laC~mflmtBn ELSEVIER Microelectronic Engineering 40 (1998) 131-146

Application of Charge Pumping Technique for sub-micron MOSFET Characterization

C.R.Viswanathan and V. Ramgopal Rao Electrical Engineering Department, University of California, Los Angeles, CA 90095-1594, USA In this paper, charge pumping technique for MOSFET interface characterization will be reviewed. The basic principles of charge pumping technique will be elaborated and its evolution as an excellent tool for a thorough characterization of MOSFET interface properties will be illustrated. Published results regarding the applicability of charge pumping technique for a study of sub-micron MOSFET interface and its degradation under various electrical stress conditions and radiation will be analyzed. The effect of geometric components on charge pumping current as well as the recent reports of single interface trap characterization in sub-micron MOSFETs will be described. The application of charge pumping technique at cryogenic temperatures and in other MOS based devices will also be included.

1. INTRODUCTION
During the last decade, charge pumping (CP) technique, originally discovered by Burgler and Jespers in 1969, has evolved as the most powerful technique for interface characterization in MOS transistors [1]-[21. Even for the present day ultra small MOSFETs, the charge pumping technique proved to be extremely useful, since no other technique allows the probing of MOS interface properties with the same kind of simplicity and accuracy. This can be gauged from the recent demonstrations [3]-[5] where charge pumping has been used to resolve individual interface traps in sub-micron MOSFETs and from the reports that single interface trap generation [3] in deep submicron MOSFETs under hot-carrier stress can be monitored with charge pumping. The major advantage of the charge pumping (CP) technique however is that the measurements are done on the actual transistor itself without needing a separate test device. The need to extrapolate the results obtained on a test capacitor to characterize the interface in a transistor is thus eliminated. The technique yields precise results. Several authors have developed different ways of applying the charge pumping principle for the determination of the average density of interface traps [6]-[9], their energy and spatial distributions [9]-[15], the geometric mean value of the capture

cross section for electrons and holes [9], and recently even the reverse short channel effect in sub-micron MOSFETs [16]. This technique is adequate for the evaluation MOSFET degradation due to electrical stress and radiation damage [17][251. Much work has been carried out over the last 2 decades in order to improve the understanding of the technique [1]-[2]. The basic principle of the technique is discussed in Section 2. The presence of geometric components in the charge pumping current was first reported by Brugler and Jespers [6]. The geometric component originates from free minority carriers that do not have enough time to escape via source and drain during the switch-off of the gate pulse. It is very important either to eliminate the geometric component, or to separate it from the actual interface trap component of the charge pumping current. Section 3 deals with this issue of separating the geometric component from the actual charge pumping current. The modifications to the basic charge pumping technique to extract energy distribution of interface traps, as well as the lateral and vertical spatial profiles of the interface traps and/or of the trapped oxide charge is discussed in Section 4. A demonstration of the extremely high sensitivity of the technique was reported recently [3]-[5] by the observation of single interface traps in sub-micron MOSFET's, which is the subject of Section 5, in

0167-9317/98/$- see front matter Copyright 1998 Elsevier Science B.V. All rights reserved. PII: S0167-9317(98)00266-4

C.R. Viswanathan,V. Ramgopal Rao / Microelectronic Engineering 40 (1998) 131-146

where Dit(E) is the energy distribution of interface states, [Iit is the average interface state density between ~ o s and ~oFF and q is the electron charge. Due to repeated charge pumping during each pulse cycle, unidirectional pulses of charge flow through the substrate and a de substrate current flows equal to
~N

133

Typically it is assumed that in strong inversion, the band bending is pinned to a value equal to twice the Fermi potential OF, and that in accumulation, the band bending is nearly zero. This makes A~, in equation (3) equal to 2 t l ) F By measuring lcp, one can, in principle, obtain information on the interface state density. 3. G E O M E T R I C C O M P O N E N T S In the description above, several simplifying assumptions have been made. One is that, when the pulse is turned off, all the mobile (inversion) carriers flow back to the source and drain regions. This does not happen always. When the length of the channel is large, the carriers in the middle of the channel, which are far away from the source and drain junctions, will not have time to flow back to the source and drain by the time the pulse is turned off. These carriers that are IeR behind will recombine with majority carriers (holes) during the OFF period and therefore an additional component of substrate current that does not involve the interface states arises. This dc substrate current resulting from the recombination of inversion carriers (mobile carriers) with majority carders in the substrate is an unwanted current which is included in the measured CP current. Since the number of carriers left behind in the channel depends on the length and the width of the channel, this de substrate current is called the geometric component. The presence of geometric components in the charge pumping current was first mentioned by Bmgler and Jespers [6] as early as in 1969 and later studied by various authors [8][9], [33]-[35]. If ct is the fraction of the inversion carriers that do not flow back to the source and the drain and then recombine in the substrate, the charge pumping current including the interface and geometric components can be expressed as

Isva = QcPf = qAf ~ , , D~,(E)dE = qAJ~,Ag's

(3)
where f is the repetition frequency of the gate pulse. This current is usually denoted lee rather

than Isus.
EC

t
Dk (E) I ,
, . .

&Ws

IEi

Ev

Figure A

',

Ev

E~ =

E~

Ec

Figure B
Figure 2. Illustration of the band of interface states participating in CP action; A) The location of the QuasiFermi level during the ON and OFF period of the pulse. B) An arbitrary distribution of the interface states in the bandgap. The shaded region represents the interface states contributing to the CP current.

