CAD of Electronic Circuits, | Modeling the Bipolar Transistor lan E.Getreu Tektronix Labors Beaverton, Oregon, U.S.A. = ‘Amsterdam — Oxford — New YorkELSEVIER SCIENTIFIC PUBLISHING COMPANY 8386 Jan van Galenatrant P.O. Box 211, 1000 AE Amsterdam, The Netherlands Distributors for the United States and Canada’ ELSEVIER NORTH-HOLLAND INC. 52, Vanderbilt Avenue New York, N.Y. 10017 Paos so7: A “Wostaing the bipolar tranctstor. (caapater-atded design of electrons cirealte 5 ve 2) Biotlogreshy traniistore-ethemstical aodele. |e Tea Tirerhge 307 den gst 15 Tenn ocltishines 2 " mane ISBN 0-444-41722-2 (Vol. 1) ISBN 0-444-41723-0 (Series) Copyright © Tektronix Ine., Beaverton, Oregon 97077, 1976 “AP vights reserved. No part of this publication may be reproduced, stored in a ‘fetrieval system of transmitted in any form or by any means, electroni jeal, photocopying, recording or otherwise, without the prior written permission Sf ike publisher, Elsevier Scientific Publishing Company, P.O. Box 330, 1000 AH “Amsterdam, The Netherlands Printed in The NetherlandsAcknowledgments ‘The author 1s indebted to many people who assisted ‘in the preparation of this book. He is grateful to his colleagues at Tektronix and several people at other institutions for their valuable feedback from earlier versions. The assistance of Fred Severson (Tektronix), Professors Robert Dutton (Stanford University), Don Pederson and Robert Meyer (University of California, Gerkeley) and Or. Richard Vaughan (University of New South Wales, Australia) in lengthy discussions and review of the manuscripts 1s greatly appreciated. The comments and advice rendered by George Wilson, Car? Bates, David Hannaford, Or. Binoy Rosario, Or. Jin Smith, Dick Hung and Jack Millay (Tektronix), Ors. Don Scharfetter and Larry Nagel (Bell Laboratories) and Johan Brinch (University of New South ales, Australia) are also very gratefully acknowledged. The contributions of several people to various sections of the book in ‘the measurement, section are appreciated and are acknow!~ edged in the appropriate places. The assistance of Joyce Lekas, Fred Severson, Or, Robert Nordstrom, Ted Niimi, Jeanne Galick, Doreen Weaver, Oiana Clark and Jane West in the preparation of the manuscripts and Growing of the Figures fas been invaluable. Sone of the materiel in Section 1, Introduction; Section 11, Theoretical Derivation of the Models, 2.1 and 2.2; and Section 111, Parameter Measurenents 3.1, 3.2 and 3.3 appeared in ¢ series of articles in Electronics, Septenber 19, October 31, and Novenber 14, 1974, Copyright © MeGraw-HitT, IncledContents SECTION 1 - INTRODUCTION Usefulness of Conputer-aided Design onjective Structure Notation and Orientation Input and Mode! Parameters Dependence of Models on the Physics of the Device « Model Notation Which Model to Use? SECTION 2 - THEORETICAL DERIVATION OF THE MODELS 2.1 Introduction 2.2. The EMy Model 2.2.1. The Injection Version 2.2.2 The Transport Version | 2.2.3. Mw Alternative Form the Non} inear Hybrid-» 1823 2.28 2.2.5 Temperature Variation . . . see Sunmary and Additional Comments The EMy Model 2.3.1 2.3.2 2.3.3 2.3.4 Improved de Characterization abrge Dirge eee ory - charge-storage Effects 2) Junction Capacitors b) Diffusion Capacitors c) Substrate Capacitor sia1T-signal (Linearized) Mg Model. Summary. . 2.4 The EM Model 2.40 An Improved dc Model at a Given Temperature - a) Basewidth Modulation... 2... ) ge Varfation with Current... 5 « i) Region II: Mid Currents . 41) Region I: Low Currents 441) Region III: High Currents iv) The Full Picture Regions I, I and HL... v) Effect of Ohmic Resistances vive 20 a 2 28 2 2 28 28 23 2 8 39 40 43 “6 “4 “8 50 50 55 87 572.4.2 2.4.3 244 2. 5 2.5. The GP 2.5.1 2.5.2 2.5.3 Wi) Bg Versus Ig Input Parameters in SLIC and SINC Vit) fgg and Its Variation with Current An Inproved Charge-storage Model at a Given Tenperature a) Improved Ce - rf Model b) Variation of 1g with Current An Improved Variation with Operating Temperature . . a) Physics-based Tenperature Variation iy 44) jp and Cig b) Temperature-dependent Paraneters that Require Extra Input Parameters smat-signal (Linearized) EM, Model Surmary and Conclusions Model ee eee eee Introduction The Physical Definition of Is The Up Concept 2 ee The Components of Op 4) Mathematical Derivation of Components of O 58 6 6 6 63 65 65 65 66 or 6 6 6 6 1” a 83 a4b) Physical Significance of Components of Q, - sees 86 5) Op + = : cee HI) Op eee eee 118) & cee 88 iv) Q Lee ee 88 WW eee eee see 89 vi) Effect of Operating Regions a) vit) Summary ee : 30 2.5.4 Evaluation of gy. - + Lees a) Component ag : ses 9 ) Component go es 2. 96 €) Component. a¢ beens 98 4) Component a. = bene 99 e) Effect of Depletion Approximation. ee +. 99 HQ vee eee ee = 100 Hi) Igy ss» eee 100 18) pe Qe . en) WW) Qs Qg erence 101 1) Solution for gy. 22 eee ees 103 4g) High-tevel Injection Solution oe ee ee 108 1) Verification of High-Current solution. . eee 105 is) Simplification of a2. +. + + 106 avi2.6 2.5.5 2.5.6 2.5.7 2.5.8 h) Final Solution 4) Comparison with Gunme1-Poon Derivation Base-widening Effects . a) de Characteristics - b) ac Characteristics i) Effect of Base-widening on tg fi) Emitter Delay (1) Versus Ic 441) tp Versus Ip Comparison of the GP Model with the EMj Model a) Basewidth Modulation b) High-level Injection ¢) @ Versus d) tg Versus Ie Smal1-signal (Linearized) GP Model Summary a) de Model b) Charge-storage Model. cc) Temperature Variation Model Limitations of the Models a) Three-dimensional Effects b) Breakdown c) Saturation 09 no m ne ne n3 ne ne Ww ne na 1 1a 1a vee 13 123 124 124 125 126SECTION 3 - PARAMETER MEASUREMENTS 3.1 3.2 3.3 Introduction - EM, Model Paraneter Measurements... . er Fe Is 1, Fy EM Model Parameter Measurements 6. eves rhe Cor tp (or a) Definitions of Two Limiting Values of ry. b) Measurenent of repgmmat and Tesat a) Introduction b) Comparison of Measurement Techniques for ry, c) Smal1-signal Measurements 4) Pulse Measurement Techniques... . =. e) Noise Measurenent Technique and @ fy at ley Meg) oo eee a) Determination of xp from fy. ees b) Determination of fy -viti- wz 129 130 132 133 136 137 138 140 144 ue 16 151 151 181 152 158 164 165 169 170 ratg (0F tsar) sup °F Cos 3.4 EM, Mode! Paraneter Neasurenents Wee pus Coo Mey and 0 Cor Seyaxs Tyan &FLow Teygye BCEC and Yop) Bayes Cq> Mey aNd Og patio L we tco Tey, Tey 3.5. GP Model Paraneter Measurements Ig ss beeen - Ween 1 lee 8 APPENDIX 1 - A Comparison of the Transport Notation With the Injection Notation APPENDIX 2 - EM, Model Basewidth Modulation Analysis 176 179 180 v2 188 200 201 202 204 205 206 207 a3 215 as a9 223APPENDIX 3 - Derivation of the Five Components of Qy in the GP Model . APPENDIX 4 - The Accuracy of the EM, and GP Basewidth Modulation Models APPENDIX 5 - The Smal1-Signal, Linearized EMy and GP Models APPENDIX 6 - Input Parameters Cross-Reference for SLIC, SINC and SPICE REFERENCES 229 233 239 246 + 255Usefulness of Computer-aided Design It has almost become a cliché to coment on the rapid arowth of the computer as a tool in the design of circuits, both discrete and integrated. Althouah the computer is often thouaht of only in terms of a cheap and fast breadboard (or dry tab), computer-aided design (CAD) enables the circuit desiqner to do thinas which are not possible with other techniques. Using the computer, he can = Observe waveforms and frequency responses of volt ages and currents without loading the circuit as a probe would in an actual circuit, = Predict the performance of an IC at hiah frequen cies, without the parasitics a breadboard intro- duces. = Use ideal devices selectively, such as one with an infinite bandwidth or a very large gain, to isolate the effects of various device parameters fon the circuit performance or to do futuristic “blue sky" analyses. = Feed into a circuit ideal waveforms, such as ex- tremely fast pulses or a mixture of ‘pulses and sinusoids = Separate out de circuitry in order to understand the basic part of the circuit. = Open 2 feedback loop without disturbing the de levels = Determine the poles and zeros of a transfer func~ tion for even larue circuits = 0 noise, sensitivity, worstcase and statistical analyses.Objective ‘The circuit designer today has programs available which allow him to do a wide variety of analyses. However, the different pro- grans with their differing input formats, rules, notations and device models can be very confusing and discouraging to the inex- perienced user. Fortunately, the program input formats and rules are normally well documented, so if mistakes are made they are relatively easily detected (either by the computer, the program or the user). The biggest problem, however, Ties in the Tack of standardization of the notation and device models being used and ‘the measurenent of the model parameters. This book 1s aimed at reducing the modeling confusion by systematically describing how to model the bipolar junction tran- sistor (BJT). Emphasis is on the nonlinear, large-signal models used in nonlinear dc and transient analyses, since this is where rmost model ing problems are encountered. The linear model (the well-known hybrid-1 model) which is already fairly well docunented and easy to work with, is simply a linearized version of the non- linear model presented. This linearization process is also described. Structure The book ties together recent developments in the modeling of the BIT with the established models in such a way that the reader can use most, if not all, of the circuit analysis programs cur- rently available, To this end, two aspects of the modeling are covered. First, in Section 2 of the book, the effects that are modeled are explained in terms of the physics of the device so that the reader obtains an understanding of the assumptions and limitationsof the model and the ability to handle the different notations that he will encounter. The descriptions start with the simplest model and build up to the most sonhisticated model The second modeling aspect is the measurement of the device parameters. Techniques for obteining these paraneters from termi~ nal. measure its are given in Section 3 Notation and Orientation The model paraneters and their notations are oriented towards three conputer prograns, SLC!" (up to version t}, sinc!?) (up to Version 0) and sPrce'?) (Version 1), all readily available From the University of California at Berkeley.” These programs are excellent vehicles for this book because (i) their built-in tran- sistor models cover the full range from very simple to very sophis- ticated, (11) the strong dependence of the models on the physics of the device Yeads to concise input requirenents, (111) the programs are relatively widely accepted, and (iv) they are readily available. The use of these prograns 2s examples does not restrict the applicability of this book. The paraneters described can normally be readily transformed for other prograns in which models are based on the Fbers-Mot model") (in the de and large-signal, non- Vinear case), the hybrid-» nodel®) (in the snall-sianal, linear case), or are user-defined. A cross-reference for the SLIC, SINC and SPICE model parameters with those described herein is given at the end (Appendix 6). Tnece prograes can be obteined for a nominal nanahing charge by contacting Professor 9. 