vision. Cones are also concentrated heavily in the fovea, the central region of our vision in which we can see detail clearly, as we see in Fig. 9.1.

Fig. 9.1 Density of rods and cones on the retina. http://hyperphysics.phyastr.gsu.edu/hbase/vision/rodcone.html. The apparent spectral response of the different kinds of cones are shown in Fig. 9.2. If we look at the overall apparent sensitivity of the human eye, regardless of color perception, there are two kinds of response. One is the dark adapted response, known as scotopic vision, which is associated with the rods only. The other is the normal daylight response, known as photopic vision. The photopic response peaks at 555 nm. When we want to know how bright a light source is, as perceived by a human, we have to weight the actual optical power using the photopic curve. There are many units used to describe the many different quantities we might think of as brightness in an everyday sense. The science of the measurement of perceived brightness is called photometry. Luminous flux and lumens One key quantity is the luminous flux, measured in lumens. The lumen is the closest analog to the concept of light power. One lumen corresponds to 1/683 W at 555 nm. At other wavelengths, we have to weight the optical power by the photopic response curve to get the perceived luminous flux. For example, at a wavelength of about 650 nm, where the photopic response has fallen to about 10% of its peak value, we would need to have 1/68.3 W of power to give a luminous flux of 1 lumen. The number of lumens in a

EE41 Physics of Electrical Engineering

Chapter 9 Light sources and optoelectronic devices spectrum of light is therefore the power spectrum, multiplied by the photopic response, and integrated, and finally divided by 683.
1

Green Blue Red

0.8

Relative response

0.6

0.4

0.2

0 300

400

500

600

700

800

Wavelength (nm)

Fig. 9.2 Relative response of the Red, Green, and Blue cones in the human eye. http://cvision.ucsd.edu/database/text/cones/ss2_10.htm.
1

0.8

Scotopic (night vision)

Photopic (day vision)

Relative response

0.6

0.4

0.2

0 300 400 500 600 700 800

Wavelength (nm)

Fig. 9.3 Relative response of scotopic vision (night vision from rods), and photopic vision (day vision). http://cvision.ucsd.edu/

EE41 Physics of Electrical Engineering

Chapter 9 Light sources and optoelectronic devices Luminous intensity and candelas The lumen is a measure of the perceived power of the source itself, considering power radiated in all possible directions. Sometimes we are interested in how that perceived power is concentrated in angle. For example, an indicator light might be designed to be seen by a person directly in front of it, and then we would want a measure that took angle into account. That measure is the candela, which is lumens/steradian, and the name given to this quantity is luminous intensity. Illuminance and lux Sometimes we want to know how brightly a scene is illuminated. We are then interested in not the total perceived power, but how much there is per unit area. Then we use units of lux, which are lumens per square meter, and the associated quantity is called the illuminance. Lux would be the units we would use to determine if there was enough light for us to read and work at our desk, for example. The table below gives some typical illuminance values (after R. W. Boyd, Radiometry and the Detection of Optical Radiation, (Wiley, New York, 1982), p.97) Source Sun at zenith Clear sky Overcast sky Full moon at zenith Moonless overcast sky Thermal and non-thermal light sources All sources of light are quantum mechanical in one way or another. They can be broadly considered as belonging to two categories. The first category are thermal sources, sources that give off light because they are hot. The second category are sources that give off light because the atoms, molecules or solids have been excited into some high state by some non-thermal means. This second category can also gives rise the most efficient artificial light sources. Thermal sources and black-body radiation We know that when we heat objects up, they can emit light. The most obvious common example of hot objects as light sources are flames, as in candles and lamps that burn oil or gas. Coals or hot ashes in a fire can also be red or orange because they are hot. Metals are also often red, orange or yellow when they are molten, as is molten glass in a glass works. All of these objects that are giving off light when they are hot are limited in the amount of light they can give off at a given wavelength by something called the black-body radiation spectrum. A black body is a thermodynamic idealization, a body that would EE41 Physics of Electrical Engineering 3 Illuminance 1.2 x 105 lux (lumens/m2) 1 x 104 1 x 103 0.27 1 x 10-4

60W incandescent light bulb at 1 m 1 x 102

Chapter 9 Light sources and optoelectronic devices perfectly absorb electromagnetic radiation of any frequency. If we were to heat up such a black body, it would emit light in a spectrum that corresponded exactly to this black-body radiation spectrum. We will not give the full derivation of the black-body spectrum here, but we will sketch the key steps. As we mentioned above, it was exactly to understand the shape of the black-body radiation spectrum that the first step in quantum mechanics was taken by Planck. Classical thermal or statistical mechanics theory would say that there is an energy kBT for every degree of freedom in a system (this is known as the equipartition theorem). Every additional possible mode of electromagnetic radiation we consider therefore is adding more degrees of freedom into the problem, and hence more energy. As we keep considering higher and higher frequencies, we therefore get more and more energy we have to consider, and, in this classical model, the amount of thermal energy would become unbounded as we ask to consider all possible frequencies of electromagnetic radiation. (This problem was known as the ultraviolet catastrophe.)
B

By postulating that optical modes could only be in states with multiples of h of energy, Planck solved this problem. As h becomes much larger than kBT, the probability that there will be even one photon in a given mode drops exponentially by the Boltzmann factor exp(-h/ kBT). The energy per mode falls below the value suggested by the classical equipartition theorem. That fall off turns out to be faster than the growth of the number of modes with frequency , to such an extent that the total energy in the electromagnetic field at all possible frequencies becomes a finite number, and so the ultraviolet catastrophe is avoided. Such a theory predicts that, at a given temperature there will be on the average a particular number of photons in a mode, a number known as the Planck distribution 1
B B

n=

1 exp ( h / k BT ) 1

(9.1)

and a consequent average energy in a mode of nh .The number of electromagnetic modes in a given frequency range grows as the square of the frequency 2 , leading, after some care to get the constants right in the derivation, to a spectrum of electromagnetic wave power (in free space) at a given temperature, per unit frequency, that is crossing a unit area

2h 3 1 L = 2 c exp ( h / k BT ) 1

(9.2)

Now we come to a key conceptual point. Imagine then that we had a gas of such photons, in thermal equilibrium at some temperature T, and that we put it in contact with a perfectly absorbing body, also at temperature T. Then, if these two are to be in thermal

This distribution is a special case of the Bose-Einstein distribution, the thermal distribution we get for bosons, just as the Fermi-Dirac distribution is the thermal distribution we get for fermions.

This number can be deduced by considering a rectangular-sided box of some finite but large size, presuming that the electric field goes to zero at the walls, and counting up all of the possible threedimensional sine-wave-like standing waves that can fit in it in a given frequency range.

EE41 Physics of Electrical Engineering

Chapter 9 Light sources and optoelectronic devices equilibrium with each other, the energy absorbed by the body in a given frequency range must be exactly equal to the amount emitted by the body in the same frequency range. Hence, Eq. (9.2) also describes the power that is emitted by such a black body per unit area per unit frequency range. Furthermore, no body that can be described by a temperature can emit more power per unit area per unit frequency than this. If it did, it would start to add energy to the photon gas, heating it up, and that is forbidden by the second law of thermodynamics we cant heat up a body based only on heat transfer from a body at the same or colder temperature. We can integrate Eq. (9.2) over all frequencies to get the total electromagnetic power M emitted per unit area. After some algebra, we would arrive at
M = SBT 4

(9.3)

where SB is called the Stefan-Boltzmann constant

SB

2 5 k B 4 = 15h3c 2

5.67 108 W/m 2 K 4

(9.4)

Eq. (9.3) is known as the Stefan-Boltzmann law. Now we have found that there is a maximum electromagnetic power that can be emitted per unit area by a body at a specific temperature. This is a key problem for thermal light sources. Essentially, they can only be so bright. To make a brighter source, it is obvious that we should heat up the source more, and we would win greatly by doing so because of the underlying T4 increase in the emitted power. But there are practical limits to heating up materials if they are not simply to melt or evaporate, and this is the core problem of the design of ordinary tungsten light bulbs. The benefits of higher temperature become even stronger as we look at the spectrum of the emitted light from our ideal, black-body emitter. It is actually more common to express this as the power per unit area per unit wavelength (rather than per unit frequency). To convert from one to the other, we have to multiply by | d / d | , i.e., since = c / , we have to multiply by c / 2 , leading to the power per unit area per unit wavelength

L =

2hc 2

exp ( hc / k BT ) 1

(9.5)

Both Eqs. (9.2) and (9.5) are known as Plancks radiation law. Some differentiation allows us to calculate the wavelength at which the power per unit area per unit wavelength is maximum, i.e., the peak in this spectrum, which leads to

L (max)

2.898 103 T

(9.6)

To get a peak wavelength of ~ 500 nm, somewhere in the middle of the visible spectrum, needs a temperature of ~ 5800 K, which, not surprisingly, is approximately the temperature of the surface of the sun!
EE41 Physics of Electrical Engineering 5

Chapter 9 Light sources and optoelectronic devices

Fig. 9.4 shows the spectrum of the sun, both above the atmosphere, and for typical conditions at the surface. Fig. 9.5 shows various blackbody spectra at different temperatures, and Fig. 9.6 compares the solar spectrum with that from a tungsten bulb and from a candle. The sun is, to a good approximation, a black-body emitter. Specific wavelengths are missing from the suns spectrum by the time it arrives at the earth because of absorption by various gasses.

Fig. 9.4 Solar spectrum. Upper line outside the atmosphere; lower line at the earths surface.
0.25 Power/unit wavelength (arbitrary units)

0.2 5800 K 3000 K 0.15 3000 K x 20

0.1

0.05

0 0 500 1000 1500 2000 2500 Wavelength (nm)

Fig. 9.5 Black-body radiation spectrum at different temperatures. We cannot in practice make objects that are at temperatures of 5800 K all of our materials melt at such temperatures, and as a result, we cannot readily replicate the suns light with artificial thermal sources. We cannot even make a very efficient visible thermal source because the radiation from hot bodies tends to consist mostly of infrared light that we cannot see, and so the usable portion of the spectrum of tungsten light bulbs, for example, is a small number of percent depending on definitions and the specific bulb. Fig. 9.6 shows how the spectra of a tungsten bulb and a candle flame compare with the spectrum of the sun.

EE41 Physics of Electrical Engineering

Chapter 9 Light sources and optoelectronic devices

All this suggests that we have to move beyond thermal sources if we want bright, efficient and compact lighting.

