Вы находитесь на странице: 1из 9

IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 48, NO.

2, FEBRUARY 2001

323

Small-Signal Operation of Semiconductor Devices Including Self-Heating, with Application to Thermal Characterization and Instability Analysis
Niccol Rinaldi, Member, IEEE
AbstractA rigorous mathematical treatment of dynamic selfheating in semiconductor devices is presented. Two formulations for the admittance parameters are given. The thermal behavior of the device is referred to device temperature in the first formulation, and to ambient temperature in the second. Contrary to previous work, nonlinear thermal effects are included. An analytical model for the thermal resistance is derived which confirms the relevance of these effects. Applications of the above results to device modeling and thermal characterization are studied in detail by means of numerical simulations. Possible sources of inaccuracies are evidenced. Finally, it is shown that the differential analysis of thermal feedback provides a general and rigorous means to determine the conditions for the onset of thermally-induced instabilities. Index TermsBipolar transistors, frequency response, semiconductor device modeling, temperature.

I. INTRODUCTION HE analysis of thermal behavior of semiconductor devices has become a critical issue in the design of modern electronic systems. For example, it is well known that the exponential dependence of minority-carrier currents on temperature variations makes bipolar devices particularly sensitive to electrothermal effects [1]. These effects not only represent a major limitation on the high power operation (giving rise to thermal instabilities, hot spot formation and thermal runaway [1][4]), but also affect to a significant extent device performance (due to self-heating and thermal coupling effects [5], [6]). These problems are further aggravated by the continuous downscaling of device dimensions, which gives rise to higher thermal resistances and increased power densities. Thermal effects are a main concern also in emerging technologies, such as SOI (silicon-oninsulator) and GaAs-based ICs, due to the poor thermal conductivity of the substrate [7][10]. As a consequence, an accurate characterization and understanding of self-heating is of utmost importance for ensuring the safe operation and predicting the reliability and performance of solid-state devices and circuits. The primary emphasis in this paper is on dynamic self-heating effects in semiconductor devices operating under small-signal excitation [11][13]. During small-signal

Manuscript received January 19, 2000; revised September 8, 2000. This work was supported by the MURST national program on microelectronics and by the Italian National Council of Research. The review of this paper was arranged by Editor A. H. Marshak. The author is with the Department of Electronics Engineering of the University of Naples Federico II, 80125 Naples, Italy (e-mail: rinaldi@diesun.die.unina.it). Publisher Item Identifier S 0018-9383(01)00775-4.

operation the variation of the electrical signal causes an instantaneous variation in the dissipated power, and hence in the device temperature. The instantaneous variation of temperature with the applied signal, in turn, influences the input and output currents, and hence the parameters of the two-port network which describes the small-signal operation of the device. These parameters, therefore, are affected not only by dc self-heating (caused by the dc bias), but also by ac self-heating. Dynamic self-heating is a frequency-dependent phenomenon. Due to the thermal inertia of the substrate and package (often described in terms of thermal capacitances), the temperature cannot follow the ac signal at high frequencies, and remains constant at the value determined by the static power dissipation. As a consequence, dynamic self-heating disappears at frequencies above the thermal cutoff frequency [8], [14]. An in-depth analysis and modeling of dynamic self-heating is important for different reasons. First, dynamic self-heating may have a pronounced influence on the small-signal behavior of both bipolar and unipolar devices. A well known example is the negative output conductance observed in heterojunction bipolar transistors [15] and SOI or bulk MOSFETs [14], [16]. As a consequence, these effects must be accounted for in the design of analog systems, as evidenced by different authors [5], [6], [8]. Secondly, analytical formulations for small-signal thermal feedback effects have been used in various techniques for the measurement of the thermal resistance and of the thermal time constants. The first analytical formulation of the two-port parameters including ac self-heating dates back to the classic paper of Mller [11], which has been adopted in various studies dealing with dynamic thermal feedback in bipolar devices [5], [6], [17]. The analysis of Mller, however, has been developed by considering a current-controlled device with a bias- and temperature-independent current gain. This approximation has not been previously noted, and limits the accuracy of this formulation. A more general treatment was later developed by Caviglia and Iliadis [13]. The analysis presented in this paper improves published work in different aspects. 1) The result derived by Caviglia and Iliadis [13] is generalized to include nonlinear thermal effects, i.e., the dependence of thermal conductivity on temperature [18]. These effects make the thermal impedance a nonlinear function . The of dissipated power and ambient temperature increase of thermal resistance with dissipated power is a well known effect which influences the thermal behavior of GaAs HBTs [9], [10], [19], MESFETs [20], as well as