0. Pederson, Clestronie Research Lat orator California at Gerketey, California, 94720, U.S.R. ATT SINC. and SPE} are written im FOKISAN IY for'a CBC 6000 Series c= Versions sf the progrars for other camuters are also available from other institutionsInput and Model Parameters to specify a transistor model completely, a program requires unree types of information: fundamental constants, operating condi- tions and model parameters. Only one type (the last) is specified by the user. the fundamental physical constants, such as Boltzmann's on- stant (k) and electronic charge (q), are normally defined inside the progran, The operating conditions define the circunstances der which the nodel equations are to be used. In a nodal anal- ysis program, for example, the operating conditions are normally the transistor's bias voltages, s8y Ypp and Vqc- These bias volt- ages are determined internally as the conputer iterates to the relation. That 1s, the program assumes ¢ set of bias voltages, tolwes the equations ané then selects new and better values unti? tt converges to an adequate solution, This is al1 done internally. tn ents book it wi11 be assuned that the operating conditions con- Hist of not only the blas volta0es Vgg and Vgc, but also the ten perature T at which the analysis 1s t0 be performed, The value of Tis normally required as an input to the prograns, and it will be assuned throughout that T has been specified. the third type of information required is the set of node? araneters for each different device tn the circuit, The weaning tnd measurenent of the nodel parameters is the, subject of this sre reasentes of coe) garanters ere fy (© fen the normals active region) and rp (the total transit time in the normal, cctive region). The values of the model paraneters must Be SUP sited by the user in a manner predetermined by the program. Sone prograns are very flexible and allow sone model paraneters to be crecitied indirectly. For example, vp i normally determined via the neasurenent of fy» the unity gain bandwidth. In SLIC and SINC, the uzer hay the option of specifving either sp or the fy value at a sias condition. AS explained in detait Tater, if fy tnformationis provided, the prograns determine the value of vf internally, taking into account the effects of junction capacitances, base- width modulation and collector ohnic resistance. Note that in this particular case vp is the model oaraneter, yet fy is the in= put. Therefore, a distinction must be made between the model parameters and the program's input parameters. This distiction is illustrated in Fig. 1.1. The nodel parameters are those param- eters used in the model equations to describe the device for a given set of operating conditions. The input carameters are the data required by the program to specify the madel parameters Some or all of the input parameters may be model parameters, de- pending on the program. The formal distinction between model and inout parancters , though appearing at first to be rather pedantic, is in fact very important. It helps to maintain the proper perspective. In the above example, the fact that +. is the model parameter underscores its importance. If rp were measured directly (rather than via fy), then there would be no need to measure fy. The measurement of fy arises only in that it is a means of determining Fe Input Parameters input. informa tion required by the ” ransistor rogram. (Sone or eatdeh | Fa Traasise proaran. (Sone araneters ae Daraneters.) COMPUTER PROGRA Fig. 1.1. The distinction between input parameters and model parameters.Dependence of Models on the Physics of the Device In general, a mode? based on the physics of the device permits a more thorough understanding and fewer input and model parameters Fo* example, some programs require a table of values to describe the variation of sp with collector current. The physics-based rode) described in this book, however, requires only four model parameters to completely specify fp versus I; over the whole range As well, in the process of determining these four parameters, an understanding is obtained of the reasons for the variation of Fe with I. For these reasons, the development of the models in this book emphasizes an understanding of the physics of the device Model Notation The nonlinear models are based on the Ebers-Moll model. The original Ebers-Mol1 model) has been modified by many people to include effects such as charge storage, § variation with current, and basewidth modulation, among others. As a result, there are any interpretations of the phrase, “nodified Ebers-¥ol model." The modifications used here have three different levels of com plexity. The notation, first put forward by Pederson,(®) is as follows: EM, is the oriainal tbers-Nol1 modet.(4) it is a non- linear de model only EM, is the next level of complexity. With the EM) mode! as its basis, it provides a first-order model of the nonlinear charge-storage effects and ohmic resist~ ance. EM, is the third level of complexity, It includes such Second-order effects as baseridth modulation, @ and t; variations with current, @ better representation of the distributed collector-base junction capaci- tance and an inoroved temperature dependence. GP is the Gurme] Poon model as imolenented in the pro- gram SPICE.(3) It differs from the IntegralCharge-Control model >y Gunmel and Poon(7) mainly in the input parameters required and the absence of base push-out modeling, ith respect to the effects that fare modeled in both the EM3 and GP models, the two models are basically equivalent None of the computer programs describe the models with this termi- nology; it was chosen merely as a convenience. For example, a program may allow the user to specify sone parameters fron EM), some from EMz and some from M3. In all the Berkeley programs any parameter not specified by the user has a default value. If no parameters are specified, the resulting model is the nodel . For example, if no capacitance parameters are specified, al} capa~ citors are defaulted to zero, since the EMy model is a de model. Which Model to Use? Use the vimplest model that wilt do the job. This saves modeling effort and computer time, and the results are east understand. For example, if a transistor is current-fed, the r to external base resistance 1s not imortant and no effort should be nade to determine its value. The usefulness of the different models can be sunmarized as follows: EM) model: This model is very useful for first order dc analyses. ©, model For wost applications (especially with digital circuits) the EM model represents @ good compromise between accuracy, ease of modeling, speed and understandable resuits-(8) It is the model mst often used EM, and GP models: These models should only be used when the extra accuracy they afford 15 necessary.2.1 Introduction The structure of this section, which presents a systematic, theoretical derivation of he BJT model, is as follows Section 2.2 The EM) model Section 2.3 The EM model Section 2.4 The EM mode? Section 2.5 The GP mode! Section 2.6 Limitations of the models In the derivations of the aodels, emphasis 1s placed on an understanding of the effect being modeled and an exoianation of the paraneters required, The technioues for measuring these parameters are then given in Section 3.2.2 The EM, Model The BH model is basicaTly the simple, nonlinear wodel de- seribed by Ebers and Mot? in 1954.4) tt is a de model in that there is no characterization of charge storage in the device. The Ebers-Moll model is valid for al] regions of operation: saturation, inverse, normal and off. These four regions are defined by the bias voltage on the junctions as illustrated in Fig. 2.1 ac i TNVERSE REGION SATURATED REGION + "gp reverse bias a Forward bias oFF 3| Nomma., REGION =| ACTIVE é REGION zl Fig. 2.1. The four regions of operation of the bipolar transistor. The model parameters that fully describe the EM) model are defined and described in this sections they are Br» Spr Is+ Thon and Eg- Virtually all de and large-signal, nonlinear models are based fon the EM) model. Currently there are two popular versions of the EM, model: the injection version (Fig. 2.22) and the transport version (Fig. 2.25). These two versions, both drawn for an npn transistor, are mathematically identical. It is not inmediately -10-Fig. 2.22, Injection version of EMy model Fig. 2.2b. Transport version of EM, model. apparent why one should be preferred over the other nor why any tine or effort should be devoted to the distinction between the two. However, there are valid reasons why one (the transport ver- sion) is preferred for computer simulations.(®) These reasons only becone obvious when higher levels of model complexity are “neconsidered.* An explanation of both versions is given next, start- ing with the injection version. 2.2.1 The Injection Version The injection version is the original and better-known Eh model. Its reference currents (those currents in terms of which al] other currents are expressed) are Ip and Ip, the currents through the diodes. The reference forward diode current, Ip is given by:** 4 Woe | a. les i Ip (2.1) were Igg 15 the enitter-base saturation current, Vg 15 the base- enitter voltage, q 1s electron charge, k is Boltzmann's constant, and T is tenperature. The reference reverse diode current, Ig. 1s given by (2.2) where Igg is the collector-base saturation current, and Vgc is the base-collector voltage. The collector terminal current can now be expressed in terms of Tp and 1p: The reasons for the preference of the transport model are (i) the reference Loreents are setually ideal over sany decades of curvents, (si) the complete Specification of both reference currents for giver volteges by one fundasen- SST constant Ig, and’ (111) the ease of ceseription ef the diffusion capaci- tances. A nore éetailed cenjorison of these bro versions of the Ey nadel 1S"ghven tn appends L “tn this book, all terminal currents are defined as positive when owing {nto thw device and the veltage Yay tS detsied ay posttive when Vx 4 equal io or greater than 4, “12carte Te (2.3) where up is the large-signal forward current gain of a conon-base transistor, Similarly, the base tenninal current can be written as: Ig = (1 = ap) tp + ( la (2.4) where og is the large-signal reverse current qain of @ comon-base transistor. And lastly, the emitter terninal current becomes: Ipe-l + al 5) ‘The mathematical derivation of these equations is given in most transistor physics text books,!") but a simple intuitive feel for ‘the model can be obtained by inspection The diodes represent the transistor's base-emitter and base- collector junctions. I, is the current that would flow across the base-enitter junction for a aiven base-emitter voltage, Vg, 1f the collector region were replaced by an ohmic contact without disturd- ing the base. Igg is the saturation current of this junction. The value of Ip for a given Vgp is given by Eq. (2.1). Similarly, Eq (2.2) describes the collector-tase junction if the enitter were replaced by an ohmic contact without disturbing the base, and Ic. is the saturation current of the collector-base junction. The two diodes alone, back-to-back, do not fully reoresent the transistor. Coupling between the junctions is provided physi- cally by the very narrow base region and is modeled by the two Current-dependent current sources Consider the transistor to be biased in the nomal, active region (B-£ forward-biased and B-C reverse-biased). Then, the colector-base diode can be approximated by an open circuit, and ‘the model reduces to the ofl, current aenerator and the base- enitter diode. Ip represents the total current flowing across the base-emitter junction while ag is the fraction of that current that is collected at the base-col lector junction oeSimilarly, when the transistor is operated in its inverse ode (B-£ reverse-biased, 8-¢ formard-biased), up is the fraction of the total current that is flowing across the collector-base junction that is collected at the enitter-base junction.” From Eqs. (2.1) through (2.5), four parameters are required to describe the injection version of the EM model at one tempera ture: Iggy Iggy of and ag. The number of paraneters is reduced by one when the reciprocity property is apolied. This property 1s defined by: ap Teg * a Tes 41s (2.6) where Ig 1s cated the transistor saturation current."" Rectproc~ ity, which is experimentally observed,''2) is easily oroven for the very simple case of a one-dimensional, constant base-doping tran- sistor under low-level injection.(13) It has also been proven for the general case under low-level injection by an application of Green's theoren, (4) The constants ap and og are related to the large-signal for- ward and reverse current gains of @ conmon-enitter transistor (@ and gs respectively) by the well-known expressions: "since, for reasonable a values, the reference currents, Ip and Tg» represent the currents injected into the base region, the mane "injection verston” 1s used. "the physical interpretation of Ig 1s as follows. A pon junction saturation Current conststs of tho terms, one fron an analysts of each neutral regton. for example, for fay constant-doping, short=base lode, the Saturation cur- rent 1s given byl ADP ag , WORN Isat + for My << 1, sa Tp 5 etn cles 1s the portion of the emitter-base saturation current (igs) that arises ‘fom the analysts of the base region. Sinilarly, aqlcg 1s the portion of the Collector=base saturation current (Ics) that arises fFom the analysts of the Stae FegTon Reciprocity, €9. (2.6), then staply means that this analysts of the base region 1s the sane for both Tes and Ics and that Ts 1s this common portion of both saturation currents aefete (2.1) (2.8) Thus, only three model paranetes are needed at one temperature. Those normally used are Sey fg, and Tz. Al) the other model paran- eters (legs Iggs aps ag) Can be obtained if necessary from these three model parameters. 