I c p : fA[q~tAgrs+aCox(V~H-Vr)]

(4)

where Cox is the gate capacitance per unit area, VT the threshold voltage and V~H the gate voltage at the top or high value of the pulse. Several techniques are employed to reduce the geometric component. One is to apply a larger reverse voltage between the substrate and the source/drain

134

C.R. Viswanathan, V. Ramgopal Rao I Microelectronic Engineering 40 (1998) 131-146

regions. This technique has the disadvantage of altering the effective channel length which has to be determined by auxiliary measurements. It also alters the threshold voltage in the channel. Some authors [9],[33] have claimed that increasing the W/L ratio reduces the geometric component. Some others [6],[8] have claimed that only reducing the length of the channel decreases the geometric component. Using transistors with a larger W/L ratio, while keeping L constant, does not alter the fraction of the measured charge pumping current that can be attributed to the geometric component. Increasing W/L ratio by decreasing L reduces the time needed for the inversion carriers in the middle of the channel to reach the source/drain region. The technique used in the earlier days to minimize the geometric component was to increase the fall time of the gate pulse. Under this condition, the inversion carriers are able to empty completely into the source/drain region by the time the gate is turned off. A typical pulse shape used in charge pumping experiment is shown in Figure 3. V~L is the voltage at the bottom of the pulse and VGH is that corresponding to the top of the pulse. In order for the charge pumping experiment to work as stated earlier VOL has to be less than the flatband voltage, VFB, in order to correspond to accumulation condition and VOH has to be larger than the threshold voltage VT in order to correspond to inversion condition.
VC-d. - . - I
J

Recently, Van den bosch [34] et al. have shown that the occurrence of a geometric component can also be studied by using an additional charge collecting junction, such as in the case of a p-well inside an n-type substrate. In this method, the free inversion electrons that do not reach source and drain are injected into the well and will be collected as majority carriers in the n-type substrate, where they are measured as a substrate current. The holes that recombine with the electrons in the interface traps are measured at the well contact. In this way, the interface trap and the geometric component can be separated. Fig. 4 shows the geometric component, measured in this way, as a function of the gate length and the fall time of the pulses [34]. The model that describes the dependence of the geometric component on the pulse shape has been improved by I-Iabas [35]. For devices with a channel length of 10 ttm, fall times larger than 100 ns should be used in order to avoid the geometric component [2].
1 0 ~z '~ 10'
-

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . .

"~OX.

-o- Lo=lO~m ,=0,= Lg.,3Opm -O-Lg.lOOpm

1 0e 0 "s

........

'

........

10-7

10 "s

........ ) 10 "s

..... 0"4

VT - - - /

Transition time [s]

.____Jt
VQL
RiseTime

I- il- To. -I-T;IONTune F=ITn OFFTime

Figure 4. Pumped charge per cycle at the substrate as a function of gate pulse transition time, measured on transistors of various gate lengths. Measurement conditions: pulse amplitude 6 V, reverse bias VR--0.2 V [34]. 4. E X P E R I M E N T A L T E C H N I Q U E S The charge pumping measurement is usually c.~'ried out using rectangular pulses in either a variable amplitude mode or a variable base mode. The rectangular pulses are also called trapezoidal pulses due to the finite rise and fall times. Various modifications to the basic charge

Figure 3. The shape of the gate pulse used in CP measurements. The finite rise and fall times, TR and TF respectively, of the pulse makes it a trapezoidal pulse. Tow and TOFFare the ON and OFF periods of the pulse respectively.

C.R. Viswanathan, V. Ramgopal Rao / Microelectronic Engineering 40 (1998) 131-146

135

pumping principle have been proposed over the years. In this section we will review the different charge pumping schemes in use for sub-micron MOSFET interface characterization.

typically over

an

energy

range

of

width

qA ~ ~q(2~) in which case the CP process can

be called voltage limited CP. The fact that the current nearly saturates in this region validates our assumption that the Fermi potential is pinned in both inversion and accumulation. The flat portion has a very small slope. In the voltage limited CP case, the surface potential increases very slowly with VG. in inversion and hence the interface states that participate in charge pumping increase by a small amount for an incremental change in VoH. It must be noted that the slight increase in current with Vail in this region can be used for determining the interface state density in the region of the band gap close to the conduction band edge from a knowledge of Us variation with Vo in strong inversion. Ignoring the small increase in the current with Vo in this region, this current is taken as the maximum charge pumping current IcPmax. The interface states are assumed to be in equilibrium with the free carriers at the oxide interface in the bottom of the pulse as well as the top of the pulse through the rapid capture of appropriate free carriers in the voltage limited CP measurement. However, situations arise when the interface states are not in equilibrium with the carders at the interface. For example, when the rise and fall times of the gate pulse are not extremely short, carriers will be emitted by the interface states during rise and fall times and these emitted carriers will flow back to the Source/Drain regions. The measured CP current will correspond to the remaining interface states that did not emit charges during rise and fall times. It is reasonable to model this process by considering an energy level, E~,,, the interface states above which have emitted electrons during the switch from inversion to accumulation (fall time) and an energy level Era,h, the interface states below which have emitted holes during the switch from accumulation to inversion (rise time). Therefore, the CP current in this case is less than the voltage limited case and is due to interface states between Era., and E,m~. This case is called emission limited CP. In the emission limited CP case, the slight increase in Ice with VGH can also be due to the fact that the time for emission becomes smaller as VGH is increased keeping the rise and fall times constant.

4.1 Variable Amplitude Mode

In the variable amplitude mode, the base level of the pulse, VoL, is fixed while the charge pumping current is measured as a function of the amplitude (Le. VGz-VGL)by varying VGn.A typical charge pumping current characteristic is shown in Figure 5. Here VGL is kept below the flatband voltage V~ so that the device is in accumulation during the OFF period of the pulse. The current nearly saturates when Von exceeds Vr.

lap

VGH

VFB

Figure 5. The CP current characteristics in the variable amplitude mode. VOH is varied keeping VOL fixed corresponding to accumulation. Vr is the threshold voltage when the entire channel is inverted.

Three distinct regions of the characteristics can be observed by reference to Figure 5. In region 1, VQH is less than Vm. Hence the surface is in accumulation during both ON and OFF periods. No charge pumping current is observed in this region. In region 3, Von is larger than Vr. Hence as discussed in the earlier section, .the charge pumping current is due to the charging and discharging of interface states between inversion and accumulation. These states are distributed

136

C.R. Viswanathan, V. Ramgopal Rao / Microelectronic Engineering 40 (1998) 131-146

In region 2 V ~ is less than VT. The current is less than Icpm= for several reasons. One is that although VGH is less than Vr, there is inversion close to the source and to the drain as shown in Figure 6 [46].

I~

,/

INVERSION CHARGE
L

\,

~1

DEPLETION
Figure 6. The formation of the inversion region near the source and the drain while the middle of the channel is still depleted at a gate voltage slightly less than VT.