Fig. 9.6 Comparison of the spectrum from the sun, a tungsten lamp, and a candle flame. http://acept.la.asu.edu/PiN/rdg/color/source.shtml
Non-thermal sources

To get lighting sources that are more efficient than simply hot bodies, the major approach has been to run an electrical discharge current through ionized vapors. Such a discharge current ionizes the atoms or molecules, or excites them to very high levels, and electromagnetic radiation is emitted as the atoms fall back into their non-ionized, lower states. Such sources need not obey the black body curve because they are generally not in thermal equilibrium. One very common example of such lamps is the fluorescent light tube, common in commercial and institutional lighting. It is much more efficient than a tungsten bulb. In the fluorescent tube, much of the light is emitted from the phosphors that coat the inside of the tube, rather than the direct emission from the gas discharge itself. The gas discharge often produces quite a lot of short wavelength light, including ultra-violet light. That light can optically excite the phosphors, which then decay to emit much of the light we see. The choice of phosphors is primarily what controls the color of the resulting fluorescent light. In applications such as street lighting, efficiency is particularly important, though color rendition does not matter so much. The lamps also need to be relatively small for the amount of light they emit. Much such lighting is done by metal vapor lamps. At their simplest such lamps contain a quantity of metal (such as mercury or sodium) that is heated up as the lamp is started, and eventually turns into a vapor that can be ionized by the electric discharge through the vapor. Low pressure versions of such lamps emit quite narrow spectral lines corresponding to the transitions in the metal vapor. Such low pressure lamps are often used as standards by which spectrometers (devices that measure wavelength) can be calibrated. Some geographical regions use sodium vapor street lamps at relatively low pressures, which emit light that is almost all from the so-called sodium D-lines, a pair of spectral lines at ~ 589 nm in the orange. Other regions have moved to
EE41 Physics of Electrical Engineering 7

Chapter 9 Light sources and optoelectronic devices

higher pressure lamps, which give broader spectra, and may also contain mixtures of different metal vapors, as well as some phosphors, so that color is somewhat more natural. The most efficient of these lamps is the high-pressure sodium lamp, which emits a somewhat peachy color. The spectrum of a high-pressure sodium lamp is shown in Fig. 9.7.

Fig. 9.7 Spectrum of a high-pressure sodium lamp. http://www.lighting.philips.com/nam/prodinfo/hid/p2055b.shtml. The mercury vapor lamps are among the least expensive of such lamps, and emit a somewhat harsh blue-white light. If not shielded, mercury vapor lamps can emit a lot of ultraviolet. Metal vapor lamps are also used extensively as grow-lights for plants because they contain more of the shorter wavelengths that are also present in sun-light, and they are very efficient.

Fig. 9.8 Conventional 5 mm LED


Light emitting diodes (LEDs)

Recently, light emitting diodes have begun to replace tungsten lamps in many specialized applications. They completely dominate the indicator light market now. Fig. 9.8 is a

EE41 Physics of Electrical Engineering

Chapter 9 Light sources and optoelectronic devices

schematic of the construction of the very common 5 mm LED indicator light. For other, higher power applications, light emitting diodes are more expensive initially per Watt of light emitted, but they are much more rugged, have much longer lifetimes, and are much more efficient. One major modern application is in traffic lights, for example. There is substantial reduction in cost of ownership for LED traffic lights, both from the energy saving, and from the reduction in maintainance and replacement required. Other common applications include tail light clusters in vehicles, and some high-brightness large-scale displays.

Fig. 9.9. Spectrum of an InGaN Green LED. http://ledmuseum.home.att.net/ledtur.htm LEDs are made almost exclusively from III-V compound semiconductors. Modern high efficiency LEDs are made from AlGaAlP alloys for the red and orange, and from AlGaInN alloys for the green and the blue. Individual diodes may not contain all of the alloy elements, e.g., we could have an InGaN green diode as in the typical green LED whose spectrum is shown in Fig. 9.9. This diode shows a typical spectral width of ~ 30 nm. This spectral width results primarily from the thermal statistical distribution of electrons in the conduction band, and holes in the valence band. The development of LEDs using the nitrogen-containing III-Vs was a major materials breakthrough in the 1990s. Such compounds permit efficient short wavelength emitters, and open up the possibility for the use of LEDs for full color applications (red, green, and blue). Fig. 9.10 shows the comparison of the luminous flux per Watt of electrical power (lumens per Watt) of modern LEDs, compared also with the corresponding numbers for various other lighting sources. LEDs are now at a level that is comparable with fluorescent tubes, and much better than incandescent (halogen and tungsten) bulbs.

EE41 Physics of Electrical Engineering

Chapter 9 Light sources and optoelectronic devices

Fig. 9.10 Comparison of the luminous flux per Watt of modern LEDs with various other light sources, plotted also with the photopic response curve (CIE). (M. George Craford (Lumileds)) http://www.lumileds.com/pdfs/techpaperspres/Opto2001.pdf

Fig. 9.11. Plot of the development of performance of LEDs. (M. George Craford (Lumileds)) http://www.lumileds.com/pdfs/techpaperspres/Opto2001.pdf It is also possible to make white LEDs by borrowing one of the techniques from the gas discharge lamps. A spectrum of such an LED is shown in In this case, an LED running in
EE41 Physics of Electrical Engineering 10

Chapter 9 Light sources and optoelectronic devices

the deep blue or ultraviolet also has phosphors added into the case. The LED light excites the phosphors, which then emit at shorter wavelengths, creating a whiter light. This particular white is relatively weak in the red compared to sunlight, for example, so it is a somewhat blue white.

Fig. 9.12. Spectrum of a white LED, based on a blue LED with added phosphors. http://ledmuseum.home.att.net/ledtur.htm In general, we can expect LEDs to take over progressively more lighting tasks as their costs are further reduced, and their efficiencies are further improved. Such use of LEDs can significantly reduce electric power consumption.
Constant brightness theorem (constant radiance theorem)

Thus far we have been discussing light sources mostly for use in indicators, displays, and illumination. All of the light sources we have discussed above in this section are what are known as incoherent sources. The light is emitted almost entirely by spontaneous emission, and that emission takes place into a very large number of modes (and hence a very large number of directions) of the electromagnetic field, with no particular relation between the phases of different photons in a given mode. For some applications, we would like to be able to take the light and concentrate it into a very small spot, for example to focus it into an optical fiber mode, or for an intense heat source for welding, or to get more light into some device for measuring, e.g., wavelength. There is a very basic principle that tells us just how well we can possibly do in such concentration, and it is known as the constant brightness or constant radiance theorem. It is a major limitation on how much we can concentrate incoherent light. Like a few other important limits, it is protected by the second law of thermodynamics. One statement of the constant brightness theorem is that no linear optical system can ever increase the power per unit area per unit solid angle per unit wavelength (or frequency), a
EE41 Physics of Electrical Engineering 11

Chapter 9 Light sources and optoelectronic devices

quantity sometimes known as the brightness. If we presume that it is the same frequencies that enter an optical system as leave it (i.e., there is no changing of power between frequencies, as is necessarily the case in a linear optical system), then we can say that the radiance (the power per unit area per unit solid angle) also cannot be increased. If the system is lossless, then the brightness and radiance are conserved (constant), hence the name of the theorem.

a1 Object (source)

a2 Image

Fig. 9.13. Illustration of imaging with a lens. It is easy to see this happening for imaging by a lens (see Fig. 9.13). Suppose we have a square area of size a1 a1 = A1 that is emitting light of power P into a solid angle that corresponds to the diameter d of the imaging lens, i.e., a solid angle d 2 / u 2 , where u is the object distance (the distance from the emitting area to the lens. (Remember that the solid angle subtended by some area relative to a point is the area on a spherical surface centered on that point, divided by the square of the distance from the point, i.e. the square of the radius of the sphere. Here we presume the angle is small, so we can approximate that spherical surface area simply by the perpendicular area.) The radiance at the original object is therefore Pu 2 / d 2 A1 . If the lens forms an image of the square area at a distance v from the lens, the magnification of the linear dimension of the image will be v/u, so the v v v2 new area will be A2 = a2 a2 = a1 a1 = 2 A1 . The solid angle of the converging light u u u going towards the image plane is now d 2 / v 2 . So, the radiance at the image square will be Pv 2 / d 2 A2 . But
Pv 2 = d 2 A2 Pv 2 Pu 2 = 2 d 2 A1 2v d 2 A1 u

(9.7)

i.e., the radiance has not changed. Though the imaging operation has changed the size of the image compared to the object, the solid angle has changed in exactly the opposite way, so that the radiance has not changed at all. If we want to conserve power, at least in this example, smaller images necessarily have larger angles of converging or diverging

EE41 Physics of Electrical Engineering

12

Chapter 9 Light sources and optoelectronic devices

light. Since there is an upper limit to the solid angle we can use (it can never exceed 4 steradians), there is a limit to how small we can focus all the power in a light beam. Of course, this statement above is just an example. Could we not manage to devise some other linear optical scheme that would get round this limit? The answer is that we cannot. We will not give the formal proof but the essence of one proof is as follows. If we could, we could devise a scheme that would enable us to start with two black bodies at the same temperature, and heat one up with the other, violating the second law of thermodynamics. Black bodies at the same temperature have the same brightness, which we could deduce directly from the Planck distribution (they all emit the same number of photons per mode per unit time at a given frequency, which we could use to show they all have the same brightness). If we could make an optical system that would take the light from one black body, and concentrate it to a larger brightness, then we could heat up the second body (since larger brightness of a black body corresponds to higher temperature). What does this mean in practice? Often we would like to take all of the light from a source and concentrate it into a small spot, for example, to focus it all down to a size of the order of the square of the wavelength, which is approximately how small we can make a spot of light, or to focus it all into a single mode fiber. The constant brightness theorem tells us that, for an incoherent source, we simply cannot do that. The only source that can do that is one that has all of its light in one single mode, not in the very large number of modes that we find in incoherent light. In fact, we could restate the constant brightness theorem in more modern terms to say that we can never combine the power of two mutually incoherent modes into one mode. The largest power we can get from an incoherent source into, say, a single mode optical fiber, is that largest power that occurs in a single mode to start with in the incoherent source. The only practical single mode light source is the laser, or sources derived from it.
Lasers

All lasers also belong to the second, non-thermal category of light source. Lasers require the stimulated emission to dominate over the absorption process, and that requires there be more population in the upper state than the lower state, i.e., population inversion. Population inversion can never be obtained merely by heating; because of the Boltzmann factor, higher energy states always have lower occupation probabilities than lower energy states. We will discuss some specific lasers in the next section.
Optoelectronic devices

Most of the modern optoelectronic devices, such as photodetectors that convert light to electrical signals, light-emitting diodes and laser diodes that generate light, and some kinds of modulators used in telecommunications, are made using semiconductor materials. Most of these devices are also based on diodes. Silicon is used very successfully for visible light detectors, as in modern charge-coupled device (CCD) arrays, which are at the core of most digital cameras and camcorders. Silicon photodetectors combined with CMOS (complementary metal-oxide-semiconductor) integrated circuits are also now being used in CMOS imagers. We will not deal further with the silicon devices in this section. Here we will concentrate on the III-V devices that are used for many high-performance applications and for all semiconductor light emitters.