2.2.2 The Transport Version sort version of the EM, model d injection vereion only in the choice of the refern In the transport version the reference currents, Ig¢ and Ipcy are those flowing through the model's current sources. They reore- sent those currents that are collected.* The reference collector source current can be written as ee eT 1 (2.9) leg = ce“ Is and the reference emitter source current is: gc Ww leet Ig le = (2,10) “br transported across the base, heice the nane "transport version: <5These two reference currents can then be used to express the tran- sistor’s terminal currents: an) (2.12) [- 4] tec lee (2.13) Yithenatically, Eqs. (2.9) through (2.13) are identical to Eqs (2.1) through (2.6). Note that the dependence of the reference curcents on the junction voltages s the sane for both Igc and Ipc: That is i } +] ret leh on (2.14) where 1 1s the reference current and V the appropriate junction Voltage. Figure 2.3 shows the variation of the reference currents for both versions as a function of the appropriate junction volt~ ages. The injection version curves are shown in Fig. 2.3a, wile curves for the transport version are shown in Fig. 2.30. Note that for the transport version (Fig. 2.35), both lines are identi- cal, That is, the vartation of both reference currents with junc- tion voltage is describes by one fundamental constant, Is. For Given values of Vgp and Vacs then, both reference currents for the transport version can be completely determined if 1g 15 known. For the injection version, however, two constants (ies and Ics) are needed to obtain the reference currents, This, together with the fact that the transport reference currents are in practice linear over many decades on a semi-log plot, is the key to the preference of the transport version over the injection version, as described in Appendix 1. “16Inlg) #5. Ips YS Yoo 7 Poly) 06. yp og scale aeCTION WESTON (a) Nece fee fri oe tog scale Inllg) ¥. Y TRANSPORT ¥ERS ION (b) erence currents with junction voltages for (a) injec Fig. 23. The variation of * ‘transport of EMy model, tion version and “WeThe difference between the injection and transport versions of the EM) model only involves a change in notation, not 2 change in the form of the model. The transport version's notation will be used hereafter. 2.2.3 An Attemative Form - the Nonlinear Hybrid-r At this point, a change in the model form is wade. The change, as shown in Fig. 2.4, consists of replacing the transport model's two reference current sources with a single current source between the collector and the enitter. To do this, the equations for the diode saturation currents must be changed appropriately. As well, the equation for the single reference current source, Tgqe must be defined. The diode currents (Fig. 2.4) becone (ote ] 5 | ae ee ¥ (2.15) and: (2.16) Fig. 2.4. The nonlinear hybrid-* version of the EM, model. -18-and the generator current, Icy is given by (2.17) The model's terminal currents can now be written as te = (gg = tag) - |S (2.18) which is equivalent to Eq. (2.11), and as 1p (2.19) which is equivalent to Eq. (2.12), and as te = [HE | Cage = teed (2.20) e* 7 ae f Gee > Fee veh ts equivaent to fo, (2.13. sodel form of Fig. 2, the linearized satt-signal waive redlces to that of the vell-om Tinea sno-stanalhybrid-r todel. As shown In Fig. 2.5, the Toy current generator becoes the gq (transconductance) current generator. The forward-biased dade between the base and the enitter terminals becnes the F, resistor (r, = EE). and the reverse-biased dade between the base and the coliectel terminals becones the r, resistor, which is Vinear noted, the equivalent circtt of Fig. 2.4 ts cattes he -19Fig. 25 A line wad version of active region, Since the EM model isthe do portion of the wellknown li 2.4 for operation in the forwa ‘de model, this model hybrid- model. nonlinear hybrid-+ model." Its similarity in form to the Tinear hybrid-r model means that a computer program can perform the small Signal hybrid-# Tinearization with very Tittle effort." Note, however, that a physical "feel" for the model is decreased since the diodes in the nonlinear hybrid-+ mode! no longer represent the actual transistor pn junctions. The diode currents now actually represent conponents of the base current. However, the change in node! form does not affect the parameters required to specify the these paraneters first-level Ebers-Nol1 model at one temperature - are Still Bps Bay and Is. 2.2.4 Temperature Variation At the EM, level, sp and gp are both regarded as constants, independent of current, voltage and temperature. This assumption is maintained until the FM, level. The only parameter which is "4 second reason for using the nontinear hybrid-r model of Fig, 2.4 Ves in A, second reas the lowecurrent variation of Bye. THis wiTl be Brought cut fater in the Ey section 20assumed to change with temperature 1s 1, and is of the form:("®) nom (2.21) where T is the analysis temperacure in *Ky Taoq 18 the nominal tenperature in °K at which the device data is taken and £, is the effective energy gap in electrowolts of the semiconductor material Two more model parameters, Tyoq and Ey, are needed to account for the variation of Ig with tenpersture assumed in the conputer programs: (Renenber, the tenperature T is regarded as an operating condition ‘and not a model parameter.) In sone programs (SLIC and SINC for exanple) E, is internally fixed at a value for silicon. In other programs (SPICE for exanple) £, 1s 2 model parameter that the user can specify. 2.2.5 Summary And Additional Comments To sumarize the important aspects of the EM model = For computer simulations, the transport notation is a better choice thar the injection notation. = At a given tempersture, only three model paraneters, fi, tgs and Ten oe needed to specify the noel Tro aaditfonal pafaneters, Tyom and eq. are required to wmode! the variation of saturation’current Ig with ten- erature. “Because the EM} mode! is adc rodel, sp and fp are dc parameters, not ac paraneters.. At the EA} evel the distinction between ac and de beta values 1S purely academic. However, at the higher-level models where ggc versus I 1s'modeled, the distinction Decones Important. veglocing th foeratane deendence of dittusivity, Aa(t/". See page 37 -2e= A transformation from the transport model form (Fig 2.2b) tothe nonlinear hybrid-»mnodel form (Fig. 2.4) mmakes the computer linearization to the sna)-signal Tinear hybridn model (Fig. 2.5) simpler, The non- linear hybrid-r model is the basis for the other models. Elements will be added to it and some of its parameters will be redefined later. ANT the model diaarans given here are for an npn transistor. For @ pnp transistor, the voltage and current polarities must be changed appropriately. In most conguter programs, the mode? paraneter values are always considered to be positive, and the appropriate sign changes are implemented internally by the progran. The collector characteristics of the Eth model are shown in Fig. 2.6 as they would appear on a curve tracer, (Note that, even for this model, there is an inherent saturation voltage, Vce(sat):) Although it is very simple in form and requires, at most, five parameters, the EM, model is quite accurate. It is useful, not only for @ dc characterization of the bipolar junction transistor. ut also as an “ideal” transistor. This “ideal” transistor can be used in CAD in a variety of ways as outlined in the introduction in Section 1. Fig. 2.6. Curve tracer collector characteristics of the EM; model 22The limitations of the EM, model Tie mainly in its neglect of transistor charge storage (1.e., no diffusion or junction capaci~ tance) and ohmic resistances to the terminals. Th effects are included in a first-order mannar in the EM, siodel described in the next section. =23-2.3 The EM Model The Ez model provides 2 first-order model ing of the charge storage effects which permits the realization of Finite frequency and time responses. It also provides a more accurate de represen- tation of the device." Eight additional components are introduced, requiring twelve more model parameters to describe then. Figure 2.7 shows the progression from the Em model to the {My model for an npn transistor. The complete EM model is given in Fig. 2.7e. Three ohmic bulk resistors (rg. rg and ri)s to diffusion capacitors (Cog and Cyc) and three junction capacitors (Cyeo Cjgr and Coyg) are added to the first-level model, which is simply the two diodes and the current generator of the non-linear hybrid-» model form 2.3.1 Improved dc Characterization ‘The inclusion of three constant resistors (r2, rg and rj) improves the dc characterization. They represent the transistor's ohmic resistances from its active reaion to its collector, enitter and base terminals, respectively. These resistors are included in ‘the model as shown in Fig. 2.7b. The internal nodes of these resistances at the active region are denoted by the letters Cc’, £* ‘and 8" in the model diagram. The voltages used in describing the ‘two ideal diodes and the current source are the internal voltages. ag The effect of the collector resistance (r!) is seen in Fig. 2.8 in which collector characteristics of the FM, model (solid Lines) are compared to collector characteristics of the Oy "Ihe saproyenent to the de representatton will also obviously affect the ac and translent responses of the model otheFig 2.7. The progression from EMy to EMz model for an npn transistor. (a) EMy model (nonlinear hybrid- ¥); (b) addition of the three ohmic resistors; {e) addition of the nonlinear junction capacitors. -25-Fig, 2.7. The progression from EMy to EMz model for an npn transistor. (d) addi- tion of the nonlinear diffusion eapacitors; (e) complete EM model. 26Fig. 2.8, The effect of ron the Ic versus Vee characteristics. Dashed lines ropresnt the EM; model {r; = 0). Solid lines represent the EM model model (dashed lines). Resistance rz decreases the slope of the curves in the saturated region r low collector-emitter vol tages. In the EM, model, rj is assumed to be constant. In an actual device, however, it will be a function of collector current and base-collector voltage. Therefore, the biggest problem in obtain- ing rj is not how to measure it, but which value to use. A more detailed description of this problem is given in the measurement section on r? The emitter is the most hesvily doped region in most present- day transistors in order to produce a high emitter injection -2r-efficiency!'®) and therefore a high Ap. For this reason, the don- inant component of resistance (rg) 1s normally the contact resistance (usually on the order of 1s). rg, which is often neq- lected, can normally be assuned to have a small, constant value. Its main effect is a reduction in the voltage seen by the enitter- base junction by a factor of rg 1 In this effect on Vge» resistance FZ is equivalent to e bose resistance of (1 + fp) rj. Therefore ry affects the collector cur- vent as well as the base current, as shown in Fig. 2.9, This effect can be significant and rj can cause substantial errors in the determination of rj.