Charge pumping current arises due to the interface states in this AL region. By studying the characteristics during the rising portion of the current characteristics, the non-uniformity of the threshold voltage can be investigated. If the channel is uniform as in a long channel device, where the threshold voltage Vr is the same over most of the channel, there is still a fmite slope to the rising portion of the charge pumping characteristics. The surface is in weak inversion in the region of the rising edge of the Icp characteristics. The current in this region is determined by capture of electrons in weak inversion during the ON period. The measured current is less than what is obtained in strong inversion for two reasons. One is that the portion of the band gap over which the Fermi energy changes between the ON and OFF periods is smaller and hence the number of interface states contributing to the charge pumping action is less. Second is that due to the competing hole emission process, some of the interface states, especially those with a short hole emission time constant, become charged by hole emission process rather than by electron capture process. Hence these interface states do not contribute to the charge pumping current.

4.2 Variable Base Mode In the variable base mode, initially proposed by Elliot [7], the gate pulse amplitude (V~H - V~L) is kept constant and the charge pumping current is measured as a function of the base level, VoL, of the pulse. A typical charge pumping current characteristic obtained in this manner is shown in Figure 7. As shown by Heremans et aL [33], five distinct regions of operation can be identified. In regions 1 and 5 where the device is in accumulation during both ON and OFF periods or in inversion during both ON and OFF periods, no charge pumping current is measured. Region 3. corresponds to the maximum CP current regime and extends over a range Vr - AV < VOL < V~ where AV is the amplitude of the pulse equal to VoH - VGL" Implicit in this condition is the assumption that the pulse amplitude zlV is larger than Vr -VFB" At VGL= Vr -AV, the surface starts to get inverted at the top of the pulse while the bottom of the pulse corresponds to accumulation. At VoL=VF~, the surface starts to get out of accumulation. The charge pumping current measured in region 3, is the same as in region 3 in the variable amplitude technique (Figure 5) and equal to Icpmo~ In region 2, the current rise with VGL is similar to region 2 of the variable amplitude technique (Figure 5).

. . . .

',

,'

sv~

ITI
4 3

VT
V FIB

Figure 7. The CP current characteristics in the variable base mode. The pulse amplitude VoH-VGLis kept fixed while VoL is varied.

C.R. Viswanathan, V. Ramgopal Rao / Microelectronic Engineering 40 (1998) 131-146

137

Although VGL is varying, the base of the pulse corresponds to accumulation in region 2 and the finite rising edge of the current characteristics is also due to the same reasons as discussed earlier in region 2 in the variable amplitude mode. In region 4, the base of the pulse corresponds to a condition when the surface is slightly depleted. Nevertheless, we call this a weakly accumulated regime to show that there is a significant amount of majority carriers at the surface. During the ON period, the interface states get negatively charged by capturing electrons from the inversion layer at the surface. The finite slope of the falling edge can be due to non-uniformity in the flatband voltage VFB i.e., a non-uniform channel. Even in a uniform channel, the finite slope can arise due the competing processes of electron emission and hole capture cited above.

will allow sufficient time for the mobile inversion carriers to flow back to the source and drain

Icp

VFB 1 2

i I
3! VGH

4.3 The Inverted Variable Amplitude Mode

Still another variation of the technique is to use the variable amplitude technique with the top of the pulse V~H fixed to correspond to strong inversion. A typical charge pumping current characteristic measured using this technique is shown in Figure 8. Again, three regions can be identified. In region 3, the surface is inverted in both ON and OFF period since VOL is larger than VT. Hence no charge pumping current is measured. In region 2, V~L is at a level corresponding to weak accumulation. The charge pumping current increases with a decrease in VaL since more and more holes are available during the OFF period with decreasing V~L. This region is similar to region 4 in variable base technique (Figure 7). In region 1, VOL is sufficiently low (VQL < VFB) to keep the surface in accumulation during the OFF period. Hence, maximum charge pumping current is measured in this region similar to region 3 in both Figures 5 and 7. By analyzing the current characteristics in region 2, the non-uniformity of the flatband voltage in the channel can be studied.

J!L,,

VT VGL

VFB

Figure 8. The CP current characteristics in the inverted variable amplitude mode.

4.4 Saw-tooth Wave

Another variation of charge pumping technique is to use a saw-tooth waveform rather than a rectangular pulse. The saw-tooth waveform is a limiting ease of a trapezoidal pulse train (Figure 3) with ToN and Tow becoming zero. Initially it was thought that the slow fall of the voltage during the decaying part of the saw-tooth

thereby eliminating or reducing the geometric component of the charge pumping current. This is not confirmed by experimental results. Increasing the rise time increases E~h while increasing the fall time decreases E~e. The charge pumping current therefore decreases continuously with an increase in the rise and fall times of the pulse. The application of saw-tooth waveform is useful in obtaining more information on the interface emission parameters. A saw-tooth waveform with a short fall time and various rise times can be used to study hole emission parameters. Or alternately, use of a short rise time with various fall times of the saw tooth wave can be used to study electron emission characteristics of the interface states. The analysis is more elaborate but can be made with simplifying assumptions such as an uniform channel. Groeseneken et al. [9] determined the geometric mean of capture cross sections of electrons and holes using saw-tooth waveforms of different rise and fall times.

138

C.R. Viswanathan, V. Ramgopal Rao / Microelectronic Engineering 40 (1998) 131-146

4.5

Three-Level Charge Pumping

Chung et al. [36] used a three level method developed by Tseng [37] to measure interface state density. In Tseng's method, the gate pulse has in addition to the base and top levels, VGLand VGH, a third intermediate level V~ as shown in Figure 9.

......................

V~

.....................

"[ ......

v,

I-

T,

T1

Figure 9. Three level gate pulse: In addition to VaL and VoH the pulse has an Intermediate level V.. During T~, all the interface states have captured holes and are in quasi-equilibrium with the holes in the accumulation region since VGL is less than VFB. During T2, all the interface states have captured electrons and are in quasi-equilibrium. When the gate pulse is switched from V~H to the intermediate level V, the states emit electrons. If T~ is long enough, all states lying above an energy level ETa, determined by Ve, will have emitted electrons and states below this level will still be negatively charged. Afterwards, during T~, the interface states lying between ET, and the Fermi energy level in accumulation will capture holes and thereby contribute to the charge pumping current. By measuring the charge pumping current as a function of V~ and therefore as a function of ET~ the interface state density distribution is obtained. Chung et al. [36] applied this technique to determine the emission time constant characteristics of the interface states by shortening T~. They applied this technique to determine the increase in interface state density due to F-N stress and hot electron stress. Saks et al [38] extended the three level technique to measure the electron and hole capture cross sections of the interface states. The energy level of the interface states up to which emission takes place is a function ofVe ifT is long enough. It will be a function of T~ if the

