Beruflich Dokumente
Kultur Dokumente
8, AUGUST 1991
1090
IF
A
V+
v-
Fig. 1. When one tub is forward biased with respect to the substrate,
electrons are injected into the Substrate and diffuse to adjacent reversebiased tubs, where they are collected.
I. INTRODUCTION
i
L e p l e tion
p - substrate
0018-92~0/91/08O~-IO90$O1.O001991 IEEE
1091
400
20
,-.
10
L
-a
TJ = 25C
/-
200
+
z
I-
-10
3
U
-20
1
-4
I
-2
VOLTAGE (V)
Fig. 4. Forward characteristics of tub-substrate diode when electronrecombination current is supplied only by holes generated in the
Schottky barrier.
111. DIE-ATTACHINTERFACE
The silicon-metal contact established at the die-attach
interface is not necessarily ohmic. Older IC wafers that
were neither backlapped nor masked against the diffusion
of impurities into the backside had a p-n junction located
within a few micrometers of the back surface where the
n+-emitter diffusion predominated.
Backlapped wafers with p-type substrates in the 5-15R . c m range appear to form a Schottky junction at the
die-attach interface. This has been observed with both
gold-eutectic and soft-solder die attach.
Properties of the Schottky barrier can be inferred from
the characteristics of the diode formed between a p +
contact on the top of a p-type substrate and the metal
case to which the substrate is attached. The results in Fig.
3 are representative of those obtained with 250-pm-thick,
6-Q .cm substrates. Forward dynamic resistance is about
what would be expected from an ohmic (noninjecting)
contact. Reverse dynamic resistance is several times the
forward.
The forward characteristic of a tub-substrate diode,
where substrate current is supplied solely through the
die-attach interface (isolation wall floating), is shown in
Fig. 4. The high dynamic resistance indicates that electron
concentration in the substrate is limited by hole generation at the reverse-biased Schottky barrier which must
I 092
10
-U
0.1
2
z
Q
08
06
04
001
02
0 001
0
-2
-1
DISTANCE RATIO ( S i t )
(a)
10
i
L
08
0.6
a
04
02
I ,=~ltJ,
(1)
-2
-1
(b)
Fig. 7. Normalized substrate voltage rise on the surface as a function
of normalized distance from edge of depletion transferring holes into
substrate. (a) Isolation wall floating. (b) Isolation wall at V - .
common isolation wall in addition to the current predicted by (1). If the tub size is not too great, this translates into an h,, near unity for the lateral n-p-n formed
by the tubs and substrate. This current gain falls off for
current levels greater than 4 A/cm (10 mA/miI) of
contacted mutual isolation.
If a third tub connected to V - separates two tubs
where the isolation walls are hardwired to V - , current
transfer between them is reduced [5]. For example, with
100 p m between the walls of the third tub, the lateral
transfer current between the outside tubs will be roughly
0.04 times the current flowing into the proximate isolation
wall of the injecting tub.
When a p-region within a tub is forward biased into the
tub as shown in Fig. 2, holes will be transported through
and around the subcollector to be collected by the substrate. The resulting substrate current will raise the substrate voltage above V-. Two-dimensional solutions for
substrate voltage near the surface in the vicinity of such a
tub are presented in Fig. 7. Assumptions made are that
the current density .ID is constant in a planar depletion
zone between subcollector and substrate, the current density in the substrate is sufficiently low to avoid velocity
limiting [61, the injection of electrons at the backside
contact is negligible, and the tub is biased at a voltage
greater than the peak substrate voltage. The results in
Fig. 7(a) ignore the effects of isolation walls while those in
Fig. 703) presume that the isolation wall at the periphery
101)3
60
-5
40
4 -
2
5
9k
2
20
-20
-2
-40
-A
?i
3
Vour
-6
-60
10
15
20
25
TIME (ms)
V ,= p t J ,
where p is the substrate resistivity. Depletion-zone current densities exceeding 50 A/cm2 are not unheard of in
practical ICs. Worst case, the peak substrate voltage rise
could be in the tens of volts. The data in Fig. 7 indicate
that tubs clamped within a few volts of V - should be
located at some distance from the source of substrate
current even when the isolation walls are connected
to
Should the substrate voltage rise be so great that the
conducting tub is no longer reverse biased (quasi-saturation of the parasitic p-n-p), the results in Fig. 7 are no
longer valid (the lateral voltage rise is increased).