(17) Resistance rg can also seriously affect the collector characteristics in the saturation region if the transistor has a low rg value. on Base resistance rj, is an inportant model parameter. Its greatest impact is normally its effect on the small-signal and transient responses. It is also one of the most difficult paron- eters to measure accurately, partly because of its strong depend- ence on operating point,(17*78) (due to crowding) and partly because of the error introduced by the sma1l, but finite value of rg-(7) In the Ey model, rj 15 assumed to be constant, The de effect of ry is seen on the In(Ic) and In{Jg) versus Yap curve ‘obtained from the EM model, as illustrated in Fig. 2.9, 2.3.2 Charge-Storage Effects Charge storage in the bipolar junction transistor is modeled by the introduction of three types of capacitors: two nonlinear Junction capacitors, tho nonlinear diffusion capacitors and a con- stant substrate capacitor. -28-Fig. 29. The effect of ri, and ce ov the Inflc) and tnllg) verwus Vag characteristics Br the Emig medel. (Note that the EMz model neglects other high-level ttfects which are treated in higher-order models.) a) Junction capacitors The two junction capacitors (sometimes called transition capa- citors) mode? the incremental fixed charges stored in the transis~ tor's space-charge layers for increnental changes in the associated junction voltages. These capecitors, denoted by Cj_ for the base- emitter junction and Cjg for the base-collector junction are 29included in the model as shown in Fig. 2.7c. Each junction capac- itance is @ nonlinear function of the voltage across the junction with which it is associated The normal, sinple analysis of Cig and Cj, makes the depletion approximation (which assumes that at the junction of interest the space-charge layer is depleted of carriers). Then, for a step (or abrupt) junction and for a linear (or graded) junction, the varia~ ‘tion of the emitter junction capacitance with base-emitter junction voltage can be written for an npn transistor*('9) as Seo Sie Maree) * Wei (2,22) my Where Cieq 18 the value of the enitter-base junction capacitance at Vgig = Oy tg 15 the emitter-base barrier potential, and me is the emnitter-base capacitance gradient factor. Likewise, the varia~ tion of the collector junction capacitance with the base-collector Junction voltage is given for an npn transistor by: — (2.23) or tc where Cyoy 45 the value of the collector-base junction capacitance at Vgicr * Os oe {8 the collector-base barrier potential, and me 1s, the collector-base capacitance gradient factor. For a step junction, m+ 0.8; for a Vinear junction m= 0.333. Since most practical junctions ie between a step and a linear unction, the above equations are assumed to be general and to apply for 11 junctions with a gradfent factor between 0.33% and 0.5. This is therefore an empirical fit Sie Mores For a pnp transistor. since Vqy 1s positive when Vy 2 Yys the minus signs in Eqs. (2.22) and (2.23) becone"blus sians. ” -30The parameters 4 and ¢ are called the transistor's built-in borrier potentials for the emitier-base and collector-base junc- tions, respectively. There can be sone confusion about these builtin barrier potentials (which are typically of the order of 0.7 V from capacitance-voltage data). It may appear at first that Since typical value of Ype at mA current levels is 0.7 V to 0.8 V, the junction applied voltage can be greater than the built in barrier potential. This does not follow, for two reasons Fiest, the externally-applied base-ewitter voltage, Yar should aot be confused with the internal, junction applied voTtaaey Ypsgs- A high current levels, the finite values of ri and r in practical transistors can cause Vpigs to be significantly 1e55 than Vag Second, there are actually tuo built-in barrier sotentials:* the well -krova barrier potential which sets up the drift component of current to oppose te ds ¢fusion component at the junction 2? (otvich equats KT tn(" 3) for the simple, constant-doping, abrupt- Junction case and is {picelly of the order of ! 1 ang the bareier potential used in the above capacitance formula?!) (wich 1s typi cally of the order of 0.7 ¥). It is the former barrier potential that 1s greater than Vpigi- Because of this condition the former barrier potential does not appear in any of the equations. The user, however, need not concern hinself with the details of this distinction between he two types ‘ier potentials. In the model described, ¢ is used only to compute the junction capacitance and therefore the value obtained for y from C-V data is (by defini- tion) the appropriate value to use.** The above functional dependencies of C,¢ and Cy¢ (Eqs. (2.22) and (2.23) are built into the EMy model. To specify the junction Taito" theranator’s Geary one Eye (TP MMEHINe he trrter eter “*hetually, ¢ may also be used in the My and GP models to describe the vari- elon ZF olsentetn wth fessersure snd non convent evelse ot rorsaly IGE Guay sca for cout inn Cy 31capacitances completely, three parameters (Cjqy ¢ and m) are required for each. Although most experimental data can be forced to fit Eqs. (2.22) and (2.23), the reduction of the measured capacy tance as a function of voltage into these three paraneters is not trivial ff any stray capacitances are present. Techniques for reducing the data are presented in the measurement section. Some programs allow the user to specify each junction capacitance at a value of its associated junction voltage, instead of requiring the Cj value Figure 2.10 shows three plots of the variation of the junction capacitance as a function of voltage. The dashed Tine represents Eqs. (2.22) and (2.23). Under forward bias these equations predict infinite capacitance when the internal junction voltage equals the built-in voltage. Chawia and Gunnet'2) nave shown that under forward bias the depletion approximation is no Tonger valid and that Eqs. (2.22) and (2.23) no longer apply. The solid Tine in Fig. 2.10 shows the (non-infinite) variation of the junction capac tance obtained by Chawla and Gurmel. An expression requiring four paraneters has been fitted to this curve by Poon and Gumet (23) The third curve in Fig. 2.10 represents the straight Tine approximation made by the Berkeley programs SLIC and SINC for V > 9/2." The equation for this straight line, obtained by match- ing slopes at ¢/2, is given by: cy (« 2 +): Big [= ad -s} (2.24) where the junction subscripts have been omitted and V must be greater than or equal to ¢/2. This approximation, while avoiding the infinite capacitance, is not as accurate as the Chawla-Gunmel curve, However, it is acceptable because under forvard bias the diffusion capacitances, described next, are dominant and inherently “ne other Borketey program, SPICE, ves sina straight Tine approximation, But shout v= OLE "Gnat iss Cgliad) = jg) # mH 232: (ah Cy. (2,22) + | oreo D siwie theory) \ chown % +N eee 0 result, |; | \ <\ Berkeley srograns (SEC 8 SINC) sinter — my Reverse bias Forward bias Fig. 2.10. A plot of three equations desc with voltage. ing the variation of junction capacitance include the effect of the mobile charges in the space-charge layers. Besides needing one less parameter than the Poon-Gunme? equation, the Berkeley model 1s accurate in the reverse-bias region where it is most importent. b) Diffusion capacitors The diffusion capacitances model the charge associated with the mobile carriers in the transistor. This charge is divided into ‘two component: one associated vith the reference collector source current (Igg) and the other wit the reference emitter source cur- vent (Igc]+ Each component is ~epresented by a capacitor To evaluate the diffusion capacitance associated with Igc, the total mobile charge associated vith Icc must be considered. There- fore, the base-enitter junction is assuned to be forward-biased and Vgc = 0- Figure 2.11a shows the minority-carrier concentrationsfor the simplified one-dimensional case of constant base doping, neg) igible base recombination, and low-level injection.” The total mobile charge associated with Teg (Ogg) can be written as the sum of the individual minority charges: Gt +0; 4% (2.25) 0 where Q; 15 the mobile minority charge stored in the neutral enit- ter region, Q, is the mobile minority charge in the emitter-base space-charge region associated with Ice (normally considered to be zero),"* Qy is the minority mobile charge stored in the neutral base region and Qy {5 the mobile minority charge in the collector tase space-charge region associated with Icc. Because of charge neutrality there will be identical majority charges stored in the neutral regions. However, to determine diffusion capacitance only one (minority or majority) needs to be considered. Charge Q, is normally considered to be zero, so that the total mobile charge associated with Igg can be expressed as +0) +05 +0 = Toe "eB, Hee * "e Fee * ToBge, ‘ec = Gy tg epee) Foc Qe 8 te Nee (2.26) where 1; 13 the enitter detay,"**(25) tggccy fs the enitter-base space-charge layer transit time, tg is the base transit time “Tne analysis 16 not restricted to this siaplified case. This case was chosen for ease of presentation only “For high-frequency devices, recent computer studies (by P. Mertens, University Gf Leuren, Belgiun) have shun that dp acy not be negligible. TiS pas also been concluded Uy Kerr and Ber? "The enitter delay, "1, $s normally given the syubol cg. However, €) is used here in order to avoid any confusion Tater in the GP node! “ue(o) Fig, 2.11. The mobile minority charge associated with (a) lec and (b) lec (not to scale) 2 ( for the constant-coping case) (9) Sepeey is the base- Yessey _\l27 conectar spaces ayer transit tine (+ ABS) cat Vim)” tg 15 the total forvard transit tine (here assumed to be @ con- stant) and 0, and y, scat. Vim 2° the diffusivity and scatter- Limited velocity for the base minority carriers, respectively.” “Wote that £q. (2.26) is 2 large-signal, or dc, definition of op ‘he normal small-signal, or ac, definition is Ae Fac 8 Mee For the Eg model where sp is assuned constant. Fac Tae F Similarly for +4 and £4. (2.28), In the Mg and GP nadels, where ty 1s not constant the dlstinetion between = Tye ahd gy ts necessary. 35A similar analysis of the total mobile charge assoctated with Teg (see Fig. 2.110) gives: Aye + Og #06 +8) #0 (2.27) whee Q5 is the mobile minority charge stored in the neutral col lector region, Qs is the mobile minority charge in the collector~ base space-charge region associated with Ipc, Q, 1s the minority mobile charge stored in the neutral base region, and Q, is the mobile minority charge in the enitter-base space-charge region associated with Igc. If charge Qg is assumed to be zero, then: doe = thee t ton, Tee * antec * 0,0, nc * ‘elec * ScB5e, Tec * "eRIEC * 'EBsq, TEC = bgt tgp * tee) | c* Ter * TEBe,) Tec Ope te Nec (2.28) were 1g 18 the collector delay, tgp 15 the reverse base transit tine, and tg is the total reverse transit tine (also assuned to be constant here) For the saturated mode (i.e. Vge and Vgc both forvard-biased), Both Upg and ge are assumed to occur independently and the total iminority charge stored in the transistor 1s the sun of components Q} through Qg. The two charges, Qe and Qc» are modeled by two nonlinear capacitors (yp and Cog respectively (see Fig. 2.74) given by: (2.29) (2.30) -36-For small-signal analyses, Cpe is Tinearized to:* “oe, a “9 (2.31) ‘small signal mr here Qgp 1s the transistor's forward transconductance 4 a i ce ( Sar 2 age | y a (2.32) Bc Similarly, for smal1-signal analyses, capacitance Coo is linearized to: . 84 | enait signar Verc = Soe OR [ere = 0 where ggg is the transistor's reverse transconductance: a, | a ee _ Tec Sok & Wg ece oT (2.34) ‘The position of these two diffusion capacitors in the model can be Justified by considering the voltages that influence the charges = Vgsgs fOr pg and Vgies FOr Oye. In present-day transistors, the contribution of the base- region terms Q3 and Q, 15 not as significant as it used to be For example, in the total forward transit tine rp, emitter delay ty can be as great as or greater than base transit tine xs(2) while in the total reverse trarsit time 1g, collector delay t¢ is invariably the dominant component “This Tineartzetion is performed inside the computer program so the user $s only required to specify tf. By specifying {y the user mudels Qye in either 3 non~ Tinear mode (for transient atlyses) ar a'Tincar node (for ac analyses) since TPs assumed to be constant G28. tye } -37-Input parameters Although x, and 1g are the only extra model parameters needed to describe Cpe and Cyc sone programs enable the user to specify others, which are more easily measured. For example, the total forward transit time 1, can be speci- fied in terms of the transistor's unity gain bandwidth, fy, at a given collector current and collector-emitter voltage. From the operating point at which f, is measured, the program first calcu- Yates ae and Vgc» then Cp and Cie. tp can then be determined from: room fey ey. vette Hy |e Sie ( ae nl (2.38) This formita 18 mdified In Ue Ey model to take basevith wodu- but only a meas to obtain. defined as the saturation tine constant and is related to the searationéetay Ces) a5 sh Inthe ester ston. og a {a obtained from sgqr BY (2.36) €) Substra capacitor ‘The substrate capacitance (Cog or Cgyg) can be important in integrated circuits. Although it is actually a junctfon capacitance in the way it varies with the epitaxial layer-substrate potential, it 1s modeled here as a constant-value capacitor. This representa~ tion is adequate for most cases since the epitaxial layer-substrate junction is reverse-biased for isolation purposes. (To include the 38variation of Coup with epitaxial Tayer-substrate voltage, a sepa- vate diode or transistor can be included in the circuit sche~ (30) 45 described in the measurenent description for ¢, in Section 3.) The placement of Coy in the madel in Fig. 2.7e 1s shown for an npn transistor. However, for a pnp transistor, Coup may not be connected to the collector. Irstead, for a lateral pnp device, Ceyg 1S connected between the tase and the substrates while for a substrate pnp device, Coyg 4s set to zero since it is already nod- eled in the Cie capacitance. (In SLIC these different connect ions are automatically made, whereas in the other Berkeley programs, ¢ matic, SUB SUB fs connected to the collector for both npn and pnp transistors.) 