emission process is not completed at the end of Te. In the plot of the charge pumping current as a function of Te, the value of Te, at which the current levels off is taken as the emission time constant. From a knowledge of the emission time constant and the energy of the interface states, which is obtained from V, the capture cross section is obtained. Cilingiroglu [39] modified the CP technique by combining it with the pulsed interface probing (PIP) technique [40]. Two different pulse waveforms are used in this technique. One is the usual rectangular (or trapezoidal including the rise and fall times) which pulses the device between accumulation (during the OFF period) and depletion (during the ON period). The other waveform has a narrow pulse of duration TI superimposed in the middle of the depletion phase as shown in Figure 10. This narrow pulse inverts the surface to charge the interface states negatively. During the ON period, the time interval between the start of the ON period and the start of the narrow pulse, denoted by TL, determines the lower level of the energy range of the contributing interface states. Similarly, the time interval between the end of the narrow pulse and the end of the ON period, denoted Tu determines the upper level of the energy range of the contributing interface states under the assumption that TL and Tu are not long enough for the interface states to reach equilibrium. The source current is measured using successively each of the two waveforms. The difference in current between the two waveforms is taken as a direct measure of the CP current due to the interface states and the subtraction procedure eliminates the leakage current in the junction. The interface state density in the lower half of the bandgap is obtained by measuring the source CP current IcPs as a function of TL and similarly by varying To the interface state density in the upper half of the bandgap is obtained. In another modification of the CP technique, Kejhar [41] replaces the reverse bias source/drain supply voltage by a second pulse generator which is synchronized to the gate pulse. This technique is called the Double-Pulse charge pumping. The major advantage of this technique over the three level charge pumping is claimed to be enhanced immunity against the parasitic geometric component.

C.R. Viswanathan, V. Ramgopal Rao / Microelectronic Engineering 40 (1998) 131-146

139

4.6 Constant Field Profiling Technique

Ancona et al. [13] measured the spatial variation of the interface states using a technique called "constant field" profiling technique.

...............

1..[

............

Vr

TON
Figure 10. The pulse on pulse waveform used by Cilingirogulu [391. The narrow pulse inverts the surface while during the rest of the ON period the channel is depleted. By varying the source/drain reverse voltage, the effective channel length is varied. The variation in the current Icp gives information on the interface state density profile. However, when the source/drain voltage is varied the threshold voltage is also varied. To obviate this difficulty the source/drain bias voltage is applied in the form of a pulse which is ON only during the accumulation phase of the gate voltage pulse and not during the inversion phase. Thus the problem of threshold voltage variation with the source/drain reverse voltage is eliminated. Chen et al. [42]-[43] developed a new technique to measure the interface characteristics in non-uniform channels as in stressed devices. During the rising edge of the charge pumping characteristics, a small increment in VGH causes an elementary increase in the current. This current is due to a small increase in the inverted portion of the channel. The gate voltage VOH at which this increment occurs is the local threshold voltage in this elementary portion of the channel. The derivative of charge pumping current will be nonzero only if VQH is equal to the threshold voltage and is given by

where Icy is the charge pumping current. W and L are the width and length of the channel, f the pulse frequency 8 the Dirac delta function. Nit(Y) the distribution of interface states along the channel integrated between Em.h and E~e. Vr(y) is the local threshold voltage at y. The contribution to dIop(VGH)/dVGH comes from those regions where VG, is equal to the local threshold voltage Va-. If the threshold voltage is uniform (i.e. the same everywhere), the derivative of Icp will be sharply peaked. If the channel is non-uniform the plot of the derivative of Icp will exhibit a finite width representing the distribution of local threshold voltages Va- (y) in the channel. The lateral non-uniformity in the threshold voltage is caused primarily by localized charges near the interface. If the reverse-bias between the substrate and the source/drain junctions is varied, the channel length is varied. For an increase AVR in the reverse bias, the channel length L is decreased by AL with a corresponding decrease in Icp. The change in the maximum value of Icp is given by

C.m= = qWf f,, ( L

(7)

dlce(Va") = qWff~LNi,(Y)o~VG,- Vr(y)]dY (6)

dVGH

whore L' = L - alL. Using a simulation program, one can determine the variation in channel length with the reverse voltage VR. In the experiment a variable reverse voltage is applied on the source/drain junction and a series of Icp vs. VGH and dlq~lVQa vs. Van curves are obtained for different reverse Junction voltages. From the incremental change of Icpm=, Nit (Y) is obtained and the voltage where the incremental change occurs gives the threshold voltage VT(y) from which the trapped charge Nit(Y) can be obtained. The constant field approach of Ancona et al. [13] was adapted by Chen et al. [42] and the drain voltage was pulsed. Li et al. [44] used a spatial profiling technique in which the source and the drain regions are independently reverse-biased. By varying the values of the source reverse bias keeping the drain reverse bias constant, the spatial variation in the interface state density near the source is obtained. Doing the reverse, the profile near the drain is obtained. Another new CP method has been developed by Tsuchiaki el al [21] and applied to study

140

C.R. Viswanathan, V. Ramgopal Rao / Microelectronic Engineering 40 (1998) 131-146

stressed devices. This technique is particularly suitable for the investigation of non-uniform interfaces and determines the distribution of fixed charge near the drain junction. In nMOSFETs, the interface state generation is a major degradation mechanism and the charge pumping technique is suitable for measuring interface state density. In pMOSFET devices electron trapping is the dominant degradation mechanism. This technique is also well suited for investigating the lateral distribution of fixed charge in stressed pMOSFET devices. Assuming a variation in the local threshold voltage VT(y) and VFBOr) in the channel, the time for emission of electrons during the switching of the pulse from accumulation to inversion denoted k~e(y) and the time for emission of boles during the transition from inversion to accumulation denoted t,~h(y) determine the contribution of interface states at y to the CP current. As VGH is incremented by a small amount, the corresponding increase Alcp gives the contribution from an elementary section located between y and y+dy. to the charge pumping current. The elementary section dy has a local threshold voltage equal to Vow. Instead of keeping the rise time and fall time constant when Vm~ and/or VoL are varied, the gradient of the rise and fall time is kept constant. This assures that the contribution of each elementary section is independent of Vail and V~L. The source and the drain CP current lcPs and IcPo are measured and the range over which Ices or IC,pD increases from zero to a saturation value yields the range of local threshold voltage variation in the channel. Combining this with a two dimensional program the distribution of fixed charges in the oxide can be obtained.