The discussion above indicates that contacting the isolation walls to V - significantly increases lateral effects
with electron currents in the substrate while reducing
them with hole currents. Further, while a lateral collection tub connected to I/- (guard) can suppress electron
diffusion between tubs, it can also result in a p-n-p-n
structure being activated should there be sufficient hole
currents in the substrate. That which mitigates one problem can aggravate another.
v-.
V. CLAMPING
INDUCTIVE
OVERLOADS
Fig. 8 shows the output waveform of an op amp that
has an inductive overload. Inductor current continues to
flow even though the output transistor cannot handle it,
because of the power limiting. Consequently, the output
must be clamped to the supplies to limit the voltage
across the transistors. A n output stage with clamp diodes
is shown in Fig. 9.
An internal clamp diode for V,,
< V - can be formed
simply by connecting the isolation wall of one output
transistor ( Q 2 in Fig. 9) to V-. This can be augmented by
1094
3.
Tc = 125C
Vs = k40V
toN= 1ms
1 1 1 2
z5
,'
9
4
0
0
0
10
15
(a)
03
TC = 125OC
Vs = +40V
n:
W
tON=
1ms
0.2
3
a
0
01
2
m
2
0
10
15
(b)
VI. DESIGNEXAMPLE
An existing design for a power op amp [4] was modified
to reduce stress with inductive overloads. The IC is rated
to deliver & 35 V at 10 A. Short-circuit current is 15 A.
Power transistor tub size was 2.7 mmX 1.2 mm. The two
power transistor tubs were separated by 300 p m along a
2.7-mm common boundary; the region between tubs was
floating n- material. Detailed drawings and microphotographs of the composite transistor structure are available in [4]. Only tub size, edge lengths, and boundary
separations of the new design are relevant here.
The VoUT < V - clamp had an effective length of 2 mm
(n+ contacted to V,,
separated by 50 p m from p +
contacted to V - ) , This clamp was part of the tub of Q ,
along the edge remote from the common boundary with
Q,. The layout gave relatively uniform conduction in the
power array metalization, so ohmic drops in the metal
were reduced. The forward characteristics of the diode
are plotted in Fig. 10(a).
The isolation wall at the common boundary was not
entirely floating because there were V - contacts to the
isolation wall at a distance comparable to a diffusion
length from the tub of Q 2 .These contacts were necessary
to control substrate voltage rise with VoUT > V+. The
cross-supply current that occurs with VoUT < V - is given
in Fig. l o b ) . The increase in positive supply current is
about 0.01.,,Z
Using the results of Fig. 10, a 15-A inductive overload
with V,,, < V - will dissipate 30 W in the clamp diode
and, with k40-V supplies, 12 W along the edge of Q ,
facing Q,. This severe case is within continuous power
capabilities with a case temperature of 125C [41.
1095
15
TC = 125C
V, = i 4 0 V
L
;
t o N =lms
10
5
>
n.
0
0
10
(a)
d
+
w
Tc = 125C
Vs = +40V
foN= I m s
15
and
high-voltage power switches has received some attention [9].
(4)
As a final detail, positive output-stage shutdown was
provided should the output go beyond either supply. With where J , is the electron current, J, is the hole current
V,,, < V - , positive shutdown of Q, in Fig. 9 does not supplying recombination losses, and b = p, / p , . The
occur until Q 3 has substantial conduction. Internal tubs variation of J , and J, with distance from the tub to the
can become forward biased before this happens. To avoid depletion is sketched in Fig. 12.