2.3.3. Small-Signal (Linearized) EM , Model A Vinearized version of Fig. 2.78 15 used for ac linear analy- ses. The elements for this linear hybrid-r model, shown in Fig. 2.12, are given by: ne (2.37) (2.38) Cy Spe te + Cye(¥gegs) (2.39) C= Gen Tat SjelMpuce) (2.40) In the normal region of overation, reverse transconductance Sp 15 e8sentially zero, so thit resistance r, can be regarded as infinite and capacitance C, = ¢,¢(Vgics)- The resultant model is the well-known Vinear hybrid-s model (5) -39-of Fig. 2.76 Fig. 2.12, The linearized version of the EM2 mod 2.3.4 Summary ‘The complete EM, model requires eight extra components (three constant resistances, four nonlinear capacitors, and one constant capacitor) to be added to the EM model. The five capacitors provide a first-order model of the charge storage in the transis~ tor. As well, the three resistors give an improved de representa- tion over that provided by the EM) mode? To characterize these eight components. a total of 12 extra node! parameters are required. These are: The three resistors are described directly: ry The (assumed constant) collector ohmic resistance 40ry The (assumed constant) emitter olic resistance rh The (assuned constant) base omic resistance For the enitter junction capacitors Ci paraneters are needed Cyeq The enitter-base Junction capacitance at Vor = 0 » three more mode) (or Cig at 2 given Yop) te. The emitter-base barrier potential ig The enitter-base capacitance gradient factor For the collector junction capacitor, Cyc, a sitilar set of three mode) paraneters is needed joy The colTector-base junction capacitance at Vgc * 0 (or Cjg at a stven Vgc} tg. The collector-base barrier potentiel nig The colTector-base capacitance gradient factor for the emitter diffusion capacitors Cye+ only one node! paraneter is required tg The (assuned constant) total forwerd transit time (or the unity-gain bandwidth, fy, at a given Tp and Vee» fom which 1 is obtained). for the collector diffusion capacitor, Cycs one additional model paraneter is required: tq. The (assuned constant) total reverse transit tine (or the saturation time constant, xp is obtained). at? from which Lastly, the substrate capacitor, Coygs is specified directly: Coyg The (assumed constant) capacitance between the substrate and the collector for npn transistors or between the substrate and the base for lateral pnp transistors. oa.In computer programs where the default model is the EM, model. parameters 12. Ths Mh» Cyeqy Cjcor Tee TRE aNd Coyp alt default to zero, if not otherwise specified. The default values for ses ¢¢+ mig and me need not be zero. It should be enphasized that the EM, model is adequate for the majority of cases, especially for analyzing digital circuits. Wow ever, there are still sone limitations, including the absence of such de effects as basewidth modulation and the variation of @ with current level. Both of these second-order effects are accounted for in the EM and GP models which are described in the next. two sections. The collector-base junction capacitance is actually distrib- uted across the base resistance. The use of a single capacitor in the &M, model is a first-order representation of this distributed capacitor. An improved, second-order model is given next in the EM, model. Other (second-order) improvements in the EN mode? are a nore complete treatnent of the effects of tenperature and the variation of x, with collector current.2.4 The EM3 Model The EM model is the third level of complexity in the non- Vinear modeling of the bipolar junction transistor. While the EM is a simple de model, and the EM model contains a first-order representation of charge-storage effects and ohmic resistances, ‘the Ey model is concerned with second-order improvements in the de aspects of the EM, model, charge-storage model ing and tenpera- ‘ture performance. The EM model adds the following features = base-width modulation and variation of « with current and voltage. the ability to split the collector-base junction capacitance across rj, + the rise of if at Fgh currents variation of device parameters with temperature. These effects are mainly incorporated by modifying existing equations. Added conponents are two diodes and a junction capacitor. Three extra model parameters ore required for the vari~ ation of forward current gain with collector current (ip ¥S Ics three for the variation of inverse current gain with enitter cur- rent (Bq VS Ig). one for basewidth modulation, one for the junc= tion capacitance split, two for the variation of xp with current level and six for the temperature variation The de improvenents described in this section are basically wodeled equivalently in the GP sodel (which is the subject of the next section). In this [My section, enphasis is placed mainly on a description of the effect be:ng modeled and a brief justifice- tion for the method of incorporating it into the wodel, In the GP sodel, 9 nore detailed derivation of the underiying physics is Is essentially available in the computer (2) given. The EM; model prograns Stic") and sinc.(2) Although its de treatment is not quite as complete as that of the Gumel-Poon model, its input parameters are easier to understand and determine, -43-2.4.1 An Improved de Model at a Given Temperature The de model is inproved by incorporating two second-order effects: basewidth modulation and the variation of § with operating current 2} Basewidth modulation Basewidth nodulation (the so-called “arly Effect")(22) describes the change in basewidth that results from a change in the collector-base junction voltage. In the normal, active region the emitter-base junction is forward biased and the collector-base junction is reverse biased. The width of the space-charge layer of a pen junction is @ strong function of the applied potential Large variations in Vgc» for example, may cause the collector-base space-charge layer to vary significantly. This, in turn, changes the normally thin basewidth The total effect of basewidth modulation on the device char- acteristics in the normal, active region is a modification (as @ function of Vgc) oF = Ig (and thereby the collector current) oe oe These three mode] parameters are affected because of their strong dependence on the basewidth, W. Only one extra parameter, the Early voltage, Va, is used to tmodel basewidth modulation in the forward, active region. Figure 2.13 shows the effect of basewidth modulation on the variation of collector current (Ig) with collector-emitter voltage (Ip ¥5 Veg) =+ a non-zero slope in the normal, active region. (The dashed Vines iMlustrate the zero slope obtained with the EMz model, where basewidth modulation is not included.) AteFig. 2.13, The effect of basowidth modelation on the I versus Voe. characteristics ‘The dashed lines represent the characteristics obtained with the EM2, model. (Not to scale.) The dependence of the above three wodel parameters on Vac (vi the basewidth modulation phenoneron) has been derived dy necattal3) and is given in Appendix 2. The analysis, which assunes that the transistor 18 operated in the linear region, first determines the effect of basewidth modulation or the basewidth and then on the three basewidth-related paraneters.* The results of the analysis are \ Wag) = (0) ( ' #) (2m) ph more detailed consideration and justification of the assumptions made in the alysis ia given in Aypendices 2 and 4 25(2,42) (2.43) (2.44) where Vp is defined, for an npn transistor,” as al 1 aw a slatoy > aft (2.45) BC | Mac = 0 Equation (2.41) describes the (assumed linear) variation of the basewidth with Var. Equations (2.42), (2.43) and (2.44), which sive the variation of the three noel paraneters with Vggs Follow directly from &q. (2.41) and the assumptions that the constant- base doping relationsnips Ty = 4s} gp « 18 and xy 228) are aso approxinately valid in generat An expression similar to Eq. (2.44) can be obtained for tgq. but since tgp is normally only 2 very sell component of tga ita not been inelued. The second forms of Eqs. (2.42) and (2.43) are preferred couputationally since the First Foras becone infinite at Voy = Yq (Weg = Yq may not necessarily be the correct value but could be a temporary value while the computer 1s iterating to the sotutton.) The second form assumes that |Vgcl << Vq- Since Vp is typically on the order of 50 V and basewidth nodulation is itself a second= and tp order effect. the analysis can give acceptable accuracy even when the value of Vgc is close to that of Vp. For a enn teansiste a #ta.s sige stould be introduced OF Vue changed £0 Veg. -46-Vq has no physical counterpart in the circuit model; only a nnathenatical effect whereby existing equations are modified. (This process of altering equations or paraneters without altering the form of the equivalent circuit will be observed for other effects jn the EMy and GP models.) With the exception of 1g, which is described later, the total effect of basewidth nodulation is accounted for if Ig and 6 are modified as in Eqs. (2.42) and (2.43). The expressions for Icy and Ig (Eqs. (2.17) and (2.19) respectively) then becone: in the first term of Eq. (2.47) the sinilar dependence of I, and Bf on Vac (since both are assuned to be « f) results in @ cancellation. Therefore, in the normal, active region where the second term ts negligible, 1g 1s independent of Vgc. The Vgc variation iM gp is achieved by Keeping Tg constant and modifying Te. This ‘trod ‘Very Vaportant Conceot which will be Wlustrated again later The correct variation of 8; (with Vgc as here, or Ip, as later) is ‘not obtained by varying é, but by modeling correctly the expressions for Ig and/or Ip. A sinilar analysis for basewidth modulation when the device is operated in the inverse mode is included in the GP model. The Early voltage Vp can be obtained directly fron the Ip vs Veg characteristics. This can be shown matheratically. The slope of these characteristics in the normal, active region, doy is ob- tained from Eq, (2.46) by first dropping the (negligible) second term and then differentiating with respect to Vee (Vge assumed a7constant). The result, derived in Appendix 2, is: 1,(0) os (2.48) gtle 90° ACE | gp = const. YR (in SLIC and SINC, rg(=1/g,) and 1(0) are acceptable input param eters, from which Vp is obtained by Eq. (2.48).) The geometrical interpretation of Eq. (2.48) shows that Yq 1s obtained from the intercept of the extrapolated slope on the Vee axis (as shown for the Vag, curve in Fig. 2.13). For example, a slope of (50 ka)" at Ig(0) = 1 mA gives, from Eq. (2.48), vg = 50¥. Amore detailed description of the geonetrical inter pretation is given in Appendix 2. bY B ge Variation with current In general there are three regions of interest in the varia- tion of 4c with current. Figure 2.14 shows a typical variation of fy with Ic. Region I is the Tow-current region in which og ‘increases with Ic, Region II is the mid-current region in which bp 15 constant (i8py)- Region ILI is the high-current region in wivich s, drops as the current is increased. Before analysing these regions, several points should be noted about Fig. 2.18. Variation with Vgc. The curve in Fig. 2.14 {5 drawn for con- stant Vgcr im this case gg = 0. The variation of fp with Vac has just been covered, In the following it will be assused that all data corresponds to Vgc = 0. Non-Zero Vgc data can be reduced appropriately by the application of Eq. (2.43). Bp Variation. For simplicity. most of the following analysis considers only the variation of f with I. A similar analysts can be performed for the variation of gq with Ip at constant Vgc. Results for gg variations are given where appropriate. otaToa seal Fig. 2.14, Typical variation of Bi with Ig illustrating the three regions of interest. (Not to scale.) Usefulness of Fig. 2.14. It will be shown that Region I ts governed by additional components of Tg, while Region III results from a change in Ic. This infomation is not evident from Fig. 2.14, Therefore, an alternative form of presenting the above infor- nation 1s used: one in which more information and a clearer under- standing of the device operation 1s obtained. This alternate form is a plot of In(Ic) and In(Ig) a5 a function of Yg.gss a8 shown in Fig. 2.15.