You are on page 1of 5

630 IEEE TRASSACTIONS os ELECTROS DEVICES, VOL. ED-13, xos.

8/9, AUGUST/SEPTEMBER, 1966

Some New Aspects of Thermal Instability of the


Current Distribution in Power Transistors
F. BERGMANN AND D. GERSTNER

Abstract-An importantcharacteristic of secondbreakdown in bility of the current distribution. These parameters are
p-n junctions is the currentconstriction to a small region. This may devicegeometry, power density,andtemperature de-
becaused by a thermalfeedbackmechanism, a s discussed by
Scarlett and Shockley, and by Bergmann and Gerstner. pendence of current. Someexperimentalresults on the
A brief review of this theory is given, illustrated by experimental temperature coefficient of current are presented
for
results of a simple model arrangement consisting of three thermally typical Si planar transistor configurations. These results
coupled transistors. The
essentialparameters influencing the allow an interpretation of the fact that the thermal sta-
thermal stability of the current distribution are device geometry, bility of the current distribution is much better in the
power density, and temperature dependence of current.
I t is widely known that second breakdown occurs at high voltages
case of high current and low voltage than itis in the case
at a much lower power level than a t low voltages. To allow a more of low current and high voltage.
detailed discussion of this effect in view of thermal stability, we
determined experimentally the temperature coefficient of transistor MODELDEMONSTRATION
OF THERMAL
INSTABILITY
current for various Si planar transistors as a function of current, A large area HF power transistor can be regarded as a
voltage, and junction temperature. The experimental procedure is
described and the results are discussed.
parallel connection of many small area transistors. Actu-
Theexperimentalvalues of thetemperature coefficient rangeally power transistors are constructed in this way, e.g.,
from 0.08 to 0.01 1 / " C . Thevalues forhigh currentsaremuch when aninterdigitatedstructure (comb structure) is
lower than predicted by the theory of Ebers and Moll. It thus can used. The question is: how will thecurrentdistribute
easily be understood why, in the case of high current, and low volt-
itself amongst the single transistors?
age, the thermaI stability of the current distribution is much better
than in the case of low current and high voltage.
Figure 1 shows asimple model arrangement [lo] to
measure the currentdistributioninthreetransistors
INTRODUCTION working inparallel. Thethreetransistorswithinthe
dashedlinerepresentherealarge areatransistor.The
HE PHENOMENOK of second breakdown in single collector currents J,,, J o , J,, of the transistor
transistors and diodes is associated with a current parts are measured. The sum of these currents and the
constriction to a small area [1]-/7]. In many cases common collector voltage V C Edetermine the power load-
this may becausedbya thermal feedbackmechanism ing of the whole transist,orconfiguration. This power
according to a theory proposed by Shockley and Scarlett loading is stabilized to a nearly constant value by means
181, [9], and independently by Bergmann and Gerstner[lo] of the emither resistor R, which is common to the tran-
in 1963. If the thermalfeedback is overcritical,the current sistors of the model.
distribution becomes unst'able,and small parts of the Figure 2 shows the measured currents J,,,J,,, Jcs
transistor bear almost all of the current. This instability as a function of t,ime after application of power. At the
can occur independently of possible diffusion defects of beginning the currents J c , , J c 2 , Jo areapproximately
thetransistor.Althoughthetransistor is loaded below equalto 10 mA. Thecurrentint'ransistor 1 increases
the theoreticalmaximum power dissipation,calculated steadily,whereas thecurrentintransistor 2 decreases.
from the thermal resistance, local overheating (hot spots) The currentintransistor3 a t first increases a small
may occur. When a critical temperature is reached at a amount, and then also decreases. Finally after 5 minutes,
hot spot, an intrinsic zone is formed which short circuits transistor 1 carries 96 percent of the total current, which
the space charge region of the p-n junction [lI], 1121. This implies also almost all of the supplied power,while the
results in a typical voltage reduction over the transistor. remainingt'ransistorscarry less than 3 percenteach.
I n some cases evenmolten zones have been observed Throughout,thesum of thecurrents remainsapproxi-
[13], [14]. With reverse-bias conditions, in addiOion to t'he matelyconstant J , w 30 mA. Thus, fromonly ex-
thermal effects, electrical fields haveto be takeninto ternalvoltage V C Rand current J , observations of our
account, but nevertheless the thermal instability seems power transistor model, the transistor seems to operate
to be an important feature of second breakdown. well.
The purposeof this paper is to point out the importance Under the special conditions of this model, transistor
of the essential parameters influencing the thermal sta- 1 mas not dest'royed because the whole system was driven
far below its maximum allowable power dissipation.
But under real, operational conditions of a power tran-
Manuscript received December 7 , 1965. sistor when loaded near the theoretical maximum power
The authors are with Telefunken Aktiengesellschaft, Heilbronn,
Germany. dissipation,calculatedfrom thethermal resistance, i t
1966 BERGMANN
INSTABILITY
THERMAL
AND GERSTNER: 631