EE41 Physics of Electrical Engineering

13

Chapter 9 Light sources and optoelectronic devices Heterostructures

The III-V devices take advantage of a major feature of III-V materials the ability to make heterostructures. Heterostructures are structures made from more than one kind of material (hetero- means different). The availability of a broad range of alloys of different combinations of the Group III materials (e.g., In, Ga, Al) and the Group V materials (e.g., As, P, N, Sb) allows a great deal of flexibility in designing and making such heterostructures. Primarily, the use of heterostructures allows us to choose the bandgap energy and also the refractive index of the different materials in the structure. Such choices are particularly useful in optoelectronic devices, as we will illustrate below.
p-i-n diode

We discussed above the idea of making a photodiode in which the photons absorbed inside the depletion region of the diode generate photocurrent that we can collect, hence changing optical power into electrical current. The very simple p-n diode we discussed above has several limitations as a good photodetector, however. First, the width of the depletion region is determined by the doping concentrations in the p and n regions. If we are to make those reasonably conducting, we cannot make the doping levels very low. As a result, the depletion region will not be very wide (e.g., perhaps a few tenths of a micron). Especially in the near infrared region where we do most telecommunications, this is not thick enough to have a high probability of absorbing a photon in most semiconductors. For photon energies just above the bandgap energies of the (direct gap) III-V semiconductors, it takes ~ 1 2 microns of thickness to have a high probability of absorbing a photon. Second, most photodetectors are illuminated from the top, and some of the light can be absorbed in the p or n layer on which the light is shining. Light absorbed in these regions is often not as efficiently collected as photocurrent, and so much of the potential photocurrent is lost. A solution to both of these problems is to make a p-i-n heterostructure photodetector, as sketched in Fig. 9.14. To increase the thickness of the depletion region, we insert a layer of undoped material (the i layer i for intrinsic) between the p and n layers. This material, with its very low background impurity concentration, is very easy to deplete, and, to a simple approximation, the depletion region is simply the i layer. We can readily make this layer 1 2 microns thick, for example, allowing for efficient optical absorption. To avoid undesired absorption of light in the p or n regions, we can make them out of a material with a larger bandgap energy than that of the i region. The InGaAs layer in the example will have a bandgap energy of ~ 0.75 eV (corresponding to ~ 1.65 microns wavelength) and the InP material has a bandgap energy of ~ 1.35 eV (corresponding to ~ 920 nm wavelength). Hence wavelengths in the range ~ 920 nm to 1.65 microns should mostly pass through the upper InP layer, and be absorbed in the InGaAs i region, from which the resulting generated photocarriers will be efficiently collected. This kind of structure is a typical one for the photodetectors used for long distance optical fiber communications, which typically use the 1.3 1.6 micron wavelength range where there are very low loss regions in the optical fiber.

EE41 Physics of Electrical Engineering

14

Chapter 9 Light sources and optoelectronic devices


Light in

p layer (e.g., InP) Electrical connections i layer (e.g., InGaAs) n layer (e.g., InP) Substrate (e.g., n-doped InP)

Fig. 9.14 Schematic illustration of a p-i-n diode photodetector.


Double heterostructure

The particular p-i-n diode we have sketched above is an example of what is called a double heterostructure. In a double heterostructure, there are two major interfaces between different materials. Here there are interfaces between the InGaAs layer and two different InP layers. Almost always, a double heterostructure will consist of a layer of narrower bandgap material between two layers of wider bandgap material, as in this example. We can view this as being like a sandwich, with the meat being the InGaAs layer in this case, and the bread being the two adjacent InP layers.
Double heterostructure laser diode

The single most important use of the double heterostructure is in the laser diode. Examples of double heterostructures are shown in Fig. 9.15, here illustrated for AlGaAs/GaAs heterostructures. There are three reasons for the use of the double heterostructure in the laser diode, one being to help contain the electrons and holes within the active region to improve the chance of stimulated emission, a second being to make a waveguide and hence confine the optical mode of the laser so that it overlaps with the active gain region, and a third being that the optical emission from the active layer is not absorbed in the other layers of the structure because they have a larger bandgap energy. There are several different ways a double heterostructure can be designed for use in a laser, two of which are illustrated in Fig. 9.15. In these examples, a narrow bandgap GaAs layer, lightly doped either n or p, is sandwiched between wider bandgap AlGaAs layers. In these examples, the narrow bandgap material is predominantly not in the depletion region. It is therefore conducting, and has therefore little field across it. Under forward bias, however, the relatively lower energy of the GaAs layer means that electrons and holes injected into it tend to be confined within the GaAs layer by the bandgap discontinuities at the edges. This greatly increases the density of electrons and holes in the active GaAs layer compared to the simple homojunction case for a given forward current density. This increased density in turn leads to higher gain.

EE41 Physics of Electrical Engineering

15

Chapter 9 Light sources and optoelectronic devices

Fig. 9.15. Illustration of various AlGaAs - GaAs double heterostructures. (a) The basic double heterostructure diode. (b) A structure where the GaAs is lightly p doped, at zero bias, and (c) in forward bias. (d) A structure where the GaAs is lightly n doped, at zero bias, and (e) in forward bias. (f) The refractive index in the heterostructure, showing the higher index in the GaAs (after Wood). At its simplest, a typical semiconductor laser is formed from a semiconductor diode and a pair of plane-parallel mirrors. In operation, the diode is forward biased so that relatively large electron and hole populations are present near the diode junction. In particular, for at least some of the possible transitions, we have injected so many electrons and holes that we have an inverted population the probability of an electron being in a particular state in the conduction band is higher than the probability that there is an electron in the corresponding state in the valence band and so we can have stimulated emission dominating over absorption. As a result, instead of net absorption, we get net (stimulated emission) gain in the material. If the gain per pass exceeds the mirror transmission loss and any other losses experienced by the beam (e.g, diffraction, absorption loss in nominally transparent parts of the structure, loss from scattering off material, or structure imperfections), the structure will lase.
Semiconductor laser structures

There are two basic configurations, edge-emitting and surface emitting.


Edge-emitting lasers

The edge-emitting laser usually is based on a waveguide structure, as illustrated in Fig. 9.16.

EE41 Physics of Electrical Engineering

16

Chapter 9 Light sources and optoelectronic devices

Fig. 9.16. Simple stripe geometry laser. First a slab waveguide is formed from different semiconductor layers to give waveguiding in one direction. In Fig. 9.16, the slab waveguide consists of the p and n AlGaAs layers, surrounding the GaAs layer. The AlGaAs is essentially transparent at the laser operating wavelength, and has a relatively lower refractive index than the GaAs. Such a structure, with a higher index layer surrounded by two lower index layers, is exactly what we need in order to make a simple waveguide. Light propagating approximately along the GaAs layer in this case will tend to bounce of the interfaces with the lower-refractive-index AlGaAs layers. Hence not only does the double heterostructure tend to confine the electrons and holes in the GaAs layer, increasing the probability of inverted populations and stimulated emission gain, it also confines the light in the same layer, further increasing the stimulated emission efficiency. The mirrors in such a laser are usually formed from the natural reflectivity of the semiconductor-air interface, which is of the order of 30%. They are typically formed by the simple technique of performing two crystallographic cleaves in the semiconductor crystal the required few hundred microns apart. Cleaves can be induced by scratching (scribing) the semiconductor surface approximately along crystallographic directions, and bending the semiconductor until it snaps. The cleaved surfaces have the advantages of perfect smoothness and parallelism. These plane-parallel mirrors form a Fabry-Perot cavity (which is the name we give to a resonator formed by two plane parallel mirrors), and such lasers are known as Fabry-Perot lasers. To obtain enough gain per pass to overcome this relatively large mirror loss, the laser cavity needs to be typically 100s of microns long. Heat dissipation is always a problem in semiconductor laser structures since relatively high current densities (e.g., 100s of A/cm2 or more) are required to generate sufficient carrier densities in the diode. In most edgeemitting laser diodes, therefore, it is desirable to confine the current injection and the optical mode in a relatively narrow stripe to minimize the total dissipation, and to allow for some heat spreading. (In modern very-high-power lasers, however, the threshold current densities have been reduced sufficiently, and the heat sinking improved, so it can be possible to run relatively large structures continuously). Also, confining in a narrow stripe gives a mode shape that is more nearly the same size in both directions, as is desirable if we want to couple into, for example, optical fibers. (In practice, though, there is still a substantial difference between the beam size in the two directions). Hence, a long narrow contact stripe is often used. In Fig. 9.16, the top metal layer only contacts the diode region in a narrow stripe opening in the insulating oxide layer. The injection of
EE41 Physics of Electrical Engineering 17

Chapter 9 Light sources and optoelectronic devices

current into this narrow stripe can itself cause a weak guiding effect in the lateral direction, giving rise to a gain-guided laser. In modern lasers, this gain guiding is usually supplemented by refractive index guiding to give index-guided lasers. A common index guiding technique is to etch a ridge in the top cladding layer of the slab guide, which tends to give a higher effective index for the laser mode in the region just below the ridge, hence giving some lateral waveguiding. A simple ridge structure cross-section is illustrated in Fig. 9.17.

Fig. 9.17. Ridge laser structure cross-section.

Fig. 9.18. Etched mesa buried heterostructure laser cross-section. A more sophisticated index-guided structure is shown in Fig. 9.18. Here the index is larger in the center partly because there is only the low bandgap active material (InGaAsP) present. In this structure, a deep mesa ridge is formed in the original layered material, down to just below the active region; then the additional InP layers are regrown on the sides, burying the active heterostructure (hence the name buried heterostructure). This kind of structure is particularly efficient at injecting carriers only into the active region in the middle of the laser mode. There are many variants of the buried heterostructure concept.

EE41 Physics of Electrical Engineering

18

Chapter 9 Light sources and optoelectronic devices

Fig. 9.19. GRINSCH structure, showing (a) the layer sequence, (b) the variation of the conduction and valence band energies through the structure (neglecting the diode aspects), and (c) the variation in refractive index. It is not necessarily optimum in a laser to use the same size of structure to confine the carriers as to confine the optical mode. A separate confinement heterostructure is a more sophisticated heterostructure in which a greater number of different layers of material are added, with some of the layers being primarily present to guide (or confine) the optical mode, and some being there primarily to position the electron and hole populations optimally for gain. One example is the Graded Index Separate Confinement Heterostructure (GRINSCH), illustrated in Fig. 9.19. In the GRINSCH, the material is graded approximately quadratically around about the thin active region. The approximately parabolic grading of index gives good control over the waveguide mode profile, and the thin active layer with deep potential wells for electrons and holes results in good overlap of the excited electron and hole populations for strong gain. The active region is also in the middle of the optical mode where the amplitude is highest, and hence the effective gain is also highest. Modern telecommunications lasers are typically somewhat more sophisticated than a simple Fabry-Perot laser, employing mirrors that reflect only at very specific frequencies. This allows greater control of the lasing wavelength, which is particularly important if we want to use multiple different wavelengths, as in so-called Wavelength Division Multiplexing (WDM), or if we simply want to control the wavelength very well for predictable long-distance communications. One trick is to use so-called Bragg mirrors. Normally, we might think of mirrors as being the metallic surfaces of normal domestic mirrors. Such mirrors are not often used for lasers because they are too lossy. The natural reflectivity of dielectric surfaces gives mirrors with low actual loss, but usually quite low reflectivities (~ 30 % for materials such as semiconductors, which typically have refractive indices of ~ 3 4, and ~ 4 % for glass, with its index of 1.5). If, however, we make multiple layers of two different materials, each a quarter wavelength thick, and with one having a low refractive index and the other a high refractive index, the various relatively small reflections at each
EE41 Physics of Electrical Engineering 19

Chapter 9 Light sources and optoelectronic devices

dielectric interface tend to add up in phase to give a high reflectivity overall, though only for a specific range of wavelengths. In the edge emitting lasers we have been discussing so far, rather than depositing layers of different materials, we merely form a corrugated interface between two materials of different refractive index. To a beam propagating sideways through this structure, it looks as if the refractive index is changing back and forth between two different values, and hence we can get the effect of a Bragg mirror. In this case, the mirror is called a distributed Bragg reflector (DBR). It has a particularly narrow range of frequencies over which it reflects, and hence can be used to control the wavelength of the laser very precisely. Fig. 9.20 illustrates a laser formed this way.