% Because of the logarithmic nature of the vertical axis, Bp 15 obtained directly from the plot as the distance between the Ig and tg curves. Wot only 1s it evident from this plot what causes the variation of #; with Ic, but all the mode? paraneters needed to characterize this variation can be obtained directly from it. One of the wost important concepts in the modeling of, “acceptabte steratives to ying te gated Tor thin ito use, Tong or ore Simply to use semiclog graph pager. However, the slopes of the Tines described fere will assume graphs of In(i) v5 -49-transistors is this preference of the data in the form of Fig. 2.155 over that of Fig. 2.14. Oata in the form of Fig. 2.15 will be used through the following analysis. It is stiTl assumed that Vgc = 0 for all points. Existence of Region II. For some transistors there may not appear to be a region in which sp is constant, For these transis- tors Regions I and IIT have simply overlapped. The analysis and subdivision into the three regions is sti11 valid, since the wode! paraneters can stil] be obtained from Fig, 2.15 even when Region 11 in the tg vs Ip curve does not exist. Ohmic resistances. The analysis of the variation of ap with 1g now proceeds, by region. In this analysis, it will first be assumed that the ohmic resistances rj, r{ and rj are all zero such that Vpp = Vgigs and Vgc = Vgices The effects of this assumption will be examined at the end of this analysis. i) Region i: mid-currents In this region, the EM, model holds; the fp used in the EM) model applies only to Region II and is now called Bey. The two currents in this region (for Vg¢ = 0) are given by: I wor (2.49) ° 1,(0) eT (2.50) From Fig. 2.15, values of 8y(0) and 1,(0) can be obtained directly, as explained in the measurenent section. ji) Region |: tow currents The drop in 8, at low currents fs caused by extra components of that until now have been ignored. For the normal, active region -50-yn REGION TTL Fig. 2.15. A plot of In(tc) and ta(lg) as a function of Vgrg’ which gives a clearer ‘understanding of Regions, Il, and II with Vgc * 0 there are three extra components which are caused pyl38 the recombination of carriers at the surface, the reconbination of carriers in the emitter-base space-charge layer, and = the formation of enitter-base surface channels AIL three components have a similar variation with base~ enitter voltage, Vag: (26)8.) Ta(surface) ~ 's(surface) \e #T - 1 (2.51) To(es sei) * s(ee seh (2.52) (4) KT Ty(channet) * !s(channet) ‘© = (2.83) ‘hese three components should each be added to the base current, tq. (2.50). Fortunately, a simplification can be made. A conpo- site current can be made of a1] three extra components which has the form: Tg(surface) * e(e8 set) * 'a(channel) ( ste \ mek 1 Te (compos ite) * I5(composite) (2.54) were ng, is called "the los-current, forward region emission <2 Fficient” and Ties between | and 4.* For most cases, a fit to Eq. (2.54) can be made with reason- Therefore, at Vgc * 0, the base current is able accuracy approximated by 1,(0) 8 Bra o wan ar1ce ©) 9, Ge enttod HE. The change in notation has been made Neve to camasize its bpticabitty to low currents only. Bi Since channeling and surface reconbination can both be made small with care: ful processing, the dominant component is normally the recombination in the emitter-base space charge layer and nay 1s nomally close to 2. 52where the term I¢(coposite) 19 £9- (2-54) has been replaced by Colg(0) (1.e., it has simply been normalized to 15(0)). The two additional model parameters are Cp and ny Inverse region model at low currents. hen the base-col lector junction is forward biased, there will generally be three similar additional conponents of Iy at low current levels -- surface recom bination, collector-base space-charge layer recombi tion and collector-base channeling. In a similar way they can be lumped together into a composite component that depends on Vac. The expression for Ig in general then becoues | : ) seagior\e )eeuaile where the two extra model paraneters ng, (the Tow-current, inverse~ region emission coefficient) and C, nave been introduced.” (2.56) Effect on equivalent citcuit. The additional components of base current Ig are included in the circuit model by means of two non- ideal diodes, as shown in Fig. 2.16. The circuit to which these diodes have been added is the EM, model, in the nonlinear hybrid-r form.** The EM, model was used only for the sake of simplicity 5 For ney» the SPICE notation has been altered from NC to ney "the simple addition of the two non-idea) diodes paints out the second advan- tage of the nonlinear hydrides fore. (Ihe First is its similarity to the fear hybrid: model.) The two ideal diodes had previously decn ident i= Hed'as'the two ideal conponents or tase current. “ihe extra cosponents of Ig mere therefore simply added by 4 eatural extension of the model. This cin'be carried further, For those rave cases where. the composite component OF Eq. (2.54) 12 not accurate enough, the equivalent circult ts very simp. extended by adding more non-ideal diodes. for each of the extra components Of Tg. A nonaideal diede 1s defined here to be one vith a non-unity emission coefficient -53-[oe cgtgton\ ot Fig. 2.16. The inclusion of two diodes that model the extra components of Ig into the EM) model (nonlinear hybrid» form), Jn iVlustration: the capacitors of the fy model have ne effect in this de analysis and the effect of the ohmic resistances rps rZ and tare considered tater, ATT these components canbe added into the titel, as was done before, The plot of Tn(Ig) vs BE for Ugg * 0, shown in Fig. 2.17 plot of In(lg) ve “PSE for Wg * 0 9. 2.17, Siiustrates the two conponents of base current Tp: the ideo? com ponent with the slope of 1 and the non-ideal conponent with the The extrapolation of these straight-line components ie \ i,(0) to the Tine defined by Vag = 0 gives the values of C)I(0) and 5 Be als Fey) slope of 5 A similar plot of In(I,) as @ function of Vgc for inverse operation yields values for the model parameters C, and ne. A typical value 3 2 for Cy (and Cy) 15 10%, and a typical value for ng, (and no) is 2.iii) Region MI: high currents” At high injection levels, the injection of minority carriers into the base region is significant with respect to the majority carrier concentration. Since space-charge neutrality is maintained in the base, the total majority carrier concentration is increased by the same amount as the total ninority carrier concentration. The effect of the excess majority carriers on the collector current has been calculated by Webster'3”) iho showed that at nigh levels the collector current asymptotes to Woe “ZT Te(nigh tever) "© (2.87) Equation (2.57) is incorporated into the My model by modifying the collector current expression for Vgg * Os Eq. (2.49), toll!) - ) (2.58) where @ is the additional model parameter. Equation (2.58) has the appropriate asymptotes (Eq. (2.49) at Yow currents and Eq. (2.57) at high currents).** ‘There are two causes for the érop_in ef at high currents: saturation and the effect of high-level injection. Since saturation 1s modefed separately iy +; {and the inherent saturation oi the ith sodel), this section concentrates only on the latter cause, It is Uierefere assuned that all data 15 obtained in the formal, active region. “*yote that Eq. (2.58) 150 provices @ matinmstical “definition” of high-level Injection. that’ is, when pe oy -55-ugg TER apHTALION Of the Migh-current asyeptate of Tn{le) ve MBE with Vgc © 0 (Fig. 2.17) to Vge * 0 gives the value 15(0)/e The nigh-current asymptote has a slope half that of the low-current curve. The intersection of the two collector current asynptotes defines the point (Ix, Vy)» This point will be used in the GP ode Fig, 2.17. The complete plot of Inflc) and inilg) versus VBE for Vac = 0. Note that here 1 and ‘are assumed to be zero. -56-iv) The full picture: regions I, Ml and Il The M3 model parameters required to describe completely the variation of bp with Ip at Yge + 0 are: Pew Region Il c @ Region 1 MeL ° Region 111 The values of these four #, v5 I, model parameters can all be obtained directly from Fig. 2.17. A Similar set (spe Sqr Moy» %R) can be defined from the variation of fg with Ir at Vay v) Effect of ohmic resistances So far, the ohmic resistances ri, rj and ry have been neg- lected. To first order, they do not affect the above analysis but they do affect the experinental data, The voltages in the above analyses should be the internal voltages Vgigs and Vpics- The externally measured value of Vp¢ should be corrected to Yyics by subtracting (Ig ri + Ip rj)s ncrmally this has a small effect since Vgp onty influences the tasewidth wodulation correction fac- tor. The esfect of rj and rj is an increase in Ypes which can be seen from Fig. 2.38. To correct for the effects of rj and rg, it is assumed that the slope of In(lg) v5 Vpug. stays constant (the dashed Vine). Therefore, the value of (Ip rj + Ip Fg) can be found from the distance marked A. The appropriate anount (wrarked 8) is subtracted from the In(ig) curve and the resultant curve (the dashed Ig Tine) becomes that of Fig. 2.17 where the horizontal axis 15 now Vg igs This technique Is explained in detail in the Measure ment Section 3.4 (p. 196). ‘One additional complicating factor aust be mentioned. The above analysis assuned constant junction tenperature (T). At high current levels, care must be taken to ensure that excessive power dissipation does not heat the junction. To this end, high-current “57Fig. 2.18, The effect of rip and fe on the lallg) and In(tg) versus Vpe character- istics measurements should be made in a pulsed mode and/or with Vgc as close to zero as possible (without entering saturation). vil ip versus Ig input parameters in SLIC and sinc>Y” In SLIC and SINC, the input parameters needed to describe the variation of g, with Ip are different from the above four model paraneters (Bpys Coy neys 0): the input parameters are designed to be more useful for the design engineer who deals with gp vs Ip rather than In(Jc) and In(Ig) vs Vge- To explain the input paran- eters it is first necessary to derive the equation for BF vs Ic. 58The equation for sf vs Ig at Vyp = 0 is built up from a considera- tion of each region, For convenience, tj’ will be found. Wee fat In Regions I and 11, g;! +s given by: 1M) (2.59) Basle (2.60) The combination of Eqs. (2-59) and (2.60) gives, for all regions. WED s aste (2.61) 59where (2.62) (2.63) or (2.64) ae Prats ° Equation (2.61) assumes no interaction between Regions 1 and IIT. From Eq. (2.61), the four input parameters could be a), ap, 23 and Ney- However, one further transformation is made to the input parameters The input parameters required are: Sruay> Towaxs BELOW Teyoye BCEC and Voge. Spygy 18 the maximum value of Bps Toypy 1 the collector current at which Bpyay OCCUrS, Spi gy 1S any value of fg at a collector current Tess than Ioyay> Icuoy 15 the collector current at which Bpygy 18 measured, BCEC equals 1/mg, and Vog is the value at which all this data is obtained.” Basewidth modulation effects are first renoved by multiplying Nec tne first fur input parameters by [1+(#8)]. the, te fort i a) ing equations are solved for 9), ap and ay. 1 . (ocee-1) , Sana 791 * 82 Fowax aston (2.65) 2 . (ecee-1) , fuow = 81 * 2 Tecow asteuow (2.66) 1 (acec-2) _ °3 ( Tenax 3 [p - ee (2.67) where Eq. (2.67) is obtained by differentiating Eq. (2.61). "im SINC and some versions of SLIC, QCEC 1s Internally fized at 0.5. -60-jon with current Vid) Begg and its va pac 18 defined as (2.68) From Eq. (2.61), a ie) 1 eh seas te (2.68) erie a Fac 91 * gy Therefore, a knowledge of the &;4, vs I¢ model parameters also gives a complete description of Bp,¢ VS Ie. 2.4.2 An Improved Charge-Storage Model at a Given Temperature The improvement in the charge-storage model is achieved by modeling two second-order effects: the ability to split the collector-base junction capacitance across rj, and the variation of xg with collector current a) Improved Cig - 1, Model Ideally, the collector-base junction capacitor should be modeled as a distributed capacitor across rj as shown in Fig. 2.192. A First-order model of this distributed capacitor is used in the -61-EH, model in which a11 of Cie Ties on the inside of rf. An im proved representation is available in the EM, model in that Cyc can be split up on either side of Thy as shown in Fig. 2.190. An extra model parameter, RATIO, is needed. RATIO Ties between zero and unity. Se Rest of in Mosel Tr Fig. 2.198, Diagram of Cj¢ distributed across ry. £ PATIO Copies Lear gtipee test of 8 an B] moaet vr Fig. 2.19b. The EMg model of Cic distributed across i 6b)_ Variation of T, with current The variation of 1, with Vgc has already been covered in Sec~ tion 2.4.1 on basewiéth modulation (Eq. (2.44)). 