problems, auxiliary circuitry was provided to turn off Q ,
The electron concentration gradient is also sketched in
before any tubs can become forward biased as V,,
Fig. 12 for J, 0. This gradient was assumed in obtaining
drops below V-. Similarly, Q2 is turned off when VoUT the solution for 7 in Fig. 6. When a hole current is
rises a diode drop above V+. Without some form of required to supply appreciable recombination losses, the
guaranteed, fast shutdown, the IC can be subject to second curve applies for the same total current. From ( 3 )
obscure or transitory power-limit failures that are difficult and (4), the diffusion gradient is reduced for x > 0 and
to locate or characterize, especially when they cannot be the electric field increases to support simultaneous hole
made to recur because of device destruction. One such and electron conduction. With the higher field, an equal
flaw, observed on a power op-amp design, was that output current is supported with lower carrier concentration in
current could become uncontrolled if a supply lead was the substrate. With n = NAs or n -K NAs, a similar situaopened, even though the device was otherwise protected. tion applies. For a given current density, minority-carrier
concentration in the substrate is reduced in the presence
of recombination losses.
VII. CONCLUSIONS
Under these conditions, 77 will be reduced, both beSubstrate currents are strongly influenced by conditions cause of reduced concentration under the injecting tub
at the die-attach interface. Backlapped wafers with a and because of recombination losses in diffusing to the
p-type substrate resistivity of 5-15 R . c m exhibit a collecting tub. Electron-diffusion length in the substrate is
Schottky barrier at the interface with both gold-eutectic roughly equal to the substrate thickness (250 pm). This
and soft-solder die attachment. This barrier has been significantly reduces the value of r ] . Nonetheless, the
characterized with sufficient accuracy to establish an up- results in Fig. 6 can be used to establish an upper limit for
lateral diffusion.
per limit for deleterious effects.
0
7
G, =~SXIO"V.S/C~~
=670A/cm
Jp =330A/cm2
F
U
< 2
F
o
Z
8
I-
o
1
U
SUBCOLLECTOR
L
I
o
6
10
J,=-.
kTP,:
G,
J p = k T p -.
p'
ws
(7)
The Gummel number is defined as [6]
(9)
kTni
J,=-.
GiJ
The Gummel number of a bipolar transistor emitter can
be determined experimentally. The measured value takes
into account bandgap narrowing and recombination losses,
factors not considered here.
In the diode of Fig. 2, J , would be the electron current
flowing from the p-n-p emitter to the subcollector (base);
and J,, would be the hole current flowing from the p-n-p
emitter through the subcollector (base) into the substrate
(collector), assuming negligible recombination in the subcollector.
The Gummel number for the p-n-p emitter has been
determined from transistor measurements, so no is obtained from (lo), Po is then obtained from (31, and G,, is
obtained from (9).
The result was G, = 6 X 10"V.s/cm4. This is near the
limit of what has been obtained for the emitter of an
n-p-n transistor 10I2V.s/cm4, indicating that a major
reduction in substrate current is not to be expected from
increased subcollector doping.
Fig. 13 gives a plot of plasma concentration in the n collector of the diode in Fig. 2, with and without injection-efficiency spoiling. The current density is equal to
that in Fig. ll(b) for a 15-A clamp current. There is
simultaneous conduction by both holes and electrons in a
drift field where diffusion forces are relatively small.
If a lateral p-n-p collector were put along the side of
the p-n-p emitter, conduction to this collector would be
Isus
0.04. I:;
(11)
1097
Robert J. Widlar was born in Ohio on November 30, 1937 and died February 27, 1991. H e
graduated from the University of Colorado,
Boulder, in 1962 while working for Ball Brothers Research.
In 1963 he joined Fairchild Semiconductor
where he headed linear IC development. In
1966 he formed the linear I C group at National
Semiconductor, Santa Clara, CA, and was responsible for new product design until 1970.
Since 1974 he had been working as an Independent Contractor, and had developed products for both National Semiconductor and Linear Technology. Since 1963 he had specialized in the
development, specification, and application of linear ICs. He designed
the first industry-standard IC o p amps, voltage comparators, and power
voltage regulators, along with generations of improvements. Over two
dozen of his products are still in volume production, some for over 20
years. He had pioneered such innovations as the bandgap voltage
reference and the super-gain transistor as well as numerous design
techniques that are widely used today. His most recent work includes
micropower low-voltage ICs, bandgap curvature correction, advanced
super-gain o p amps, and improved class-B amplifiers and high-power
techniques. Bob Widlar will always be remembered as a great innovator,
linear circuit designer, and tenacious problem solver. He will be missed
by many in the industry.