Fig. 9.20. A distributed Bragg reflector laser structure (after Chuang) [1.4].
Vertical-cavity surface-emitting lasers

The vertical-cavity surface-emitting laser (VCSEL) is like a DBR laser, but made in the vertical direction. The motivations for making the VCSEL are not so much to obtain narrow-linewidth, single-frequency operation, but more to make lasers that can have intrinsically circular beam profiles, therefore making them easier to interface to fibers, and to allow the construction of arrays of lasers. VCSELs are also very small compared to edge emitters, because they avoid the long waveguide region. Fig. 9.21 illustrates some VCSEL structures. VCSELs have been enabled partly by the development of low-loss mirrors made integral to the semiconductor structure. These mirrors are formed from alternating quarter-wave layers of low and high index transparent semiconductors. Such a mirror, in addition to being an example of a Bragg reflector, is also know as a quarter-wave stack. Most modern semiconductor lasers use very thin active regions. This is partly for simple classical reasons. If we only have a small volume of active material, then we only have a small volume that we have to excite in order to get gain to exceed loss (at least to exceed the loss of the material itself). If we only have to excite a small volume, then we only need a small amount of injected carriers, and hence we may only need a small drive current, allowing lower threshold lasers. This benefit of thin gain regions can be used to special advantage if the cavity mirrors have high reflectivity so that there is little cavity loss also, a fact employed in vertical cavity lasers. These very thin regions are so thin (e.g., 10 nm) that they also acquire additional quantum mechanical effects. The electrons in these layers behave like particles in a box, just like the simple quantum mechanical model we have discussed above. Such thin layers of a
EE41 Physics of Electrical Engineering 20

Chapter 9 Light sources and optoelectronic devices

low bandgap material sandwiched between two larger bandgap materials are known as quantum wells. There are various additional benefits of this quantum mechanical confinements, including the ability to tune the effective band gap energy by yet another means, and also some additional benefits in improving laser gain. Almost all modern high-performance laser structures now use quantum-well active layers.

Fig. 9.21. VCSEL structures. (a) Emitting through an etched hole in the substrate. (b) Emitting through the top of the structure. (c) Emitting through a transparent substrate.
Modulators

Optical modulators can change the transmission or reflection of a light beam, or, in some cases, re-route the light beam along a different path, usually by applying a voltage to the modulator. They can avoid some of the problems encountered when modulating lasers at high speeds, such as "chirping" (imposing a frequency sweep as the beam power changes). Such chirping increases the frequency bandwidth of the light, which can lead to increased dispersion of the light pulses as they travel through long optical fibers, for example, limiting the bit rate that can be transmitted over long distances. Modulators do not have the gain dynamics found in lasers that results from the interplay of carrier injection and cavity photon build-up, and as a result may be able to be used at higher speeds in some situations. Modulators also allow the beam from a single optical power source to be split up and used for a large number of information-carrying beams, and in general give us an alternate way of imposing information on light beams. Semiconductor optical modulators can be broadly separated into two categories absorptive, which operate by changing the amount of the incident light that is absorbed in the structure, or refractive, which work usually by changing the optical path length of a beam and relying on a wave interference or propagation effect to thereby change the

EE41 Physics of Electrical Engineering

21

Chapter 9 Light sources and optoelectronic devices

transmission, reflection, or direction of a beam. Semiconductor modulators are usually capable of high speeds, and in some cases the underlying modulation mechanisms can operate even into the sub-picosecond regime, much faster than is typically possible with direct modulation of lasers themselves.
Quantum well modulator

Here we will discuss just one example of a modulator, the quantum well electroabsorption modulator. These are used in some high speed telecommunications systems, and are the subject of research for other applications, such as interconnects within computers.
Quantum-confined Stark effect

If we make semiconductor quantum wells, and if we apply electric fields perpendicular to the layers, then we effectively skew the potential well, as shown in Fig. 9.22. If we do so, we have to solve a somewhat more complicated differential equation than we did for the simple potential well, because of the linearly varying potential inside the well. There are various ways of doing such solutions, including an analytic answer (the result is technically a set of functions known as Airy functions). The net result is that we obtain the skewed wavefunctions shown on the right of Fig. 9.22.

Fig. 9.22. Electron and hole wavefunctions for the first few states in an "infinite" quantum well. Without field, the wavefunctions are sinusoidal; with field they are Airy functions.

EE41 Physics of Electrical Engineering

22

Chapter 9 Light sources and optoelectronic devices

Note also, however, that, in addition to skewing the wavefunctions, the electric field has also changed the energies for the various energy levels. In particular, the energy separation between the lowest electron level and the highest hole level has decreased. These changes in the electron and hole wavefunctions lead to changes in the absorption spectrum of the material. This effect is known as the quantum-confined Stark effect (QCSE).

15000 Absorption Coefficient (cm-1) 0V 3V 5V 10000

5000

0 825

835

845 855 W avelength (nm)

865

Fig. 9.23 Typical QCSE absorption spectra for a quantum well diode at various bias voltages. We can see the results of these shifts in the absorption spectra of Fig. 9.23. The absorption coefficient is the probability per unit length that a photon will be absorbed in the material. An absorption coefficient of 10,000 cm-1 means that there is a large probability (strictly 1 1/e where e here is the base of the natural logarithms) of the photon being absorbed in 1 micron of thickness. As the voltage is increased across the quantum wells, the absorption at, for example, 850 nm wavelength increases substantially. The amount by which the peaks are shifting in this effect is predominantly from the shifts of the energy levels as in Fig. 9.22. The QCSE is particularly attractive for optical modulators. A simple device structure is shown in Fig. 9.24. A quantum well region, typically containing 50 to 100 quantum wells, is grown as the undoped "intrinsic" ("i") region in a p-i-n diode. This quantum well region will therefore have a thickness of about 1-2 microns. In operation, the diode is reverse biased. Substantial QCSE shifts and changes in the absorption spectrum can be made with applied voltages of the order of 5-10 V. The reverse biased diode is convenient because there is no conduction current that needs to flow in the diode in order to apply these relatively substantial fields. As in the p-i-n photodiode, the "p" and "n" regions of the diode are made out of a material that is substantially transparent at the wavelength of interest. In the case of GaAs quantum wells, they would most likely be made out of AlGaAs, the same material as used for the quantum well barriers.

EE41 Physics of Electrical Engineering

23

Chapter 9 Light sources and optoelectronic devices

Such modulators are very fast and compact. They are also made in waveguide structures, in which case they are rather like quantum well lasers, but are reverse biased rather than forward biased. Such waveguide modulators are the ones used for telecommunication applications. n
+ve i p -ve

lig ht o ut

lig ht in

to p c o nta c t (p -AlG a As) q ua ntum w e lls (und o p e d ) sub stra te (n-G a A s) b o tto m c o nta c t (n-AlG a As)

Fig. 9.24. Quantum well modulator diode structure. Some current research is investigating whether such modulators would also be good ways to get information out of CMOS integrated circuits. Future CMOS circuits, with high clock speeds and high densities of information to be communicated in and out of the chips, will be increasingly constrained by the limited ability of wires to carry information, especially over longer distances. One potential solution to that is to use light beams instead. There are many challenges in making practical systems. Quantum well modulators have been successfully attached to silicon CMOS chips, and used for interconnections between chips. Arrays of modulators, each with metallic mirrors beneath them to allow reflective operation, are shown in Fig. 9.25.

Fig. 9.25 Arrays of quantum well reflection modulators solder-bonded to silicon CMOS chips (Gordon Keeler and Noah Helman, Stanford University)

EE41 Physics of Electrical Engineering

24

Chapter 9 Light sources and optoelectronic devices Summary Rods and cones

Rods and cones are the light receptors in the eye. The three different cones, red, green and blue, are responsible for color vision, and are concentrated in the center (fovea) of the visual field. Rods are more sensitive, but give no color information, and are found over the entire visual field, but with relatively few in the center.
Photopic and scotopic response

Photopic and scotopic response are the relative response as a function of wavelength of the human eye in day and night vision respectively.
Luminous flux and lumens

Luminous flux is a measure of optical power, weighted by the photopic response, and measured in lumens. At 555nm, one lumen is 1/683 W.
Luminous intensity and candelas

Luminous intensity is luminous flux per unit solid angle, measured in candelas (lumens per solid angle), and is a measure of the perceived brightness of a source when looking directly at it.
Illuminance and lux

Illuminance is luminous flux per unit area, measured in lux (lumens per square meter).
Black body

A body that would perfectly absorb electromagnetic radiation of any frequency. Such a body would also emit the maximum electromagnetic radiation possible over any frequency range for a body at equilibrium at a given temperature.
Planck distribution

At a given temperature, the average number of photons in a mode of frequency is


n=

1 exp ( h / k BT ) 1

(9.1)

Plancks radiation law

The electromagnetic power per unit area per unit frequency emitted by a black body is

L =
or per unit wavelength

2h 3 1 2 c exp ( h / k BT ) 1

(9.2)

L =

2hc 2

exp ( hc / k BT ) 1

(9.5)

The power per unit area per unit wavelength is a maximum at

EE41 Physics of Electrical Engineering

25

Chapter 9 Light sources and optoelectronic devices

L (max)
Stefan-Boltzmann Law

2.898 103 T

(9.6)

The total electromagnetic power emitted per unit area by a black body is
M = SBT 4

(9.3)

where the Stefan-Boltzmann constant is

SB =

2 5 k B 4 15h3c 2

5.67 108 W/m 2 K 4

(9.4)

Constant brightness and constant radiance theorem

No linear optical system can ever increase the power per unit area per unit solid angle per unit wavelength (or frequency), a quantity sometimes known as the brightness. If there is no changing of power between frequencies, as is necessarily the case in a linear optical system, then we can say that the radiance (the power per unit area per unit solid angle) also cannot be increased. If the system is lossless, then the brightness and radiance are conserved (constant). In particular, this theorem limits how tightly we can focus or concentrate the light from light sources, especially incoherent light sources. A modern restatement is that we can never combine the power of two mutually incoherent modes into one mode.
Heterostructures

Structures made from more than one kind of material.

EE41 Physics of Electrical Engineering

26

Chapter 10 Transistors Chapter 10 Transistors So far, we have discussed the physics behind a number of common electronic and optical devices, including electrostatic sensors, diodes, and optoelectronic devices. Perhaps, however, the most important of all of the devices based on this science is the transistor. It has driven the electronics and information revolution that has changed life dramatically in the past few decades. Every reader of this book likely owns several billion transistors. There are many different kinds of transistors, but here we will concentrate on the one that dominates nearly all modern electronics. This is the metal-oxide-semiconductor transistor, known as MOS for short, which consists of a gate, sometimes made from metal, an insulator layer, usually made from an oxide, all on top of the semiconductor, usually silicon. Actually in modern versions of this transistor it is more likely that the gate is not a metal material, being instead highly doped poly-crystalline silicon (polysilicon for short, or just poly), but the metal name remains. These transistors exist in two flavors, one associated with electron conduction (NMOS), and the other with the hole conduction (PMOS). Together these two kinds of transistors make up the complimentary MOS or CMOS technology that is used for the vast majority of present integrated circuits. The base material of all of these devices is silicon, together with some metals and other insulating materials. The reason that we can each have so many of these transistors is that they are incredibly cheap, costing perhaps $10-7 to $10-8 each. The reason why they are so cheap is because they can be manufactured with very high yields in large numbers at once to make the integrated circuits. Here are we will discuss some of the history, show how modern transistors are made, discuss the physical principles that underlie them, and discuss briefly the digital models and simple logic circuits for which they are most commonly used. Because of the background we have already discussed in electrostatics, quantum mechanics, semiconductor physics, and various devices, we need no new physical principles to understand how these transistors work. History of the development of the integrated circuit We have seen the history of the development of our understanding of electricity and magnetism, primarily in the nineteenth century. Also in the nineteenth century, the ideas of calculating machines were being considered. For example Charles Babbage designed a Difference Engine that could at least in principle perform calculations. A picture of a version of that engine built much more recently by the Museum of Science in the United Kingdom is shown in Fig. 10.1. In the nineteenth century however we still did not have the technology needed to make the kinds of switches that would allow us to make any electrical version of calculating machine, and, indeed, only in the twentieth century did even mechanical technology become good enough to make practical mechanical calculators. The technology of electrical communications was, however, developing in the nineteenth century. Telegraphy was common, and by the end of the nineteenth century even long undersea cables over very long distances were able to carry useful information. Of course, also, the telephone had been invented and brought to practical reality. The need to make switching systems for telephony stimulated much electromechanical technology. EE41 Physics of Electrical Engineering 1

Chapter 10 Transistors Such technology in the form of electrically operated switches, such as relay switches, was sophisticated enough to make simple electromechanical machines. In the Second World War, the electromechanical technology was developed to the point of making relatively sophisticated computers that were used to break codes.