5. APPLICATION OF CHARGE P U M P I N G TO DEEP SUB-MICRON MOSFETs

The charge pumping current is proportional to the pulse repetition frequency, and the number of interface states involved in the ehnrge pumping process. The sensitivity of the de current measuring instrument in the experimental set up determines the sensitivity of charge pumping technique in measuring the interface state density. Modem current meters can measure down to a few

femtoamperes. However, measurement of such low currents requires a longer integration time and hence cannot be used at higher pulse repetition frequencies. It is possible to measure current levels of the order of a fraction of a picoampere with pulse repetition frequency of the order of 100 kHz. The charge pumped per pulse is Qcp = Icp/f. It is equal to 10"tz C (62 electrons) for a current of I pA at 100 kHz frequency. Assuming W=20 ~tm and L=5 ~m, the number of interface states measured is equal to 6.2 x 107/cm2. If we assume AWs is 0.6 V between accumulation and inversion, an average surface state density Dit o f 10s/ cmZ.eV can be measured. For typical values of the device dimensions W = 5 Ima and L = 1 ~tm, the number of interface states that can be measured at this frequency and current level is equal to ~ 1.2 x 109/cm2. In scaled devices AWs is closer to 0.8 between accumulation and inversion. Therefore an average surface state density of 1.55 x 109/cmZ.eV can be measured. We calculated these values assuming a value of I pA for the charge pumping current. But as stated earlier, it is possible to measure charge pumping current in the order of 0.1 pA. Hence a sensitivity of the order of I0S/cm2.eV can be realized even when using small geometry transistors. In the present ULSl era of IC technology, the geometry of MOSFET's is entering the deep submicrometer regime. Hence the number of traps present at the Si-Si02 interface of the MOSFET becomes very small. For example, for a MOSFET with a W/L of 0.5 x 0.5 lxmz and with an average interface trap density of Ix 109 cm z a total number of about 2 to 3 traps is expected for the whole transistor area. Therefore, we can no longer consider the interface traps as being continuously distributed in space and energy, as was done until now. In contrast, each individual trap is localized at a certain position in the transistor channel area and at a specific energy level in the bandgap. Therefore, in the case of individual interface traps charge pumping current can be represented as,

lcpm. = qfN

(8)

where N is the total number of traps that is contributing to the charge pumping current [2]. For a frequency of 3 MHz, the charge pumping current for one trap (N=l) is q.f=-0.48 pA, which can

C.R. Viswanathan, V. Ramgopal Rao / Microelectronic Engineering 40 (1998) 131-146

141

readily be measured with e.g. a Keithley 617 electrometer. The effect of spatially localized individual traps on charge pumping becomes visible as a stepwise instead of a gradual change of the CP base level characteristics, as illustrated in Figure 11. The maximum current is about 8 pA, which means that, according to (8), a total of about 16 traps is probed for the whole transistor area. The steps in the base level characteristics display the contribution of individual traps at various locations in the channel. Each location has its own threshold and flatband voltage where the traps start contributing to the CP current which leads to the stepwise behavior. The rising and falling edge of the base level curve for each individual trap can be described based on the SRH capture and emission theory [3]-[5]. For individual traps not only the spatial location but also the energy levels are localized. Therefore the charge pumping current is also expected to change in a stepwise manner with fall and rise times or with temperature, which can also be described by the SRH emission time constants. Saks et al. [3] have demonstrated recently the first experimental observation of a FET with only one active interface trap. In a FET with only two traps, they could also demonstrate the electron emission of one of these traps by changing the fall time of the gate pulse. The ability to measure single traps in a MOSFET illustrates again the extremely high sensitivity of the CP-technique.
8x10 "la
6
A

6. C H A R G E DEVICES

PUMPING

IN DEGRADED

,9,0
2

-3.0

-2.0

-1.0

vt~ (v)
Figure 11. Base level charge pumping curves showing the presence of individual interface traps: S=drain disconnected, D= source disconnected, S+D = source and drain connected. Frequency = 3 MHz [4].

The charge pumping technique has been employed to study the damage to the interface in devices exposed to ionizing radiation or subjected to electrical stresses such as Fowler-Nordheim ( F - N ) or hot carrier (HC) stress. Numerous authors have adapted the CP technique either directly or in a modified manner to investigate the effect of the stress. The interface damage is more or less uniform over the entire channel length under exposure to ionizing radiation or under F-N stress. The damage is not uniform in the case of HC stress. Hot Carrier stress is produced in one of two ways. In one, called the Drain Avalanche Hot Carrier (DAHC) stress, electron-hole pairs are created due to the high electric field present in the drain depletion region and these hot carriers are injected into the oxide aided by the polarity of the electric field which is in such a direction as to drive the hot carriers towards the surface. At low gate voltages and for high drain voltages, for riMeS devices, the polarity of the electric field is such as to cause hot holes to be injected into the oxide. At higher gate voltages approximately equal to half the drain voltage, the polarity is such that hot electrons are injected into the oxide. In either case, the hot carriers are injected into the oxide in a region close to the drain. This causes localized damage to the interface in the vicinity of the drain. It is believed that hot holes are injected closer to the drain than the hot electron injection. The non-uniformity in damage to the interface is again different between hot hole and hot electron injection. The second mechanism of hot carrier injection is called the Channel Hot Carrier (CHC) stress or Channel Hot Electron (CHE) stress in the case of nMOS transistors. In this type of injection, channel inversion carriers are accelerated by the drain voltage and thus become hot. Some of these hot carriers get injected into the oxide. The region over which they are injected is larger in this case and thus the interface damage occurs over a larger region than in DAHC. Still, the damage is not uniform over the entire length of the channel. The charge pumping technique is better suited to the study of non-uniform interface damage in MOS transistors than conventional techniques. As pointed out by Heremans et al. [33], the channel