1. 2. 3. IJC, thermal coupling with its neighbors, it will increase its


temperature more thantheothers,Thisinturn will
cause a further increase of the current, and so on. It is
notimportant how large the temperature difference or
the other differences between the various transistors are
at the beginning. When the feedback mechanism is over-
critical, the currentdistribution will beunstable,and
current crowding will occur.
STABILITY PARAMETERS

From a qualitative discussion of our 3-transistor model,


Fig. 1. Arrangement for demonstration of currentdistributionin three parameters can be predicted that will influence the
three thermally coupled transistors.
thermalstability of currentdistributionin our model
or in a real large-area transistor.
A. Device Geometry.
Device geometry determines whether or not tempera-
A ture differences between the various active parts of the
power transistor will be leveled out. If thethermal
couplingbetween thethreetransistorsand/ortheir
thermal coupling totheheat sinkisverystrong, the
temperature differences will besmalleven when the
transistorsdonotdissipatethesame power. Suppose,
0 2.5 5.0min as a limiting case, the whole transistor is concentrated at
Time t
one mathematical point; then no temperature differences
Fig. 2. Unstable currentdistribution among threetransistorsin
parallelfor two different initial conditions. Atthe right-hand within this point would be possible.
side, transistor 3 had been preheated.
B. Temperature Coe$icient of the Transistor Current.
may be that one part of the transistor would have burned A currentinstability of the described type is only
out. possible when there is an increase of transistor current
It is important to note that this overloading of tran- with a rising temperature. Instability will be more serious
sistor 1 in our model arrangement is not predetermined when the temperature coefficient of the transistor current
by adifferentelectricalcharacteristic of the transistor, is high.
which would,forexample,correspond to aweakpoint
in the diffusion of the power transistor. This is demon- C. Power Level.
stratedattheright-handside of Fig. 2. Here we had For a certain transistor system, thermal stability will
preheatedtransistor 3 a short whilebefore application be better a t a low rather than a t ahigh power level.
of the power. Now transistor 3increases itscurrent Suppose in our 3-transistor model, there is a temperature
steadily and finally carries approximately 97 percent of unbalance that results in a 5 percent increase of current
thetotalcurrent, while transistors 1 and 2 retainthe for one transistor. At a power level of, say, 20 mW, this
remaining current.Thus,thecurrentinstabilityis a means additional power of only 1 mW,whereas a t 200
fundamental one, and can not be prevented by carefully mW this means 10 mW additional power.
avoidingweakpointsin thetransistor electricalstruc- For geometrically simple transistorstructuresand
ture only. simple boundary conditions, the conditions for stable or
This behaviorcanbeexplainedbya thermal feed- unstable current distribution can be calculated by solving
backmechanism, which ismade possible bythevery the differentialequation of heat conduction [lo], [ls].
rapid increase of the collector current J , (approximately For a power transistor with a simple rectangular shape
equal to the emitter current J,) with a rise in tempera- withlength a, thickness h, the result of thestability
ture.For abroadtemperature range, thistemperature analysis is shown inFig. 3. At the top surface,a dis-
dependence may beapproximatedby an exponential tributed heat source (e.g., a very fine interdigitated tran-
relation. sistor structure),and at the undersurface, a flatand
In our 3-transistor model the three transistors receive perfect heat drain, have been assumed.
the samevoltage V,, because theyare connected in Again, here are the three parameters that we already
parallel. So a transistor that has a temperature slightly know from the qualitative discussion. Using the stability
higher than the others will bear more c,urrent than the chart, (Fig. 3), it is possible to point out the relative im-
others.Thismeansthetransistor will dissipatemore portance of theseparameters. Devicegeometry for this
power, and if the temperature unbalance is not leveled transistorstructureis described by a,%, the ratio of
out either by thermal coupling to the heat sink, or by length to thickness of the transistor crystal. This indi-
632 IEEE TRANSACTIONS ON ELECTRON DEVICES AUGUST/SEPTEMBER

cates that in a large areatransistor,thermal coupling 10


distributed heat source
depends on the greatest distance between the active parts
of thetransistorandthe thickness of the chip. The
temperaturedependence of t'hetransistorcurrent at the
certain operating point is characterized by the tempera- 55
ture coefficient B, which is defined by

B=L%l . .
dT V B E - C O a S t . . .... . . , , :
JC . ;.