Fig. 10.1 Picture of a version of the Babbage difference engine built by the Museum of Science UK. The calculating section of Difference Engine No. 2 has 4,000 moving parts (excluding the printing mechanism) and weighs 2.6 tons. It is seven feet high, eleven feet long and eighteen inches in depth Also in the early to mid twentieth century, vacuum tubes were invented and brought to quite a high degree of sophistication. They allowed the construction of practical electronic amplifiers and other circuits, and could also be used to make digital switches. In the 1930s and 1940s, the core ideas of modern computing were laid down by mathematicians such as Alan Turing and John Von Neumann. At the end of the Second World War, therefore, various pieces of technology were in place to attempt the construction of the first machines that would look something like a modern electronic computer. It was, however, true that even the pioneers of that day had little realization of how computers would be used in our modern world. Famously, Thomas Watson Sr., then an executive at IBM, predicted in the 1940s that there might be a need for only five or so computers in the entire world. At that time, computers were still envisaged as machines to perform calculations, such as the calculation of ballistics tables for military gunnery. Serious attempts were made to make computers with vacuum tubes, though it soon became clear that the limited lifetime of vacuum tubes and their power consumption limited the practical size that such machines could be made. Vacuum tubes must heat a wire filament to emit electrons, and a good vacuum, usually inside a glass container, is required for the operation of the devices. Such devices are therefore intrinsically fragile, and also the heated filament would eventually burn out just as in a light bulb. Despite these problems, vacuum tubes have successfully survived even to the present day, where they are still used for some television tubes or computer monitors, for some high powered devices, and for the klystrons that power microwave ovens. EE41 Physics of Electrical Engineering 2

Chapter 10 Transistors The desire to make a more robust replacement for the vacuum tube and the strong advances that had been made in quantum mechanics and, hence, solid-state physics, in 1920s and 1930s set the necessary preconditions for the invention by Shockley, Bardeen, and Brattain in 1947 of the first transistor. This device worked with electrons in solids, and no heater was required. Transistors were successfully developed into small and reliable components, though in their discrete form, the circuits in which they were used were at least initially not much more complex than the vacuum tube circuits that they replaced. Transistors in this form were, however, viable as ways of making quite functional small computing systems. To make the kinds of computers and complex systems we work with today required another breakthrough. There needed to be some way of making progressively larger and larger numbers of switching devices, in complex circuits, in a way in which the cost did not simply scale linearly with the number of devices we wished to make. That subsequent breakthrough was the invention of the integrated circuit. The integrated circuit has multiple electronic components and their interconnecting wires manufactured on a single substrate. Jack Kilby at Texas Instruments produced the first integrated circuit in 1958, as shown in Fig. 10.2.

Fig. 10.2 Picture of the first integrated circuit, as made at Texas Instruments in 1958. Robert Noyce was also one of the key inventors of the integrated circuit. His patent was a little later than Kilbys, but he used silicon and also he used a so-called planar process. In a planar process, all of the parts of the transistor and the wiring essentially are laid out on a plane, instead of some of them being stacked vertically as in a layered structure. Of course, even in a planar process, there are layers underneath, but the whole circuit can be fabricated by a series of steps performed plane after plane. This ability to make the circuit solely on the basis of a succession of planar processes is what enables us to use the technique of lithography in making very large integrated circuits. Noyce had been one of the founders of Fairchild Semiconductor, and went on to found Intel. The early transistors had been made using germanium. Use of germanium transistors continued on into the 1960s and beyond, and transistors made from III-V materials such

EE41 Physics of Electrical Engineering

Chapter 10 Transistors as gallium arsenide or indium phosphide have also been made and are used today in particular high-performance applications. Compared to these other materials, silicon has several positive features. First, it is inexpensive. There is no shortage of silicon, which, of course, is one of the main elements in sand (sand is primarily silicon dioxide). It also turns out not to be very difficult to grow silicon crystals, even in very large physical sizes. Perhaps the dominant reason, however, for the use of silicon for integrated circuits is that the interface between silicon and silicon dioxide can be a particularly clean one, with very few interface states. This fact is crucial for the practical fabrication of the MOS transistor, especially when it has to be made in the very large numbers needed for modern integrated circuits. Silicon dioxide (glass) is also a very convenient and effective insulator. It is very stable, and can be patterned quite effectively. There are chemical etches that can remove silicon dioxide without harming silicon, which is very convenient for planar processing.
Start with wafer at current step

Spin on a photoresist

Pattern photoresist with mask

Step specific processing etch, implant, etc...

Wash off resist

Fig. 10.3 Generic set of processing steps in fabricating an integrated circuit using lithography. Fig. 10.3 illustrates the kind of sequence of steps that is involved in the lithography used to make integrated circuits. At any given step in the process, we may start with a planar layer of the material that we wish to pattern on the surface of the semiconductor wafer. Then we spin on a photoresist, which is a liquid polymer material. In such a process we can effectively pour some of the photoresist liquid onto the surface of the wafer, and then spin the wafer, rather like a high speed record turntable, to leave a thin layer of approximately uniform thickness. The photoresist on the wafer is then baked in an oven to make it solid. Then short wavelength light, usually in the ultraviolet, is shone through a mask onto the photoresist. This exposes the photoresist in the same way as a piece of photographic film. The photoresist is then developed, just like a film, and the exposed material can be removed by a chemical etch, leaving the desired pattern on the surface. Sometimes this remaining layer is itself a hard mask for some other process such as implantation or diffusion of dopants. The photoresist itself will be washed off before further processing. There are, of course many variations on this process, and what we show here is merely the underlying concept of such lithographic fabrication. EE41 Physics of Electrical Engineering 4

Chapter 10 Transistors As illustrated schematically in Fig. 10.4, there will be many successive steps of such planar fabrications, using many different masks successively on the same wafer to fabricate the transistors and the wiring that goes between them. The final processed wafer will contain many replicas of the chips, laid out in a regular pattern on the surface. Finally, the wafer will be cut up to make the chips. The number of transistors made will depend on the resolution of this lithographic printing process, and the size of the chip.
Masks

Chips Processed Wafer

Processing Wafers Chemicals

Fig. 10.4 Illustration of the sequence of steps from the starting masks and silicon wafers to the final dicing of the processed wafer into chips.

Fig. 10.5 Original graph from Gordon Moores 1965 paper [1] showing the empirical Moores Law of the exponential growth of the number of transistors on a chip. A key point about this whole lithographic approach is that it has proved possible to keep scaling the fabrication to ever smaller dimensions, while retaining the same basic approach to the manufacture. Fig. 10.5 shows the 1965 version of a graph drawn by Gordon Moore [1] that shows the exponential increase of the number of components on a EE41 Physics of Electrical Engineering 5

Chapter 10 Transistors chip. This exponential growth has come to be known as Moores law. This early graph shows a growth by a factor of two every year. That growth has slowed down somewhat to about a factor of two every year and a half, but it has been overall a remarkably constant growth over the last four decades. Early integrated circuits used bipolar transistors. The bipolar transistor was the first transistor invented, and they are still used extensively today. They led to many different families of logic circuit, including resistor-transistor logic (RTL), diode-transistor logic (DTL), transistor-transistor logic (TTL), current-mode logic (CML), and emitter-coupled logic (ECL). Compared to these bipolar transistors, MOS transistors were initially hard to make. They had in fact been proposed before the bipolar transistor, but no one could make them work initially. One key point about them is that they require conduction at or very near to the surface of the semiconductor, which requires a very clean and stable surface to be prepared. That surface problem was eventually solved because of the quality of the silicon to silicon dioxide interface.

Fig. 10.6 Graphs showing the proposed scaling of transistor performance from Dennards 1974 paper [2], comparing a technology with 5 m gate length, and one with 1 m, showing the scaling of the resulting set of curves. MOS transistors also have a large advantage in scaling. Basically, it was possible to keep essentially the same design of transistor, but keep simply scaling it down. MOS devices operate on electric fields. Electric fields are simply a voltage divided by a length. Current densities, for a given doping of semiconductor, are simply proportional to electric field, at least at low to moderate fields. This is the drift model of transport, governed by a mobility, . So the relation between electric field and current density is the same EE41 Physics of Electrical Engineering 6

Chapter 10 Transistors regardless of the size of the transistor. Hence, if, as we reduce the size of the transistor in all three dimensions, we also reduce all the voltages proportionately, then the current vs. voltage characteristic of the transistor retains the same shape, but is simply scaled (see Dennard [2]). Fig. 10.6 illustrates this point with curves from this original scaling paper [2]. The consequences of this scaling for the transistor are that the transistor can get approximately proportionately faster as the size scale is reduced, and the energy required to turn the transistor on and off, which is essentially the energy required to charge its internal capacitances, can reduce by as much as the cube of the size reduction.

Fig. 10.7 The CRAY-1 computer The results of this continued scaling in the number of transistors and in their performance has been dramatic in the development of the technology. For example, in 1976, the CRAY-1 (Fig. 10.7) was the worlds fastest computer, with 65 MB of memory, and approximately 1 million gates. It ran at 80 MHz clock speed, and consumed 115 kW of power. In a typical technology in the year 2004, which would be a 0.13m length scale technology, we could put all of this memory, and all of these gates, on one silicon chip of size approximately 5 x 5 mm. Essentially, we could put the entire CRAY-1 on one chip, and probably run it much faster too, with substantially less power. We are now at the point where this so-called constant field scaling model is ending. There are several reasons for this, one of which is that we cannot simply continue to reduce operating voltages. Eventually we run into the physics that the thermal energy at room temperature is equivalent for an electron to ~25 mV of voltage. Another reason is that making the transistor smaller and smaller leads to increasing quantum mechanical tunneling through the gate oxide. Other aspects of the integrated circuit also do not scale in the same way as the transistors. For example, scaling a wire down in all three dimensions leaves us with exactly the same resistive-capacitive time constant for the wire by a simple scaling argument, which means that the wires do not keep up with the transistors in scaling the speed.