142

C.R. Viswanathan, V. Ramgopal Rao / Microelectronic Engineering 40 (1998) 131-146

can be treated as a composition of several regions each with a different value of threshold voltage. For example, the damaged region and the undamaged regions have different threshold voltages. The total charge pumping current is the sum of contributions from each of these regions. It is possible therefore to analyze the charge pumping current more directly or using some of the CP techniques discussed in section 4 to investigate the interface stage damage. Heremans et al. [ 17] determined the increase in interface state density after hot electron injection and after hot hole injection by measuring the difference in the maximum charge pumping current, between virgin and stressed samples. The difference is only due to the contributions arising from the damaged region near the drain. Heremans et al. [33] showed that the presence of fixed oxide charge due to the charge trapping as well as generation of interface states can be investigated using charge pumping techniques under all kinds of aging conditions such as exposure to radiation, F-N stress and HC stress. By dividing the channel into two regions, one damaged (near the drain) and the other undamaged (rest of the channel) and each with a different threshold voltage. They carried out a theoretical analysis of the charge pumping current and compared it with experimental results. The difference between the current measured during the increasing (rising) portion of the charge pumping characteristics with the drain connected and that measured under the same conditions with the drain floating is attributed to the contribution of interface states in the damaged region near the drain. In the range of voltages used in the measurements, the gate voltage is able to invert the region near the drain but not the rest of the channel. When the drain is connected, electrons are able to flow only from the drain (the source is not able to supply electrons because the rest of the eharmel is not inverted) and charge up the interface states giving rise to charge pumping current. When the drain is not connected, there is no way for electrons to come to this region and hence no charge pumping current is measured. Vuillame et al. [45] used simulations and charge pumping technique to study the creation of acceptor-like traps in the gate-drain overlap region of an nMOSFET subjected to hot electron and/or hot hole injection.

Using the method developed by Groenseneken et al. [9], Chen et a t [18] determined the interface state parameters in stressed samples. From the plot of the charge pumped per pulse as a function of frequency, the geometric mean of the hole and electron capture cross sections of the interface states are obtained. They recorded the charge pumping current as the pulse frequency was swept over a range of two decades. This method assumes that the emission process is dominant and that the interface states are not in equilibrium during the transition from accumulation to inversion or from inversion to accumulation. Another implicit assumption is that the capture cross section is independent of energy and can be described by a single value for the interface states distributed over a range of energy. Paulson et al. [22] have extended the charge pumping theory from a study of traps at the SiSiO2 interface to a study of traps in the oxide. Their analytical model, based on a tunneling theory, allows the spatial distribution of nearinterface oxide traps to be determined from variable frequency charge pumping data. They have been able to successfully apply this technique for a study of near interracial traps in irradiated MOSFETs. Recently, by combining the charge pumping technique with gate-to-drain capacitance measurement Ang et al. [19] have shown that the effect of oxide and interface state charges on charge pumping can be clearly distinguished and the spatial profiles of the two charges can also be separately determined in stressed sub-micron MOSFETs. Based on all of these extensive studies it is now possible to gain an in depth understanding of the type of hot-carrier and radiation damage in current state of the art field effect transistors using the charge pumping technique. 7. CHARGE PUMPING TEMPERATURES AT LOW

Till recently, charge pumping experiments have been confined mostly to temperatures above 77 K. The difficulty in using charge pumping technique at lower temperatures arises because of substrate freeze-out effects. At very low temperatures, the substmte behaves like a leaky insulator and the flow of majority carriers to the

C.R. Viswanathan, V. Ramgopal Rao / Microelectronic Engineering 40 (1998) 131-146

143

surface during accumulation is limited by substrate time constant effects. Recently Nguyen-Duc et al. [47] demonstrated for the first time that charge pumping measurements are feasible down to liquid helium temperature. They avoided the problem due to substrate freeze-out by having both electrons and holes needed for recombination travel along the surface. They used gate controlled MOS diodes in which the holes and electrons required for charge pumping action were supplied by the source and drain contacts of the opposite type located on the surface. For example if the source region was n +, the drain region was p+. They used both n+np* or n+pp+ gate controlled diodes. The electrons flowed to the surface from the n + region during one phase of the pulse and holes flowed to the surface from the p+ region during the opposite phase. At low temperatures, the time constant for the emission of holes and electrons is large and hence no emission takes place during rise and fall times in the charge pumping process for either carrier. The charge pumping action proceeds only by successive capture of surface electrons and surface holes from the surface. The channel length was chosen small enough to minimize geometrical component. They also studied the influence of the rise and fall times on the charge pumping current at 4.2, 77 and 300K. The Groeseneken analysis [9] to obtain the interface state energy profile from the derivative of the charge pumping current with respect to the rise or fall time was found to be valid only at higher temperatures. This is because at lower temperatures, the emission process is weak or almost non-existent whereas the Groeseneken analysis assumes that the emission process determines energy band width within which the interface states contributing to the charge pumping current lie. Hsu et al. [48]-[49] carried out charge pumping measurements on a small geometry MOS transistor at very low temperatures. In MOS transistors scaled down to small sizes, the substrate impurity concentration is made high as determined by an appropriate empirical scaling relationship. Increased substrate doping in scaled devices reduces the freez-out effects. Hsu et al. used MOS transistors with a substrate doping level o f - 4 x 10m~cm'3and L = 0.5pro. Due to the high doping in the substrate, the freeze-out effect of impurities is not severe and hence charge pumping current can be measured. The current measured using the

variable base technique at six different temperatures between 295 K and 15 K is shown in Figure 12. The average interface state density corresponding to the maximum charge pumping current measured at low temperature is roughly 5 x 10~/cm2eV~. The maximum charge pumping current is higher at lower temperatures since a larger region of the band gap is swept in going from accumulation to inversion and therefore more interface states contribute to charge pumping current. Furthermore, at room temperature, the emission process takes away the contributions of some of the interface states as discussed earlier. But at lower temperatures, since the emission probability is decreased, the states which previously became negatively charged by hole emission now do so by capturing electrons and thereby contribute to charge pumping current. This is another reason why charge pumping current increases at lower temperatures. In a properly scaled device, the non-uniformity in Vr and V ~ is much less. The emission process also determines the finite slope of the rising and falling edges of the CP current. The rising and falling edges of the bell-shaped charge pumping current become steeper at lower temperatures due to the weakening of the emission process. The increasing edge is steeper because the electron density increases only when near threshold conditions are reached. Similarly, the decreasing edge becomes steeper because the majority carrier density is negligible until VoH exceeds the flatband voltage at which point the carrier density suddenly increases. Low temperature charge pumping can be used to distinguish generation-recombination (g-r) centers among interface states, g-r centers are those that have comparable capture cross-section for both electrons and holes. On the other hand, traps can interact only with one types of carriers. An electron trap will capture electrons and become negatively charged during the ON period and become neutral during the OFF period by electron emission. A hole trap will similarly capture a hole during the OFF period and emit a hole during the ON period. Due to the decrease in emission probability with a decrease in temperature, the traps will contribute to charge pumping current at low temperature only if sufficient time is allowed for the emission to take place. If the pulse repetition frequency is varied (corresponding to varying the ON and OFF periods) in a low