-
b<a
Thistemperature coefficient depends on the semicon- .B.AT
ductormaterial(bandgap),the operatingpoint of the Fig. 3. Stability chart for a power transistor of rectangular shape.
transistor (especially current density), and the junction
temperature. AT is the temperature difference between
the transistor junctions and the heat sink. AT may be
calculated as the product of the heat produced in the
collectorjunction, andthethermal resistancebetween
the transistorjunctions and the heatsink. AT is a measure
for the power density at which the transistor is operated.
In the stability chartwe find three regions.
1. Stable mgion, the lowpower region. In this region,
stableoperation of transistors is possible, regardless of
their geometry (large or small area).
2. Conditionally stableregion. I n this region, stable
operation is possible if the total power is limited; that is,
if the transistor is stabilizedagainst thermal runaway.
As canbe seen from the stabilitychart,stabilization
against thermal runaway a t a certain value of B - AT can I I
be achieved only for suitably low values of a/h; that is, 50 100 2 0
only for transistors of sufficiently small area. Collector voltage V,, ---b
3. Unstableregion. Inthis region, thermal coupling Fig. 4. Safe operating range of a power transistor.
between the active parts of the transistor is too weak.
Thus the current distributionin the transistor is unstable
loaded to approximately 100 watts, whereas at 100 volts
and current crowding will occur. This type of instability
the power allowable is only 20 watts.
cannot be avoided by limiting the total dissipation to a
How canthisbe explained? We believe that it is a
constant value,as, for example, by an emitter resistor.
consequence of thecurrentdependence of temperature
Whencomparingthe concept of thermalinstability coefficient ontransistorcurrent.Thisparameterhas
with experimental results on the phenomenon on second already been mentioned when we were discussing the
breakdown, one finds that the thermal concept fits very stabilitychart.Remember, for example, a reduction of
well the main features observed in forward-bias second the temperature coefficient by, say, a factor of two will
breakdown. A complete understanding of second break- allow thetemperature difference, and, therefore, also
down, including reverse-bias second breakdown, has not the power dissipation,toincreaseby thesamefactor
yet been achieved 1161. But it is hoped that by proper withoutchangingthestabilitybehavior of the current
refinement of the existingtheories, further peculiarities distribution. Indeed a current dependence of this temper-
will be understood. An example is the temperature de- ature coefficient can be predicted by t'he simple theory of
pendence of triggering energy or of the delay time [5], [16]. the junctiontransistor,forexamplebythetheory of
It seems possible that this temperature dependence can Ebers and Moll 1171. By this theory, one would expect
be explained by taking into account thevariation of heat for reasonable current densities, a value of B = O.lO/OC,
conductance with temperature, which is quite important, which means that the transistor current is increased by
for example, in silicon. Another well-known feature is 10 percent when the junction temperature is increased by
the depen.dence of forward-bias second breakdown be- 1°C. Concerning the current dependenceof B, this theory
havior on operating point (Fig. 4).As can be seen from predicts that B decreases by 2.3/T; that is, about 0.007/"C
the lowest curve, which is for the dc case, the tendency for a ten times increase in current density [8]. Obviously
of thetransistortobedishrbedby unstable current thistheoreticalcurrentdependence of the temperature
distribution ismuchhigher at highvoltages and low coefficient B is too weak and would not explain the large
currents than it is at low voltages and high currents. For differences of current stability which are observed experi-
example,in the 10 voltrange, thistransistormaybe mentally.
1966 BERGMANN AND GERSTNER: THERMAL INSTABILITY 633