EE41 Physics of Electrical Engineering

Chapter 10 Transistors There are various other challenges to the continuing scaling of silicon CMOS. As we are capable of making more transistors, the difficulty of designing each chip increases. Hence chip design becomes more expensive as chip complexity rises. As we try to make ever smaller transistors, we need a printing process with ever finer resolution, and, ideally, ever increasing yield in working transistors. The factories to make these integrated circuits become extremely expensive, with costs in the scale of billions of dollars. Additionally the cost of making the photolithographic masks for each new chip also rises. The results of these increasing costs do still allow us to make transistors very cheap individually as long as we can sell very large numbers of them. This in turn means that, to be economically successful, we need to create silicon integrated circuits that have very large markets. Processors and memory are universal components that can be used for many applications. Hence they fit the necessary model of parts that we can sell in extremely large numbers. The consequences of the exponential growth of silicon integrated circuit technology and markets lead to remarkable numbers in terms of production, some so large that it is difficult for us to comprehend them. A few comparisons, due to Gordon Moore [3] may help here. Even by the mid 1990s, more transistors were made each year than raindrops in California. 1 Moore also reckons that as of 2003 there were about 1018 transistors made each year, and each year we make more transistors than existed at the beginning of the year. Moore also noted that the Harvard biologist E. O. Wilson, a famous expert on ants among other biological things, estimated there are about 1016 to 1017 ants in existence, and so each year we make about 10 to 100 transistors for every ant on the face of the earth. The expected progress of future generations of chips is charted in the International Technology Roadmap for Semiconductors, an evolving document that is regularly revised and updated by a team from the semiconductor industry. [4] MOS transistors basic structure and a very simple model Now we will look at how an NMOS transistor works, starting with a very simple model. (The PMOS transistor works in the same way but with the sign of the dopings reversed and with holes instead of electrons.) A schematic diagram of the transistor is shown in Fig. 10.8. Here we see a structure with the electrical terminals, a source, a gate, and a drain. This source and the drain are each formed by heavy n-type doping (n+) of the semiconductor in these regions. Typically for an NMOS transistor, the underlying semiconductor material is lightly p-doped. The basic idea is that we are going to control the amount of current that can flow between the source and the drain by changing the voltage on the gate. The gate structure itself is made from a layer of conducting polysilicon above an insulating gate oxide layer. That gate oxide sits on top of a region known as the channel. In the channel, in the operation of this device, we will have a layer of electrons in the channel right next to the gate oxide. Since we want the device to be off when there is no voltage on the gate, we need the region between the two diffusions to be p-type material. This will create two back to back diodes that prevent current flow. To get current to flow through the transistor, we will need to locally change the surface region under the gate so that it contains electrons and hence forms a conducting channel that connects the two diffusion regions. We will
1

That is not meant to be a statement about how good California weather is!.

EE41 Physics of Electrical Engineering

Chapter 10 Transistors discuss the formation of this so-called inversion layer of electrons in a p-type material in much more detail a little later in this section. The net effect of this inversion layer is that it is almost as if we are inverting the effective doping of the silicon from p to n, though we accomplish this without actually changing the chemical doping itself, as we will see below. For now, we will make a very simple model of the transistor, viewing it as a voltage controlled switch. The gate is one terminal of a capacitor and the channel (the region between the two diffusions) is the other. Changing the voltage between the gate and the channel will change the amount of charge on both plates of this capacitor. A positive voltage on the gate will increase the number of positive charges in the polysilicon layer and as is usual in a capacitor, correspondingly (and equally) increase the amount of negative charge (electrons) in the channel. The first amount of negative charge accumulated in the channel is fixed, and not mobile electrons. This is the ionized p dopant near the surface. When the gate voltage reaches some threshold, usually called the threshold voltage or Vth, the thin layer of mobile electrons forms. As the gate-tochannel voltage increases above Vth, more mobile charge is present, which creates more carriers to carry current between the source and the drain. Hence we expect that increasing the gate-to-source voltage will increase the ability of the transistor to conduct current between the source and the drain.

Gate Source
n+ poly gate oxide channel p

Drain
n+

G S D

Fig. 10.8 An idealized NMOS transistor schematic cross-section, showing the gate (G), source (S) and drain (D), and a circuit schematic picture corresponding to this transistor. In this very simple model, then, we can look upon the NMOS transistor as being like a resistor between the source and drain, with the value of that resistor controlled by the gate voltage. This is illustrated schematically in Fig. 10.9. The way that we design a transistor for the common case of a so-called enhancement mode device is one in which the threshold voltage is positive for an nMOS device, which means that the resistance is very high when the gate voltage is zero or below Vth, and above this threshold gate voltage, we have conduction between the source and the drain. In this very simple model, then, we can view the NMOS transistor as simply being like a voltage controlled switch that is turned on when the gate is connected to the positive supply voltage (Vdd) and the source is connected to ground (Gnd). Of course, we must choose Vdd so that it is larger than Vth. For PMOS devices the situation is completely EE41 Physics of Electrical Engineering 9

Chapter 10 Transistors analogous, but the voltages are all negative, since we need to accumulate positive charge in the channel. Hense the threshold voltage for a PMOS device is negative, and we need a negative voltage on the gate to turn it on. While generating these negative voltages sounds difficult, remember that voltages are all relative. If we connect the source and the substrate of the PMOS device to Vdd, then a gate voltage of Gnd is a gate to source voltage of -Vdd. While this sounds awkward, it is actually very helpful since it creates a switch that is turned on when the gate is connected to ground (Gnd). This kind of simple model is illustrated in Fig. 10.9.
Gate

Source

Drain

Fig. 10.9 A MOS transistor with low source-drain voltage, operating as a resistor whose value is controlled by the gate voltage.

D G S

D G=1 (Vdd) S NMOS transistors

D S PMOS transistors G=0 (Gnd)

Fig. 10.10 Simplest circuit model for NMOS and PMOS transistors, viewing them as simple switches turned on by setting the gate voltage either to the top supply voltage (Vdd) in the case of NMOS, or to 0 V (i.e., ground (Gnd)) in the case of PMOS. Logic circuits with CMOS With the simple switch model considered above, we can think of Vdd as corresponding to logic 1, and ground (Gnd or zero Volts) as corresponding to logic 0. In making logic functions, we want to build simple Boolean operators using the MOS devices. In doing so, we need to follow a few simple rules. First of all, we must make sure that the output of our circuit is always connected to some voltage rail, either Vdd or ground. We must also make sure that inner circuits logic zero (Gnd) and logic one (Vdd) are never connected to one another. The basic idea of how we construct a circuit that always obeys these rules is shown in Fig. 10.10. We need some object that, when the input is of a particular logic value, pulls the output to Vdd, and another object that pulls the output to ground when the input is the opposite logic value. Connecting these two objects in series as shown in Fig. 10.11 will achieve the desired results without breaking the rules. The simplest case of such a circuit is the inverter, shown in Fig. 10.11.

EE41 Physics of Electrical Engineering

10

Chapter 10 Transistors

pull to 1

in

out

pull to 0

Fig. 10.11 Basic idea of the use of transistor groups in logic, with the upper transistor or group pulling the output to logic 1 if it is turned on, or the lower transistor or group pulling the output to logic 0 if it is turned on instead.

in

out

Fig. 10.12 Schematics of a CMOS inverter, with the upper schematic showing the transistors explicitly, and the lower schematic showing the inverter as an inverting amplifier (with the circle at the output indicating inversion). In this case of an inverter, when the input is high (logic 1) then the PMOS transistor is off and the NMOS transistor is on, hence driving the output to a low value (logic 0). On the contrary, when the input is low, the PMOS transistor is on and the NMOS transistor is off, hence the driving the output to a high value (logic one) . Note that there is no steady current that flows from Vdd to ground in the circuit in either of the two cases of the output being either logic 1 or logic 0. This feature of no steady current in either logic state is a particularly important benefit of the CMOS circuitry in reducing power dissipation in logic circuits. There is of course current that flows whenever the circuit is changing from one state to another, so there is some power dissipation in the use of these circuits, but there is no static power dissipation if the circuit is not changing state. An example of a NAND gate circuit is shown in Fig. 10.12. Here we see an interesting combination of parallel PMOS transistors and series NMOS transistors. The reader can check that both the A and B inputs must be high if the output is to be low, which is exactly the function required for a NAND gate. The number of parallel PMOS transistors and series NMOS transistors can be increased to obtain a larger number of inputs for the NAND gate. We see again with this circuit that the output is either always driven by one

EE41 Physics of Electrical Engineering

11

Chapter 10 Transistors or more PMOS transistors pulling the output high, or by the NMOS transistors pulling the output low.

out

Fig. 10.13 Schematics of a 2-input NAND gate. The circuit corresponding to a NOR gate is shown in Fig. 10.13. In this case, if any of the inputs a, b, or c. is high, the output, y, is pulled low. Again there is no set of inputs that will result in both an NMOS transistor and all of the PMOS transistors being on at the same time
a b c y

Fig. 10.14 Schematics of a 3-input NOR gate.


A out A B B out

Fig. 10.15 Proposed schematics for AND and OR gates that are in fact bad designs that will not work in general because of the actual behavior of the transistors. Based on the simple model for the transistors we have considered so far, one might imagine that the circuits shown in Fig. 10.14 would lead to working AND and OR gates, but in fact, because of the actual working of the transistors, these circuits will not work as EE41 Physics of Electrical Engineering 12

Chapter 10 Transistors drawn. The basic reason is that one needs a finite voltage,Vth, between the gate and the channel for the transistors to conduct. If you look at these circuits carefully, you will notice that when the output is Vdd Vth the NMOS transistors cannot be on, and when the output is Vth above Gnd the PMOS devices turn off. Thus the output swing is less than the input swing, which is very bad for a logic gate. Of course, for any detailed understanding of how are these circuits work, we need a more sophisticated model of the MOS transistor. We will discuss this later below.
Implant N-Well

Etch away silicon where there will not be transistors

Deposit oxide in trenches

Polish top of wafer flat and expose silicon

Implant gate and grow oxide


Implant sets turn-on voltage of transistor to the correct value.

Deposit and Etch Polysilicon for gate

Implant source and drain


Diffusion self aligns to edges of gate

Coat Poly and Silicon with metal to reduce Resistance (C i)

Fig. 10.15. Schematic drawings for a CMOS fabrication process for transistors. Left - cross-sections. Right - plan views of the different levels of mask layouts Fabrication of CMOS transistors

A typical sequence for the fabrication of CMOS transistors is shown in Fig. 10.15 above. Because we want to fabricate PMOS transistors as well as NMOS transistors, we need to EE41 Physics of Electrical Engineering 13

Chapter 10 Transistors implant n-dopants to make an n region, the n-well, into the p-doped substrate material. This implantation effectively turns what was p material into n material in that well region by overwhelming the p doping with n doping. (The technique of using an amount of opposite doping to counteract the existing doping is called compensation). We will form the PMOS transistor inside this well, just as we formed the NMOS transistor inside the p-substrate material. The next step is to etch away the silicon material in the regions where there will not be transistors formed. Next an oxide is deposited and the surface is polished flat again, exposing the top of the silicon regions. Next the gate is formed, first by implanting to control the precise dopings in the channel region, followed by growth of the gate oxide, then deposition of the polysilicon for the gate. This polysilicon is then etched to form the shape of the gate. The gate itself is then used as a mask in what is called a self-aligned masking step for the implantation of the source and drain regions with appropriate dopants. Finally the polysilicon and the silicon source and drain regions are coated with metal to reduce the resistance. In actual fabrication of transistors the resulting structure is somewhat more complex than shown on the simplified diagrams given here but the basic form of the fabrication sequence is similar. Pictures of actual transistors are shown in Fig. 10.16.
TSMC 0.13 TI 0.09

Source: Source: Nikkei Nikkei Microdevices Microdevices 11/00 9/02

Fig. 10.16. Cross sections of actual transistors from two different manufacturers. Fabrication of wires The other aspect of integrated circuits that is of equal importance to the fabrication of transistors is the fabrication of the wires that will connect them. As mentioned above, one of the key points about an integrated circuit is that both of these kinds of components can be fabricated within the same set of fabrication steps. A fabrication sequence for the wiring is shown in Fig. 10.17. First, a dielectric layer is deposited. Then the dielectric layer is patterned as shown to create dielectric structures that will eventually lead to separation of metal regions, a step shown as the etching of metal trenches in the figure. Then trenches are etched in the dielectric to expose the contact metal on the source and drain in this case. Metal is then deposited everywhere, and the whole wafer is polished to remove the very top part of the metal, hence isolating different regions of the metal. Similar processes can be repeated for multiple different layers of metal and wiring, including the vias, the metal regions that connect one layer of wiring to another. Actual fabricated wires are shown in Fig. 10.18.