144

C.R. Viswanathan, V. Ramgopal Rao / Microelectronic Engineering 40 (1998) I31-146

temperature charge pumping experiment, the contribution of traps to the charge pumping current will decrease at higher frequencies while the contribution of the g-r centers will remain frequency independent. In Figure 13, Q~p, the charge pumped per pulse measured at 15K, is shown for different frequencies and no frequency dependence is seen.

Hence it can be concluded that the CP current is due only to g - r centers and not due to traps in this device. 8. APPLICATION TO OTHER DEVICES Besides its use in classical bulk CMOS transistors, the charge pumping technique has also been successfully applied for other types of devices, such as SOl-transistors [26]-[29], EEPROM cells [30]-[31] and vertical power (VDMOS) transistors [32]. In the case of SO1 devices the MOSFET performance is determined by both the front and back interfaces. The devices are leaky and unstable in comparison with the bulk devices and hence the typical capacitance measurement techniques are not successful. The CP technique is therefore preferred. The front interface is studied by pulsing the gate while the back interface is studied by pulsing the substrate. Wouters et al. [27] showed, using the CP technique, that the front interface in a laserrecrystallized SOl MOSFET is almost as good as the interface in a bulk device while the back interface is of lower quality. Ouisse et al. [26] measured CP current in thin film SOl MOSFET as a function of the voltage on the back gate with different rise and fall times for the gate pulse. They found evidence of geometric component at high back gate bias values. They also studied the influence of channel length on CP current. Seghir et al. [28] compared the leakage current measurement technique and the CP current measurement using a gate controlled diode structure on SOl material and concluded that the two techniques yield complementary results and that both techniques are useful. Recently Li et al. [29] proposed a simple front-gate charge pumping technique for measurement of interface traps at both the front and the back interfaces of a fully depleted SO1 MOSFETs. Their technique is based on the strong coupling between the two interfaces for the ease of fully depleted MOSFET. In the field of EEPROM and Flash technologies, the charge pumping technique has proven its advantages as well [2]. Witters et al. used the charge pumping technique to investigate the degradation of FETMOS and FLOTOX types of EEPROM cells with cycling [30]. They demonstrated that the initial window opening when cycling FLOTOX EEPROM cells is not

:50

'

'

t6g

40

60K

3O

0
-e --4

-2

GaLe pulse base level (V) Figure 12. Plot of the CP current characteristicsof an n M O S short channel transistorat various temperatures from 15 K to 295 K. The pulse amplitude was kept at 4V.
A

7.0

'1

5.O
im

3.0

m 14

1.0

~ Z
0

O.O
--6

2/,
--4 --2 0 2

i,'?

o llGlz 101R'lz
IOOIQIz

Gate P u l s e b a s e level (V)

Figure 13. The plot of charge pumped per cycle in the variable base mode measured at 15 K temperature with the pulse repetition frequency varied between 1 KHz and 1 MHz in the same device.

C.R. Viswanathan, V. Ramgopal Rao / Microelectronic Engineering 40 (1998) 131-146

145

caused by Fowler-Nordheim induced positive charge, as was concluded before from capacitor experiments, but by band-to-band tunneling induced hot hole injection. This effect can only be observed by measuring directly on the cells. Also in split-gate type mash memory cells, charge pumping is the only technique that allows to investigate the write/erase induced damage in both parts of the channel separately. Wellekens et al. [31] used the technique to make a full analysis of the write/erase degradation in a split gate sourceside injcetion type HIMOS Flash memory cell, demonstrating the amount and the location of the generated and trapped charge and interface traps. Finally, Habas et al. [32] have shown recently that charge pumping can also be applied to standard power Vertical Double-diffused (VDMOS) transistors. The measurement can provide information on the interface between the oxide and the epitaxial-region, and on part of the channel region of the VDMOS device. Like in the ease of bulk CMOS, the shift of the charge pumping characteristics can also be used for the charge trapped in the oxide. This method has been applied for studying the irradiation behavior of VDMOS devices.

recently applied to various MOS based devices such as SOI-transistors, EEPROM cells and vertical power (VDMOS) transistors. Hence the charge pumping is a valuable characterization tool to study MOS interface properties and its degradation under stress. REFERENCES 1. C.R.Viswanathan, Characterization of Submicron Transistors, Chapter-3, edited by H.Haddera, Kluwer Academic Publications, Boston, 1994 2. G.Groeseneken, and H.Maes, Proe of the 21 st International Conference on Mieroeleetronies, vol. 2, p. 581, Yugoslavia, 1997 3. N.S.Saks, G. Groeseneken, and I.De Wolf, Appl. Phys. Lett., vol. 68, p. 1383, 1996 4. G. Groeseneken, l.De Wolf, R.Beilens, and H.E.Maes, IEEE Trans El. Devices, vol. 43, p. 940, 1996 5. N.S.Saks, Appl. Phys. Lea., vol. 70, p. 3380, 1997 6. J. S. Burgler and Paul G. Jespers, IEEE Trans. El. Dev., ED-16, p. 297, 1969. 7. Alexander B.M. Elliot, Soid State Electron., 19, pl 241, 1976 8. W.V. Backensto and C.R. Viswanathan, Proe. lEE, Voi 128, p. 44, 1981 9. Guido Groeseneken, Herman E. Maes, Nicolas Beltran, and Roger F. De Keersmaecker, IEEE Trans. El. Dev., ED-31, p. 42,1984. 10. Guido Groeseneken, Herman E. Maes, Nicolas Beltran, and Roger F. De Keersmaecker, in Proe. of INFOS conf, p. 153, 1983 11. U.Cilingiroglu, Solid-State Electron., vol. 28, p. 1127, 1985 12. W.L. Tseng, J.Appl. Phys., vol. 62, p. 591, 1987 13. M.G. Ancona, N.S. Saks, and D. McCarthy, IEEE Trans. El. Dev., ED-35, p. 2221, 1988 14. R.G.Lee, J.Su, S.S.Chnng, IEEE Trans on Elcetron Devices, vol. 43, p. 81, 1996 15. S. Okhonin, T.Hessler, and M. Dutoit, IEEE Trans. on Electron Devices, vol. 43, p. 605, 1996 16. S.S.Chung, S.M.Cheng, G.H.Lee, and J.C.Guo, 1995 Symposium on VLSI Technology Digest, p. 103

9.