1 also varies with collector current There are three basic causes for the increase in ig with Ip at high currents: a reduction of tre low-level aiding-field effect in drift transistors at high-level injection,(25) an effective bas widening described by Kirk(29) and a two-dimensional spreading effect described by van der Ziel and Agouridis. (4°) There is stint some controversy a5 to which of the base-widening effects domi- (41,42,43) rate. The true cause for the increase in x4 at high cur= rents probably lies in a combination of these effects ranging all the way from mostly one effect in one device to mostly the other effect in another device There are several techniques for modeling if vs Ip. Gunel and Poon'?) use a multiplier 8 which requires 4 model parameters. Jn SLIC and SINC, it is assumed that the two-dimensional effect dominates. In the normal, active region, this results in the equa- tion: 1)" where rp,(0) is the low current value of the forward transit time (previously just called rp). Lz fs the smallest width of the enit- ter, Wis the basewidth, and pg is the current at which rp starts to rise. The variation of tr,, with Ig as described by Eq. (2.70) is sketched in Fig. 2.20." Two extra model parameters are “Equation (2.70) actuslly applies to the tp component of tp. It 1s assumed that ip can also be Fitted to this expression. "in the nomal, active region, Tec = Tc -63-(0 Fig. 2.20, The variation of fg with Ic as described by Eq, (2.70). ) curve-fitting the experimental variation of tra, with Ig to Eq. (2.70) at high currents. The rise in tra. at high currents is observed as a drop in f, at high currents. The variation of %f,¢ with current is caused by an effective increase in the basewidth. This will also affect the de character- istics.” To take this variation into account, an appropriate change can de made in and é in the EM, model (which are proportional to Fig and “iggs respectively) “Since sp {5 no Yonger equal to tp it 1s given in this case bys 4 41 for Ieee Neg2.4.3. An Improved Variation with Operating Temperature!>") Thus far the only model paramter that has varied with tenpera- ture has been Ig, which was described in the EM nodel. An inproved temperature model is introduced here. Two types of improved tenpera~ ture variations are considered: those that can be expressed in terns of the device physics and those that are modeled by an enpirical data Fit, Only the latter variations require extra model parameters a) Physics-based temperature variation The model parameters whose tenperature variation can be modeled by consideration of the device physics are tp. Cp and Cyc T. niccartal3!) nas shown that the temperature variation of the basewidth is given by tc) c ow) ny where %¢ is the collector-base barrier potential Thom 15 the temperature at which the model parameters non were measured T is the temperature at which the parameter values are to be calculated {the operating condition) vi 4s the sensitivity of the dielectric constant, to T temperature (typically 200 ppm/*C for silicon) and yo 1s tne sesivivity of # to temeroture and is siven'bytr, é tee Le. nom (3 _ , vos pL. ae ( +3 (2.72) ron \Tnom eT, Equation (2.72) assumes the impurity concentrations on either side of the junction are approximately constant. Other assunptions involved in the derivation of &q. (2.71) are: a Taylor expansion of W about Vag = 0 and T + Tygqr IVgcl << Yq and the tenperature cor- rection is small. The temperature dependence of xp is then given by 15 t 2 E01) = 1g Tao) | (t) (2.73) where the W2 term results from the dependence of t, on W ang the => term results from the dependence of tg on D (Yue.. rpeke) Equations (2.71), (2.72) and (2.73) give the temperature variation of up 1 Cj¢ and Cie Equation (2.72) (and 2 similar one for vy) is used in the determination of the temperature variation of Cye (and Cye) (31) 3 4 Hecatia'") has also shown that for each junction capacitor am E41 OTpqq) ® (246-4) (2.78) AN] these temperature dependences are built into the program SLIC. “hetually, the ip. component of rp is the term that is proportional to W/O. It {eacsmed here@that ty hac the! came temperature dopendence ote that it becones necessary to have an accurate value for ¢ Sf the correct, tenperature varfation of rp 1s required. Previously, ¢ was only used to con- pate Cy¢b) Temperature-dependent parameters that require extra input parameters The paraneters described in this section require extra adel parameters to model their temperature variation. The extra param eters are used to fit the general empirical relationship. Par(T) © Par(Tyog) [1 * 1Cy(TTygq) * Tp -Fyeq)?] (2.75) where Par is the parameter being varied TC, is the first-order temperature coefficient, and TC, is the second-order temperature coefficient In SLIC and SINC Eq. (2.75) 1s applied to the parameters ips rh and rf. A theoretical expression for the tenperature variation of these parameters (which model the effect on emitter injection efficiency, transport efficiency, crowding and variation in base conductivity) 1s not available, The extra model parameters are TC, and TC) for sp, rf and r2 (a total of 6 extra input parameters] The model parameters that have nct been mentioned in this Section are assumed to be invariant with temperature. nal (Linearized) EM Model The small-signal, Tinearized EM nodel, which is described in Appendix 5, is very similar to that given for the EM, model in Sec tion 2.3.3. There are only two basic differences. = as a result of the Cio - rj split, an extra capaci- tor (equal to RATIO Ci (Vgic,}) is placed between nodes 8 and C', and = the equations determining r,. r,. © y and C,, are changed to include Ure effects incorporated tn the IM, model “672.4.5 Summary and Conclusions The My model has improved the de characteristics (basewidth f split and tg v5 Ig) and the variation with operating tenperature Basewidth modulation is included by modifying Igy fp and 1p Thu: these parameters must be specified at Veo = 0. An Early volt~ age, Vg» 1s used to describe basewidth modulation by the coltector- base junction (the Early effect). Whereas, in the normal, active modulation and f vs 1), the charge-storage model (Cy¢ ~ region, the In{Ic) V5 Vgp curve 1s affected by basewidth modulation, the In(Ip) vs Vge curve is, to first order, unaffected “v5.1 fs a result of the correct modeling of In({) vs V. The Tow-current drop in @ 1s caused by extra components of Ig which can be described by four made} parameters, Cy and ng (For 6¢) and Cy and gy (for fq). Two non-ideal diodes were added to the circuit model, The drop in # at high currents is caused by high-level injection effects. in the base which can be described by two model paraneters (0 for op and Og for fp). A set of input parameters more famiTiar to the circuit designer is used in the EM, model. The usefulness of the In(1) vs V curve was established: All of the nodel and input paraneters are directly obtainable fron it. Effects of otmic resistances and change of tenperatures must be considered Improved charge storage is obtained by splitting Cy¢ across ri, (requiring one extra parameter, RATIO) and by including the variation of 1p with T¢ (the two-dimensional effect of van der Ziel and Agouridis,(49) avaitable in the EM, model, requires two extra param- eters, Igg and ye) Inproved tenperature yariation is obtained in the EM, model by the use of 6 paraneters (3 sets of first- and second-order tenpera- ture coefficients for rj, ri and ap). The effects of tenperature on ss Gyp and Cig are inherently included in the fy mode? without the need for extra parameters. -68-2.5 The GP Model 2.5.1 Introduction The Gurmel-Poon (GP) model is the fourth and Final nonlinear nodel of the BJT described in this book. It is based on the mode? formulated by Guariet and Poon in 1970.'7) some modifications have been made, mainly with respect to terwinology, the region of the transistor to which it is applied and the required ode! param eters.(443) The GP model, which is available in the computer progran sPIce,(3) is described in some detail here. The GP mode! is almost entirely concerned with improvements in the dc charact- erization of the &M, model. These inprovenents, while appearing at first to be rather radical, are normally minor in Thus the GP model is basically equivalent to the EM; model The £M, model made improvenents in three areas: de performance (basewidth modulation and p versus 1), ac performance (the Cyc ~ rh split and tp versus Ic) and variation with anbient temperature. These effects were treated separately and the model was altered piece by piece. The GP model in this chapter, covers only three effects (basewidth modulation, Figh-injection effects and if versus Ic), but they are all treated together, This unified treatment provides @ slightly more accurate and complete model than is pro- vided by the EM, model. The GP model, however, is also more mathematical, less intuitive ané less convenient (in the input parameters required). A comparison of the GP and EM; models is included at the end of this chapter. A tabular surmary of the effects covered by the GP and EN, models is given in Table 2.1. The reader may well ask why one should know about the GP model if the EM, model is simpler and almost equivalent. The answer Ties ‘in the understanding of the operation of the BJT that is obtained: A better appreciation of the linitations and assumptions involved in the models is obtained from the GP model. This will be itlus~ trated further in this chapter -69-ie em, T=] p versus t= tow currents | Yes Yes Thigh currents | _ Yes Sasevidth Wodulation ves Yess in 8 ner Tp versus Tp Yes Th Oye sorte ves wo [Temperature Variation Yes wo TABLE 2.1 To make this chapter as “appetizing” as possible, the mathe~ natical derivations are kept toa minimum. Although the expressions ray at times look formidable, their manipulations are relatively simple and enphasis is placed throughout on an appreciation of what the equations mean and the assumptions inherent in their use. It should be pointed out that it is possible to make success~ ful use of the GP model in computer programs without fully under- standing the model. As Tong as the measurenents of the GP parameters as described in this book are understood and used properly, the GP mode! can be used. This approach, however, is not desirable since an appreciation of the accuracy of the results and the limitations of the model is not obtained. As well, computing time may be wasted by using an unnecessarily complicated model. ‘The starting point for the GP model lies approximately half- vay between the EM, and EN, models. It 15 assumed that the EM, pode! has been established and that the low-current drop in 8 is modeled by the inclusion of the two extra nonideal diodes. This assumed initial model 4s shown in Fig. 2.21. The improvements to the model of Fig. 2.21 afforded by the GP model will be incorporated by modifying existing equations (for Icy and Cog only) rather than by adding extra elenents. Note that the expression for 1, (Eq. (2.56)) 4s unaffected by these modifications. -10-Fig, 2.21. The assumed “starting point” model for the GP derivation (shown for {an npn transistor), The effects included in the following derivations are: 1 = a complete descristion of basewidth modulation 2 = the effects of high-level injection 3 = base-widening effacts which result in the variation of zp Involved in the derivation is a new definition of I, in terms of the internal physics of the transistor. Previously it had been defined (in the EM, model of Section 2.2) in terms of a terminal measurement which was @ consequence of reciprocity (a property that is stil assumed valid for the GP model) It should be emphasized that the GP derivation, like all the previous derivations in this book, assumes a “one-dimensional” netransistor (as iTlustrated in Fig. 2.22). The extension of the GP mode? to the three-dimensional (real life) case is looked at briefly in the next chapter which is devoted to the limitations of the models described in this book ‘The notation and model parameters described in this chapter are basically those enployed by the program spice.