MEASUREMENT B
OF TEMPERATURECOEFFICIENT

Since the transistor theory is based on many simpli-


fyingassumptions which are generally not fulfilled, we
havemeasured thetemperature coefficients for silicon
planar transistors and their dependence on current level
and junction temperature.
I n order to avoid instability of current distribution a t
too low power levels, we used small area transistors. But
these transistors were similar to typical large area tran-
sistors when junction depth, base width, diffusion layer
sheet resistance, and so on, are considered. Tj -
On the right-hand side of Fig. 5, a sketch of our meas- Fig. 5. Base emitter voltage VBEvs. junction temperature Ti
uringcircuit is given. Thetransistors were putin a for various collector currents J c for a silicon planar transistor
2 N 1613.
thermostat, and for fixed collector currents J , the base
emitter voltage V B , was measured with a high precision mA
digital voltmeter as a function of junction temperature
Ti.It should be mentioned that the voltage drop over
the base resistance, which has a stabilizing influence, has t lo
been subtracted from the measured values of V B E Thus . u
7

the following considerations are also valid for open base


conditions, which are veryimportantin practice. To 1

obtain the right value of Ti, we firstmeasured V B Ea t


several collector voltages for fixed collector current as a
-
function of case temperature T,. This measurement gave
-
".I ,
0 50 - 100 150 200 "C
aset of parallelcurves which could be extrapolated t o Tj
zero dissipated power using the relation' Fig. 6. Collector current J c vs.junction temperature T i for a
silicon planartransistor 2 N 1613. Parameter 1s base emitter
Tcz - T c , , which has been derived in voltage VBE.
T i= T,, + ~- 1
[lS].
-
V C-
El -

vCE2

In this way we obtained the set of curves shown on the


left-hand side of Fig. 5 for a 2 N 1613 transistor.
I n order t o determine the temperature coefficient B,
this set of curves can be replotted in the manner shown
in Fig. 6. Thisis a log plot of collector current J , against
junction temperature T ifor constant base emitter volt-
age V B E . By graphical differentiation of these curves, the - - theoretlca
experimental values of the temperature coefficient B are 0.00 !
obtained. Throughout the whole curve set, it can be seen 0.1 1 10 100 mA
Jc
that the slope is much less steep a t high currents than it
Fig. 7. Temperature coefficient B vs. collector current JC for a
is at low currents, and that theslope is not very tempera- silicon planar transistor 2 N 1613.
ture dependent.
I n Fig. 7, B has been plotted for twotemperatures
decreases from 0.045 t o 0.015 according to the 100°C curve;
(50 and 100°C) as a function of transistor current. For
that is, by about a factor of three.
comparison, twotheoretical curves-the dotted ones-
Our experiments show that at high current densities,
are given. These show the weak current and temperature
the temperaturedependence of current is muchlower than
dependencementionedabove. I n contrasttothe theo-
predicted by theory. Thus can it be easily understood why,
retical curves, the experimental values of B tend t o much
in a large-area transistor a t acertain power level, the
lower values when the current density is increased. For
stabilityagainst second breakdown is much better for
an example, when J , is increased from 10 to 100 mA, B
high current than for low current operation.
CONCLUSIONS
1 This measurement procedure is actually a simple static method In large-area transistors, an instability of current dis-
'or the determination of the thermal resistance between the transis-
;or junctions and the case [18]. We noticed that the thermal resis- tribution can occur due to a thermal feedback mecha-
jances given by the datasheets were not accurateenough to calculate nism. This type of instability which may initiate second
;he junction temperature from case temperature and power dissipa-
ion, especially a t high power levels. breakdownhas been demonstratedexperimentallyina
634 IEEE TRANSACTIONS OK ELECTRON DEVICES AUGUST/SEPTEMBER

stable cells are combined to a large area configuration with


stabilizing emitter resistors for each cell. Of course, the
internal B within these cells is not influenced, but that is
not necessary because the single parts are designed in an
undercritical size. Figure 8 shows, as an example, a power
transistor which is divided into eight small-area devices,
eachhaving a stabilizing emitter resistor. The resistors
in this example are evaporated KiCr resistors.