CONCLUSIONS

It is shown that charge pumping can be used for a thorough characterization of MOSFET interface properties. The energy as well as the spatial profiling of MOSFET interface states is possible using charge pumping technique, thanks to the many sophisticated modifications of the basic charge pumping technique that have been developed in the last two decades. Charge pumping is ideally suited for study of non-uniform degradation in MOSFETs generated due to electrical stress or radiation, lcp~, the rising and falling edge of Icp characteristics and their dependence on rise and fall times give complete information on the interface conditions in the device. While CP measurements are normally used to measure fast interface state characteristics, CP teclmique is also being increasingly used for the profiling of fixed oxide charges and slow states. Cryogenic CP measurements enable the scanning of the region of the band gap closer to the band edges. Charge pumping technique has also been

146

C.R. Viswanathan, V. Ramgopal Rao / Microelectronic Engineering 40 (1998) 131-146

17. P. Heremans, H.E. Maes, and N. Saks, IEEE El. Dev. Lett., EDL-7, p. 428, 1986. 18. Wenliang Chen, Artur Balasinski, Binglong Zhang, and Tso-Ping Ma, IEEE El. Dev. Lettr., EDL- 13, 1992. 19. D.S.Ang, and C.H.Ling., IEEE Elect. Dev. Lett., vol. 19, p. 23, 1998 20. H.Li, Y. Chu, C. Wu, IEEE Trans on Electron Devices, vol. 44, p. 782, 1997 21. M. Tsuchiaki, H.Hara, T.Morimoto, and H. Iwai, IEEE Trans. on Elect. Devices, vol. 40, p. 1768, 1993 22. R.E.Paulsen, M.H.White, IEEE Trans. on Elect. Devices, vol. 41, p. 1213, 1994 23. C.Chen, and T.P.Ma, 1996 Symposium on VLSI Technology Digest, p. 230 24. J.L.Autran, C.Chabrerie, P.Paillet, O.Flament, J.L.Leray, and J.C.Boudenot, IEEE Trans. on Nuclear Science, vol. 43, p. 2547, 1996 25. R.Ghodsi, Y.Yeow, C.H.Ling, and M.K.Alam, IEEE Trans on Electron Devices, vol. 41, p. 2423, 1994 26. Thierry Ouisse, Sorin Cristoloveanu, Tarek Elewa, Hisham Haddara, Gerard Bonel, and Dimitris E. loannou, IEEE Trans. El. Dev., Vol 38, p. 1432, 1991 27. Dirk J. Wouters, Marnix R. Tack, Guido V. Groeseneken, Herman E. Maes, and Cor L. Claeys, IEEE Trans. El. Dev., Vol 36, p. 1746, 1989 28. H.Seghir, S.Cristolovoanu, R.Jerisian, J.Oualid, A.J.Auberton-Herve, IEEE Trans on Electron Devices, vol. 40, p. 1104, 1993 29. Y.Li, G.Wang, and T.P.Ma, 1995 IEDM Technical Digest, p. 643 30. J.Witters, G.Groeseneken, H.E.Maes, IEEE Trans. on Electron Devices, vol. 36, p. 1663, 1989 31. D.Wellekens, J.Van Houdt, L. Faraone, G.Groseneken, and H.E. Maes, IEEE Trans. on Electron Devices, vol. 42, p. 1992, 1995 32. P.Habas, Z. Prijic, D. Pantie, and N.Stojadinovic, IEEE Trans. on Electron Devices, vol. 43, p. 2197, 1996 33. Paul Heremans, Johan Witters, Guido Groeseneken and Herman Maes, IEEE Trans. El. Dev., ED-36, p. 1318, 1989. 34. G.Van den bosch, G.Groseneken, H.E.Maes, IEEE Electron Device Lett., vol. 14, p. 107, 1993

35. P.Habas, G.Groseneken, and G.Van den bosch, Proc. 21 st International Conference on Microelectronics, vol. 2, p. 599, Yugoslavia, 1997 36. James E. Chung and Richard S. MuUer, Solid St. Electron., Vol 32,, p. 867, 1989 37. W.L. Tseng, Jour. AppL Phys., Vol 62, p. 591, 1987. 38. N.S.Saks, and M.G.Ancona, IEEE Electron Dev Letters, vol. 11, p. 339, 1990 39. Ugur Cilingiroglu, IEEE Trans. El. Dev., ED37, p. 267,1990 40. Ugur Cilingiroglu, IEEE Trans. El. Dev., ED35, p. 2391, 1988. 41. Martin Kejhar, IEEE El. Dev. Lettr., EDL-6, p. 344,1992. 42. Wenliang Chen, and Tso-Ping Ma, I'EEE El. Dev. Lettr., EDL-I 2, p. 393, 1991 43. Wenliang Chen, Artur Balasinski, and TsoPing Ma, IEEE Trans. El. Dev., ED-40, p. 187,1993 44. X.M Li and MJ. Deen, Sol. St. Electron., Vol 35, p. 1059, 1992 45. Dominique Vuiilame, Joan Claude Marchetaux, and Alain Boudou, IEEE El. Dev. Lett., EDL-12, p. 60, 1991 46. C.R.Viswanathan, B.C.Burkey, G.Lubberts, and T.J.Tredwell, IEEE Trans. on Electron Devices, vol. 32, p. 932, 1985 47. CH. Nguyen-Duc, G.Ghibaudo, and F.Balestra, Phys. Stat. Sol., a,126, p. 139, 1991 48. Jen-Tai Hsu, C.R.Viswanathan, R.Divakanmi, and XLi, Proceedings of the 23 rd European Solid State Device Research Conference, Grenoble, France, p. 527, 1993 49. Jen-Tai Hsu, and C.R.Viswanathan, Japan. Joum. Appl. Phys., voi. 33, p. 247, 1994