EE41 Physics of Electrical Engineering

14

Chapter 10 Transistors

Deposit Dielectric

Etch metal trenches

Etch via trench

Deposit metal

Polish surface back

Fig. 10.17. Schematic drawings for a fabrication process for wires on an integrated circuit. The drawings on the left are cross-sections, and those on the right are plan views of the mask layouts in the different levels required.
TSMC 0.13 8 layers Cu TI 0.09 6 layers Cu

Source: Nikkei Microdevices 11/00

Source: Nikkei Microdevices 9/02

Fig. 10.18. Pictures of actual wires and wiring layers in integrated circuits.

EE41 Physics of Electrical Engineering

15

Chapter 10 Transistors Finally, in Fig. 10.19 we show a picture of an actual fabricated circuit shown from the top. Note that, in this actual circuit, there is considerable rounding of the various edges that we have shown as idealized rectangles in all of the diagrams.

TSMC 0.13 SRAM cell (Diffusion and poly)


Source: Nikkei Microdevices 11/00

Fig. 10.19. Picture of an actual fabricated SRAM (static random access memory) cell. MOS transistors a more advanced model

gate oxide

F
Ec

VFB polysilicon gate (n-doped)


t ox

Ei

p-doped silicon substrate

EF Ev

Fig. 10.20. Band diagram of an NMOS transistor at the flat band condition. To understand the MOS transistor in more detail, we need to look more closely at what happens near the interface between the channel and the gate oxide. Fig. 10.20 shows an idealized band diagram under flat band conditions. The oxide has an energy gap of about 9 eV, making it a very good insulator. Electrons in silicon see a barrier of about 3.2 eV, and holes see about a 4.7 eV barrier. (Silicon itself has a ~ 1.12 eV bandgap.) Fig. 10.20 is not to scale, especially for the insulator, which will have a much higher conduction band edge than shown.

EE41 Physics of Electrical Engineering

16

Chapter 10 Transistors The polysilicon gate is highly n-doped so that it is conducting. The substrate is p-doped. We presume that we have applied a flat-band voltage VFB to the gate to get to this condition. For the situation drawn here, since we are plotting electron energy in band diagrams, VFB is a negative voltage compared to the substrate voltage. This diagram indicates the position of the Fermi levels, EF, in both materials. 2 As usual, because we have applied a voltage to the structure, the separation of the Fermi levels between the two sides is equal to the applied voltage. The conduction band edge, Ec, and the valence band edge, Ev, are both shown. (In all of these band diagrams, we can use voltages and energies interchangeably as long as we think of the energies in electron-volts, and we will take that approach here.) Conventionally in discussions of MOS transistors, the position the Fermi level would have in an undoped material, Ei, the intrinsic Fermi level, is shown, and is used as a reference for the potentials. This is an arbitrary choice we could also have used the conduction band edge, the valence band edge, or any other point as the reference for the potentials but it does lead to a symmetry in the discussions of NMOS and PMOS transistors since this point is essentially in the middle of the band gap.

F
Ec

Vg

Ei

EF Ev

Fig. 10.21. Band diagram for more positive (less negative) gate voltage, but still below the point of the formation of the inversion channel. Now let us change the voltage across this junction, moving from the negative flat-band voltage VFB to some more positive voltage V. Because of the insulating properties of the oxide, there will be no current flow across it (or we may at least presume so for the purposes of this model). The structure therefore behaves like a capacitor. Making the potential more positive means that we are adding positive charge to the capacitor plate just to the left of the gate oxide in the conducting polysilicon. We must therefore induce some corresponding equal amount of negative charge in the p-doped silicon substrate (channel) side. The process by which this takes place is similar to the formation of a depletion region in a diode. We end up removing the mobile holes from the region just next to the gate oxide, leaving a depletion region containing the ionized, negatively charged acceptor dopant atoms. Since the density of these dopant atoms is fixed, to get more charge we need a larger depletion width. The result is a band curvature in the substrate region next to the oxide, a curvature that exists for the width of the depletion
2

Importantly, because there is negligible injection of charge through the insulating gate oxide, the electrons and holes are in equilibrium with each other in the substrate that is, we do not have separate quasi-Fermi levels for electrons and holes in the substrate layer, so we only need to talk about one Fermi level in this material.

EE41 Physics of Electrical Engineering

17

Chapter 10 Transistors region. (For simplicity here, we have assumed the polysilicon is very highly doped, so the depletion that might otherwise have formed near the surface there is neglected.) This is illustrated in Fig. 10.21. Now if we keep making the voltage more positive, at the interface with the oxide, the potential of the conduction band edge will move up in voltage (and thus down in electron energy), and eventually it will get close to the Fermi level. At this point, even though we are dealing with a p-doped material, there will start to be significant numbers of electrons in the conduction band. This is called the formation of an inversion layer, inversion because we have ended up changing the important charge carriers from being holes to being electrons. It is this thin layer next to the oxide that becomes the conducting channel in an NMOS transistor. This point is illustrated in Fig. 10.22.

inversion layer

F
Ec Ei EF Ev

Vth

Fig. 10.22. Formation of in inversion layer of conducting electrons in the p-doped silicon substrate. The gate voltage needed to form this inversion layer is the threshold voltage, Vth, that we used in our simple switch model of a transistor. This voltage is typically designed to be in the range of ~ 0.2 - 0.6 V in modern devices. After this voltage is reached, the extra negative charge that is needed to match or image the positive charge on the gate comes predominantly from the mobile electrons in the channel. Thus the amount of mobile charge in the channel can be related to the capacitance of the gate structure, and the voltage from the gate to the channel minus the threshold voltage. It is not quite true that this threshold is an absolutely abrupt process. Remember that the density of carriers is set by Fermi-Dirac statistics, and is approximately exponential on the voltage difference between the Fermi energy and the conduction band edge. There are therefore always some mobile electrons near the surface, and their concentration just approximately exponentially increases as we increase the band bending. Even when the gate-to-channel voltage is below Vth there still are mobile carriers in the channel that exponentially increase with voltage, so there is not a specific voltage where the transistor turns on. 3 Yet the definition of Vth is useful since it demarks a significant change in how the channel charge changes with voltage.

In fact this current that flows is an important problem with modern circuits, since it causes transistors to leak even when their gate to channel voltage is at zero. This leakage is called sub-threshold conduction.

EE41 Physics of Electrical Engineering

18

Chapter 10 Transistors When the gate voltage is increased and hence charge is added to the gate, an equal amount of charge altogether must appear on the substrate side of the gate oxide. There are two types of negative charge in the silicon, ionized fixed dopant atoms, and mobile electrons in the channel. When the gate-to-channel voltage is below threshold, the amount of mobile charge is so small that changing the surface potential (the voltage at the silicon surface, right at the interface between the substrate and the oxide) by bending the bands more creates only a small amount of additional mobile charge. The positive gate charge is mostly imaged by more fixed ionized dopant atoms being exposed as the depletion thickness is increased. As we continue to increase the positive gate bias, because of the increased band bending, the position of the conduction band edge approaches the Fermi level. Remember that moving the conduction band nearer to the Fermi level by only ~25 mV at room temperature will increase the electron concentration by approximately a factor of e ~ 2.7 (at least for Fermi levels that are still several kBT below the conduction band). As a result, the number of mobile electrons in the channel is rising approximately exponentially, and eventually we reach the point where moving the conduction band edge by just a very small amount causes a large change in the mobile channel charge. At this point the situation is reversed, with the additional negative charge coming primarily from the additional mobile charge (electrons) in the inversion channel, rather than from additional fixed charge (ionized dopants) being exposed by further depletion. As the voltage gets larger, the mobile channel charge therefore becomes approximately linearly dependent on the gate voltage. This situation is illustrated in Fig. 10.23.
B

inversion layer

F
Ec Ei EF Ev

Vg >Vth

Fig. 10.23. Inversion layer for gate voltages Vg above (more positive than) the threshold voltage, Vth. The threshold voltage, Vth, is often defined as the zero intercept of the channel charge vs. gate voltage, if one extrapolates from the region where the transistor is in inversion. 4
It is also common in discussions of transistors to consider the threshold in terms of the Fermi potential F, the Fermi level relative to Ei far to the right in the p-doped substrate. We know that this position of the Fermi level was sufficient to give us moderate p-type conduction. If we apply sufficient positive bias voltage to the gate, we can pull the conduction band just inside the substrate down to such a point that the Fermi level is ~ F above the intrinsic level, Ei. If having the Fermi level F below Ei was enough that the material could be called p-type and have reasonable hole densities for hole conduction, then having the
4

EE41 Physics of Electrical Engineering

19

Chapter 10 Transistors The magnitude of the charge Q on either plate of a capacitor of value C is

Q = CV
for a voltage V, or equivalently, the magnitude of the charge per unit area, QA, is

(10.1)

QA = C AV

(10.2)

for a capacitance CA per unit area. Hence, for a gate voltage Vg, then the electron charge per unit area in the inversion layer will be, for Vg > Vth ,

Qn

Cox (Vg Vth )

(10.3)

where Cox is the capacitance per unit area of a capacitor formed by the oxide and two plates on either side of it. The minus sign in Eq. (10.3) is because we are explicitly considering the charge on the negative plate of the capacitor. The model we have been considering so far has been assuming no drain-to-source voltage on the transistor. We can sketch a picture of the transistor in this condition as shown in Fig. 10.24. Here we have a uniform charge density, Qn, under the gate, as given by Eq. (10.3) for a gate bias, Vg (relative to the source), that is above threshold.

Vg +

Qn y y =0 y =L

Fig. 10.24. Sketch of an NMOS transistor, with the gate biased with a voltage VGS, above threshold, and with no drain-to-source voltage (drain, source and substrate are all connected to ground), showing a uniform charge density, Qn (per unit area), in the inversion layer in the channel under the gate. For zero or small drain-to-source voltage VDS, the transistor behaves like a voltage controlled resistor, with the value of the conductance between the source and the drain

Fermi level F above Ei should be enough to say that we have sufficient electrons present in the conduction band for electron conduction. Conventionally, then, it is also common to say that the threshold voltage, Vth, is the gate voltage that corresponds to this condition, as illustrated in Fig. 10.22, at least for the purposes of simple models.