REFERENCES
[I] C. G. Thornton and C. D. Simmons, “A new high current mode
of transistor operation,” I R E Trans. on Electron Devices, vol.
ED-5, pp. 6-10, January 1958.
[2] R. Greenburg,“Breakdownvoltagein power transistors,,’
SemiconductorProducts, vol. 4, pp. 21-25, November 1961.
[3] J. Thire, ‘[Le PhhomBne de pincement sur les transistors de
puissance en commutation,” Colloque Internat’l sur les Dis-
positifs ci Shmiconducteurs, vol. I : Production, pp. 277-293,
1961.
_-
[41 F. Weitzsch. “ZumEinschnureffekt bei Transistoren, die im
Durchbruchsgebietbetriebenwerden,” Arch. elektr. ‘ Ubertr.,
vol. 16, pp. 1-8, January 1962.
Fig. 8. Second breakdown proof power transistor. The large-area [5] H. A. Schafft and J. C. French, “Second breakdown in transis-
device is divided into eight cells, each having a stabilizing emitter tors,” I R E Trans. on Elecfron Devices, vol. ED-9, pp. 129-136,
resktor. March 1962.
[6] J. Taucand A. Abraham, ‘‘Der elektrische Durchschlag an
p-n-Ubergangen in Silizium,” Abh. Akad. Wiss. Berlin, Klasse
simple model arrangement.Therearethreeessential f. Math. Phys. u. Techn., no. 1, pp. 27-33, January 1960.
r71 J. Tauc and A. Abraham. “Thermal breakdown in silicon z7-n
parameters influencing the stability of the current distri-
L _ I

junctions,” Phys. Rev., vol. 108, pp. 936-937, November 1957.


bution: devicegeometry, temperature dependence of [SI R. M. Scarlett and W. Shockley, “Secondary breakdown and
hot spots in power transistors,” 196.9 I E E E Internat’lConv.
current, and power level. Rec., pt. 3, pp. 3-13.
I n order to improvestabilityagainst second break- 191 R. M. Scarlett, W. Shockley, and R. H. Haitz, “Thermal in-
stabilities and hot spots in junction transistors,” in Physics of
down initiated by this thermal mechanism, the following Failure in Electronics, vol. 1, M. F. Goldberg and J. Vaccaro,
precautions may be taken. Eds. Baltimore, Md.: Spartan Books, 1963, pp. 194-203.
[lo] F. Bergmann and D. Gerstner, “Thermisch bedingte stromein-
schniiru!Ig bei Hochfrequenz-Leistungstransistoren,” Arch.
1) Device geometry should be designed so as to pro- electr. Obertr., vol. 17, pp. 467-475, October 1963.
vide strong thermal coupling. [11] H. Melchior and M. J. 0. Strutt, “Secondary breakdownin
2) The temperature coefficient of current should be re- April 1964:
I __
transistors.” Proc. IEEE (CorresDondence). vol. 52. nu. 439-440,
ducedfor the single active parts the transistor is [12] H. belchior and M. J. 0. Strutt, “On the initiation of second
breakdown in diodes and transistors,” Scientia Electrica, vol. 10,
composed of. pp. 139-141, December 1964.
3) The power level should be reduced to a value which [13]A. C. English and H. M. Power, “Mesoplasma breakdown in
silicon junctions,” Proc. IEEE (Correspondence), vol. 51, pp.
assures stable current distribution. 500-501, March 1963.
[14] A. C. Enghsh, “Mesoplasmas and second breakdown in silicon
Thethird possibility does not seem to beverysatis- junctions,” Solid-SlateElectronics, vol. 6, pp. 511-521, Sep-
factory, since it does not cure the inst’ability.It is merely tember-October 1963.
F. Weitzsch, ‘<ZurTheorie des zweiten Durchbruchs bei Transis-
a withdrawal from the dangerous high power region and toren,” Arch.elektr.gbertr., vol. 19, pp. 27-42, January 1955.
must be paid for by inconveniently high capacitance in H. A. Schafft and J. C. French, “A survey of second break-
down,” presented a t the 1965 Internat’lElectron Devices
thetransistorandbyadditional yield problems when Meeting, Session 14.1, Washington, D. C .
producing such overdimensioned devices. But there is a J. J. Ebers and J. L. Moll, “Large signal behavior of junction
transistors,” Proc. IRE, vol. 42, pp. 1761-1772, December 1954.
very promising possibility forthe construction of a second D.Gerstner, “Statische Bestimmung des thermlschen Wider-
breakdown proof power transistor by combining the first standes R i t h und der Sperrschichttemperatur T i von Transis-
toren,” Nachrichtentechn. zeitschr., vol. 19, pp. 195-199,
tm7o stability design concepts [19]. The large-area power April 1966.
transistor is divided into single cells which are small D. Gerstner, ‘[HF-Leistungstransistoren-Keuere Entwicklungen
mit Techniken der Mikroelektronik,” Internationale Elek-
enough to assurestabilitywithineach cell. Then these tronischeRundschau, vol. 19, pp. 495-501, September 1965.

You might also like