(“*3) the extra model parameters required for the GP model, in addition to those for the EM, model are: leg (which replaces the 1, parameter) Coe Cage Mays Mey (for low-current 8) % (Earty voltage) % (inverse Early voltage) 1 (knee current for In Ie versus x c Ypres) yg (inverse knee current for In Tp versus Vgics) B and its model parameters (for zp versus Ic) Space Charge Layers Foy tnitter base coNector bres A 1+ - Fig. 222. Simplified illustration of the one-dimensional device assumed for the GP modal oTThe four parameters (Coy Cys Mey and ng, ) were described in the previous chapter on the &, modcl: they are included in this Tist for completeness only and will rot be described further. The Early voltage, Vp, also described in the previous chapter, will be rede Fined for the GP model. Icc+ Vgs Ts Iya and 8 are the new model paraneters;* B, the last one, itself requires several parameters to describe it The following derivation of the GP model proceeds in basically three major steps. First, (in Section 2.5.2), from an exanination of the current density equations in the transistor a new expression for 1g is obtained, This new expression inherently incorporates the effects of basewidth modulation, high-level injection and tp YS. Tg. The second stage (covered in Section 2.5.3) consists of artificially separating out these effects so that they can be modeled and understood. The physical significance of this artifi- cial separation is explained and then (in Section 2.5.4) each artificial conponent 4s modeled in terms of measurable parameters Probably the most significant concept arising fron the GP deriva tion is the new importance given to the majority carrier (hole for an npn) concentration profile in the base. The energence of the majority carrier as inportent arises basically fron the use of a simple mathenatical “trick” in the derivation of I, Finally, to place the following analysis in its proper perspec~ tive, it is mainly concerned with inprovenents to the wodeling of second-order effects in the 8JT. It must be renenbered that the seemingly significant differences between the GP and EM, models are normally minor in their effect on the transistor's overall choract- eristics. The enphasis on the differences between the EM, and GP models are stressed here mainly as an aid to the understanding of the models and not as an indication of gross errors. If points are belabored (such as the accuracy of the Lesewidth modulation models) "Ihe parameter Ty was actusTly explained briefly in the EM sodel but ft wil) be treated in a*tuller manner in this chapter. “nethey are done so not because they are important on a first-order or even, perhaps, a second-order basis but to justify the approach used. 2.5.2 The Physical Definition of Ig'7"44-3) The derivation starts with the one-dimensional, de equations for the electron current density, Jn, Eq. (2.76), and the hole current density, dp, €q. (2.77), in an npn transistor. (4) aw,n(x) 6x) + q0, S509 (2.76) = x) = qo, A(X) ppl) G(x) - a0, (2.77) where E(x) is the electric field, n(x) is the free electron concen- tration and p(x) is the hole concentration. No restriction is, placed here, or later, on the variation of the carrier concentra tions, so that the folowing analysis applies for any doping profile, Equations (2.76) and (2.77) apply for both nigh- and ow Tevel injection ‘At this point it is assuned that the hole current is zero. This is not exactly the case, but is normally a reasonable approxi- mation. The approximation is justified by showing that there 1s no place where a large hole current could go. (a) The base-emitter junction is normally designed for high emitter-injection efficiency (high enitter doping with respect to the base doping).(16) this means that the current injected into the enitter from the base is smal] even when the enitter-base junction 1s forward-biased (b) In the nomal, active region the collector-base juncti; is reverse-biased and therefore no significant current Flows across it from the base to the collector, In the inverse and saturated regions of operation, however, the -14collector-base junction is forward-biased. It is assumed that for most cases of iaterest, the hole current that Flows from the base to the collector is still small Obviously, then, the following analysis wilT only be reasonably valid for the normal, active region and when the device is “weakly” saturated. For cases of strong saturation and inverse operation, it may not hold.* If the hole current is assuned to be zero, Eq. (2.77) becomes Quy 4x) Etx) = at, SHAD =0 (2.78) which can be rearranged to solve for the field, &. D 6x) = Bs ty. HE) (2.79) Substitution of Eq. (2.79) into fq. (2.76) gives 0, , ay voy hy Bel oid ae, 9 en ° “ lx) Jy = a, 6x) $20 + go, pix) ante) (2.81) where the Einstein relationship 0 ’p (= Hy ya has been used. The multiplication of Eq. (2.80) by p(x) is the rathenatical “trick,” which, with the following use of the differ- ential product rule will result in the importance of the majority carrier concentration, p(x), in the base. The use of the product rule on Eq. (2.81) gives ‘hetually, the assunption of Jp, = 0 is better than was explained: If the Jp ter is Petained in the folloving derivation, it will be sultipiies by nx}/pix) which reduces its effect even further (since, except at high level injection, n(x) €© pLx})px) Jy = a, Se {rte bs} (2.82) Both sides of this equation are now integrated from xj to x¢ where xz is the position of the enitter side of the emitter-base space- charge layer and xj is the position of the collector side of the collector-base space-charge layer, as illustrated in Fig. 2.23. Figure 2.23 also defines xg and xc, the positions of the base sides of these space-charge layers. Since current density Jq is constant for dc and independent of x (assuming negligible reconbination in the base region), it is taken out of the integral. Fig. 2.23, Bate profile for an npn transistor def xg and xc. Jing the intagea limits xe. Xe. Recombination in the space-charge layers is independently modeled by the tho non-ideal diodes in Fig. 2-21, $0 15 assumed to be zero here. -76-*c f px) dx = 99, & [over] ax =, [rts ot = alse! Iebag')! (2.83) a, (mag Dal") - five a The integration limits in the above (and following) equations need further explanation and justification. In their paper, Gunmel and Poon'7) perform the integration over the entire transistor. That is, xp" represented a point in the neutral emitter region near the emitter contact and xc’ represented a point in the neutral col- lector region near the collector contact. Their analysis then assumes that the minority carrier quasi-Fermi levels in the neutral enitter and collector regions stay constant. This assumption is (2.84) implicit in the next step of the derivation. The approach taken here, however, is slightly different. The integration is actually performed from the outside edges of the two space-charge layers, -- that is, from the enitter side of the emitter-base space-charge layer, x", to the collector side of the collector-base space-charge layer, xc'. AS a result, no assumption is made about the minority carrier quasi-Fermi levels in the neutral emitter and collector regions in the next step. (In fact, the original derivation of the new expression for Ig, as presented by Gunmel,'*6) performed the integration from the outside of the space-charge layers and then ‘assumed the result to be valid for the whole transistor). A com- parison of this approach with that of Gunmel and Poon is given later in this chapter.A further simplification can be made to Eq. (2.84) by applying Boltzmann statistics to the pn products: Were 2 ails! Dole!) = nj (xg! )oOne") (2.85) Therefore (2.86) At this point, the depletion approximation is mde. this assunes that there are no (or negligible) mobile carriers ina space-charge layer. That is, the field that is experienced by the carriers and the thickness of the space-charge layer are such that the carriers are transported "instantaneously" across it. This approximation, as has been pointed out in the EM, model (Section 2.3.2), 18 not valid for junctions under forward bias. The analysis that follows therefore takes the form of first assuming the depletion approximation to be true and then later fixes up the solution to take the mobile charges in the space-charge layers into account. In effect, the depletion approximation will only be applied for the thermal equilibrium condition. this is iTustrated {in Appendix 3, which gives an alternative, mathematica? and ore rigorous derivation of one aspect of the model. The treatment given in this chapter, however, while equivalent to that given in -78-Appendix 3, concentrates more on an understanding of the concepts involved. The depletion approxination is therefore made so that. the introduction to the concepts is kept as simple as possible The application of the depletion approximation results in the limits of the integral in Eq. (2,86) being replaced by xp and x, and the integration being performed in the neutral base region only. The change in integration limits results, of course, from the assumption that p(x) is approximately zero inside the space- charge layers. Equation (2.86) :an therefore be rewritten as: where A is the one-dimensional cross-sectional area (which converted current density to current) and the "-1" terms on the right hand side have been introduced for a later comparison 1, represents the total de minority current in the positive x direction that results from minority carriers injected into the base at the emitter and/or the collector. It is represented in the model of Fig. 2.21 by the current generator Igy. (The other compo- nents of the collector current in Fig. 2.21 are components of base current resulting from the injection of holes (for an npn) from the base towards the collector.) The equation for Icy, previously obtained in the EM, model, is 1, Tervem, moder) ~ "cc ~ Fee (2. (2. J tele a de (2.98)The sign of Icy is opposite to that of 1, (Ea. (2.87) since 1, has the opposite direction (out of the collector terminal) to that assumed for Ig (into the collector region from the terminal). A direct comparison of Eqs. (2.87) and (2.88) yields a physi- cal definition of Is. However, care must be exercised in the comparison. In previous work, 1, has been considered as a funda mental constant of the device. Yet the integral in Eq. (2.87) is not constant in that under high-level injection p(x). the majority carrier concentration, is a function of the applied bias. To reconcile this difference, a new symbol, Iggy 15 used in the GP model and is defined from Eq. (2.87) under Tow-Tevel injection con- ditions only. At low-level injection, the combination of Eqs (2.87) and (2.88) becomes: 1 . Tow level) f E (2.89) where p(x) has been replaced by Nq(x), since, at low current levels 20) joy Tevet 7 NalX) in the neutral base region, where Ke sk em . . (2.90) Before the definition of Igc is made, though, more attention must be paid to the limits of the integration, x; and xp. Because of the variation of the space-charge layer widths with applied voltage, Xp and xp are functions of the appropriate bias voltage (and, in fact, will be seen later to incorporate the effects of hasewidth modulation). The fundamental constant. Ig, is therefore defined at zero Vgp and Vgc as: -80-a0,n to f WO) ox 0 where gp and Xp are the values of xp and xe when the applied Junction voltages are zero. The fundamental nature of Ios is seen jmediately from Eq. (2.91) since it is uniquely detersined once the base-doping profile is Fixec. (As well, at zero bias voltages, £q. (2.90) becones alnost exact.) In the derivation of Eq. (2.91), the diffusion constant, Op. has been assumed to be constant and independent of x. In practice, this assunpticn is not valid. The diffusion constant should be (2.9) included in the denoninator integral. Instead, 0, is interpreted in Eq. (2.91) as an effective diffusion constant in the base. 7.44) The Og concept 7:44) The general expression for Ig (which is now a function of bias voltages) can be obtained in terms of the zero-bias constant Ig. Wowever, before this 1s done, it will be worthwhile to make a few definitions and to introduce sone new concepts bihen multiplied by q and A, the integral in fq. (2.87) repre- sents the total majority charge in the neutral base region and is given the symbol Oy xclVprcs % 4 p(x) dx (2.92) repre The dependence of the integration limits on the junction voltages has been enphasized here. The zero-bias majority base charge, Ogg. 1s defined by: co S80 of GAN, Ls) de (2.93) *0 ale