EE41 Physics of Electrical Engineering

20

Chapter 10 Transistors being approximately linear with respect to the gate bias because the charge density is approximately linear with respect to gate bias.
3

Drain current, iDS

saturated regime
2

linear regime
0 0 1 2 3

Drain-source voltage, VDS Fig. 10.25. Idealized MOS transistor curve of drain current (in units of (W/L)Cox) as a function of the drain-source voltage, VDS, for a particular gate voltage above threshold (here Vg = Vth + 2 Volts). The kind of characteristic we get from a MOS transistor, in a somewhat idealized form, is shown in Fig. 10.25. The drain current, iDS, here is plotted in units of (W/L)Cox, where W is the width of the transistor, L is the gate length (i.e., the separation between the source and the drain), and is the electron mobility. We do indeed see a linear region, with the drain current rising proportionately with the drain voltage, VDS, for low drain voltages. As the drain voltage increases, however, we see the current not growing as fast with drain voltage. The curve turns over, and at some particular voltage, it becomes rather flat (it is not exactly flat in any real transistor it will continue to rise, at least slowly, with drain voltage.) To understand why the current does not continue to rise linearly with drain voltage, we need to look more closely at the behavior of the charge in the inversion layer, and the whole channel in general. The situation with a finite, but moderate, drain voltage is shown in Fig. 10.26. Now, because we are carrying finite current from source to drain, there is a resistive voltage drop in the channel, and so we now have a voltage in the channel, VC(y), that varies with the position y along the channel. To construct a simple model for this situation, we first make a simplifying approximation (known as the gradual channel approximation) that, as far as calculating the charge density is concerned, the electric field is vertical in this diagram, so the charge density at any point y along the channel is simply the one given by the voltage across the gate-to-channel capacitance at that point, i.e., Qn ( y )

Cox (Vg Vth VC ( y ) )

(10.4)

EE41 Physics of Electrical Engineering

21

Chapter 10 Transistors

There is, of course, now actually some horizontal component of the electric field, because there is now finite resistive voltage drop in the channel, but we assume this electric field component is small compared to the vertical field (which is usually a good assumption).

Vg + + VDS

Qn(y), VC(y) y y =0 y =L

Fig. 10.26. Sketch of an NMOS transistor, with the gate biased with a voltage Vg, above threshold, and with a moderate drain-to-source voltage, showing a charge density, Qn(y), in the inversion layer in the channel under the gate that progressively reduces as we move towards the drain. The actual voltage at any point in the channel, VC(y), also progressively rises because of the resistive voltage drop in the channel from the conducted current. We can now construct a simple mathematical model for this situation. The current flowing in the channel at any point y along its length is given by
iDS = WQn ( y ) v ( y )

(10.5)

where W is the width of the transistor, and v(y) is the electron velocity at this point y. (There is technically a minus sign in Eq. (10.5) because the electrical current in the positive sense is flowing from the drain to the source, i.e., in a negative y direction.) The electron velocity, on a simple drift transport model, is
v = E y ( y )

(10.6)

(there is technically a minus sign in Eq. (10.6) because a positive electric field will drive electrons in a negative direction) where
Ey ( y ) =

dVC dy

(10.7)

is the y-component of the electric field, i.e., the electric field in the channel that is driving the electrons. Substituting using Eqs. (10.4), (10.6) and (10.7) in Eq. (10.5), we have iDS = W Cox (Vg Vth VC ( y ) ) dVC dy (10.8)

EE41 Physics of Electrical Engineering

22

Chapter 10 Transistors

We can now perform a formal integration of this differential equation that will enable us to deduce iDS if we know VDS. We integrate both sides of this equation over y from y = 0 to y = L, where we note that iDS, whatever value it has, in the steady state must be the same for all y, so we can take it outside the integral, i.e.

iDS dy = LiDS = W Cox (Vg Vth VC ( y ) )


0 0

dVC dy dy

(10.9)

Now we can change the variable in the integration on the right. We know that, at y = 0, VC = 0, and at y = L, VC = VDS, so we have
LiDS = W Cox
VDS

(V
0

Vth VC ) dVC

(10.10)

i.e.,

iDS = or equivalently

W 1 2 Cox VDS (Vg Vth ) VDS L 2

(10.11)

iDS =

V W CoxVDS (Vg Vth ) DS L 2

(10.12)

We see from Eq. (10.11) or Eq. (10.12) that the drain current has a linear increase at low VDS, with a quadratic term that will make the curve progressively flatten out with increasing VDS. This equation will explain the basic form of the curve right up to the beginning of the saturation regime in Fig. 10.25. Note that there is a maximum drain-source voltage VDS for which the Eqs. (10.11) or (10.12) have any meaning. Once we get to the point that VDS = Vg Vth at the drain end of the channel, a voltage we refer to as the saturation voltage, VDsat , the charge density in the channel will have dropped to zero on this simple model. This phenomenon is called pinch-off. At this point, incidentally, as the reader can check, the current from Eqs. (10.11) or (10.12) has reached a maximum. For drain-source voltages beyond pinch off, our model, Eq. (10.13), no longer holds, since the term in the integral becomes negative, which is not physical (we cannot have a negative number of electrons in the channel). The situation is somewhat like that sketched in Fig. 10.27, and the transistor is said to be in saturation. Clearly, for any quantitative understanding, we are now pushing our model too far. The basic problem is that at the drain, our gradual channel approximation breaks down. When the transistor is saturated, the horizontal drain field can be as large, or larger than the field from the gate. Qualitatively, what is happening now is there is a depletion region formed next to the drain, of thickness L as shown in Fig. 10.27. The additional voltage above VDsat is taken up approximately by the voltage drop across this depletion region. At some point near to the nominal pinch-off point in the channel, where the charge density is still finite, the electrons are swept out into this depletion region, rather like water falling over the edge of a waterfall. Since the channel voltage at the edge of the waterfall does
EE41 Physics of Electrical Engineering 23

Chapter 10 Transistors

not change much with drain voltage, in saturation the primary effect of the drain voltage is to change the depletion width, and thus modulate the effective channel length. 5 If we are allowed to assume that the depletion region is thin compared to the overall channel length, i.e., L << L, then the length of the region with finite charge density in the channel has not really changed very much, and so the current flowing has only a small dependence on drain voltage after saturation is reached.

Vg + + VDS

Qn(y), VDsat VC(y) y y =0 L y =L

Fig. 10.27. Sketch of the situation in the transistor beyond pinch-off, in the saturation regime.
3

iDsat =

2 W Cox (Vg Vth ) 2 L

Drain current, iDS

Vg Vth = 2.0 Vg Vth = 1.5 Vg Vth = 1.0

Vg Vth = 0.5
0 1 2 3

Drain-source voltage, VDS Fig. 10.28. Plot of the drain current voltage characteristic (with current in units of (W/L)Cox) for various gate voltages, together with the calculated maximum drain current iDsat based on the model in the text.

Actually the drain voltage has another effect in modern transistors, and that is to effect the overall threshold voltage of the transistor. This effect is called drain induced barrier lowering (DIBL) and results from the drain field changing how the gate charge is imaged in the substrate. Higher drain voltages lower the nMOS threshold voltage.

EE41 Physics of Electrical Engineering

24

Chapter 10 Transistors

Fig. 10.28 plots a set of characteristics for an MOS transistor on the basis of this simple model. The dashed line in Fig. 10.28 is the maximum value of drain current on this model, calculated from Eq. (10.12) as the point at which VDS = VDsat (= Vg Vth), i.e.,
iDsat =
2 W Cox (Vg Vth ) L 2

(10.14)

Note that, as the gate voltage is increased above threshold on this model, the maximum current grows quadratically. The model presented here is a simple introductory model of an MOS transistor. There are many detailed aspects found in real transistors that are not captured in this model. This is especially true in modern transistors where the fields are so high that it is not correct to assume that the electron velocity is proportional to field. Actually, at high fields (e.g., >> 104 V/cm), the velocity tends to limit in most semiconductors, for both electrons and holes, to a so-called saturated drift velocity, which is ~ 107 cm/s in silicon. That fact means that transistors do not enter saturation when the channel charge density goes to zero (as we have implicitly been assuming), but rather when the channel charge gets small enough it must move at the saturated drift velocity. This means that modern transistors saturate at drain voltage much less than Vgs Vth. This makes calculating the saturation voltage more difficult, but there have been a number of simplified models proposed that match data reasonably well. It is also true that, increasingly, the quantum-mechanical tunneling of electrons or holes through the gate oxide is becoming significant. Additionally, with the continued reduction in threshold voltage, the finite thermal energy of the electrons is no longer negligible compared to the threshold voltage, and there is significant current flowing between the source and the drain as a result, even for voltages nominally below the threshold voltage. Such phenomena can be modeled and understood using the basic physics we have introduced in this and previous chapters, though the details of this are necessarily beyond what we have space to discuss here.
Summary MOS transistor

The MOS (metal-oxide-semiconductor) transistor consists of a conducting gate, often polysilicon, on top of a gate oxide, usually silicon dioxide, on top of a channel region formed in the substrate between doped source and drain contacts. Modern MOS transistors rely on an inversion channel formed at the oxide to substrate interface for the conduction of current between source and drain, with the gate controlling the amount of charge available for conduction. The very high quality of the silicon-to-silicon dioxide interface is what makes practical MOS transistors possible, and these have scaled over many orders of magnitude to enable the fabrication of billions of transistors at once.
CMOS devices NMOS and PMOS

Complementary MOS consists of NMOS and PMOS transistors respectively, in which electrons and holes respectively are the charge carriers, with NMOS transistors being

EE41 Physics of Electrical Engineering

25

Chapter 10 Transistors

turned on by positive gate to source voltage, and PMOS transistors being turned on by negative gate to source voltage.
CMOS logic

Complementary MOS logic is formed with two complementary groups of devices, one pulling the output up, and the other pulling it down, with the two never on simultaneously in the steady state.
CMOS fabrication and planar lithography

CMOS devices and the wiring between them are made by a succession of planar fabrications, using lithographic processes with spun-on or deposited layers of material, and optical exposure of photosensitive resist layers that are subsequently developed chemically to create the desired patterns.
Threshold voltage and inversion layer

When the gate voltage is biased above a particular, threshold voltage (positive for the normal enhancement-mode NMOS transistors), an inversion layer is formed at the oxidesubstrate interface that contains electrons in the NMOS case, and these electrons enable conduction between the source and the drain. PMOS transistors have holes in their inversion region.
Linear region of operation

For small source-to-drain voltages, the transistor behaves as a resistor whose value is controlled by the gate voltage.
Saturated region of operation

For large source-to-drain voltages, the transistor has an approximately constant current above a pinch-off or saturation voltage. The value of this current is approximately proportional to the square of the gate voltage above the threshold value, at least in a simple model.
References

[1] Gordon Moore, Cramming More Components Onto Integrated Circuits, Electronics, Volume 38, Number 8, April 19, 1965 http://www.intel.com/research/silicon/moorespaper.pdf [2] R. H. Dennard, F. H. Gaensslen, V. L. Rideout, E. Bassous, and A. R. LeBlanc, Design of ion-implanted MOSFET's with very small physical dimensions, IEEE J. Solid-State Circuits, Vol.SC-9, no.5, p.256-68 (Oct. 1974) http://ieeexplore.ieee.org/iel5/4/22538/01050511.pdf [3] Gordon Moore, talk given at International Solid-State Circuits Conference, San Francisco, Calif., USA February 10, 2003 http://www.intel.com/pressroom/archive/speeches/moore20030210.htm [4] 2001 International Technology Roadmap for Semiconductors http://public.itrs.net/Files/2001ITRS/Home.htm

EE41 Physics of Electrical Engineering

26

Das könnte Ihnen auch gefallen