00189383/01$10.00 2001 IEEE

324

IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 48, NO. 2, FEBRUARY 2001

silicon devices [1], [4]. The dependance of thermal resistance on ambient temperature has been less documented. It will be shown that both effects must be taken into account for an accurate extraction of the thermal resistance by means of small-signal techniques. The first effect has been neglected in all previous papers, while the second effect was accounted for only in the paper of Jomaah et al. [21], where a method for the extraction of the thermal resistance in thin-film SOI MOSFETs based on the measurement of the output conductance is proposed. Furthermore, to more closely examine the impact of nonlinear thermal effects, a simple closed-form expression of the thermal resistance is derived. 2) The mathematical treatment presented here also includes a detailed analysis of the dependence of device currents and temperature on variations of ambient temperature. This analysis is of practical relevance, as it leads to simple techniques for the determination of the thermal resistance. All previous results based on this approach will be reviewed and shown to be particular cases of our more general formulation. 3) Most published papers used the frequency-dependence of the output conductance to determine the thermal time constants [8], [14], [22], [23]. In this paper, we analyze in detail the frequency-dependence of the imaginary part of the admittance matrix (i.e., the capacitances), and propose an accurate method for the extraction of the thermal time constants. Moreover, it is shown that the analysis of the frequency response of both conductances and capacitances allows the separation of the thermal spreading resistance from the package thermal resistance together with the associated time constants. 4) Finally, it is shown that the differential analysis of thermal feedback provides a general and rigorous means to determine the conditions for the onset of thermally-induced instabilities. This model can be therefore used to predict the boundary of the safe operating area related to thermal instability. II. TWO-PORT REPRESENTATION OF SMALL-SIGNAL OPERATION INCLUDING SELF-HEATING Following the approach of Mller [11], the currents at the input and output terminals can be considered as functions of the input and output voltages and device temperature: (1) (2) The dependence of device temperature on dissipated power and ambient temperature is usually described [11]: by introducing a thermal resistance

A closed-form analytical model of the thermal resistance is developed in Appendix. With reference to the admittance representation, the currents and , and the temperature may be regarded as dependent variables. The functional dependence of these variables on the is implicitly defined by the independent variables , and nonlinear system of (1)(3). The derivatives of , , and may be found on application of the theorem of the differentiation of implicit functions [24], or by directly computing the derivatives of the above system of equations. A straightforward calculation yields

(4)

(5) (6)

(7)

(8) (9) represents the thermal impedance [11]. Here the where , terms denote the admittances for the isothermal case, i.e. the temperature does not vary with the ac signal and remains fixed at the value determined by the dc bias. Therefore, the parameters , by definition, are affected only by dc self-heating, while are also affected by ac self-heating. The term parameters at the denominator of (4)(9) is defined as (10) represent the temperature where coefficients of the input and output currents, while the term is related to the dependence of on dissipated power, and is given by (11)

(3) Nonlinear thermal effects cause the thermal resistance to depend on dissipated power [19] and ambient temperature.

RINALDI: SMALL-SIGNAL OPERATION OF SEMICONDUCTOR DEVICES

325

An analytical expression for is given in Appendix. This , effect was neglected in previous work [11], [13]. For (4), (5), (7), (8) reduce to the results of Caviglia and Iliadis [13]. Equations (6) and (9) have not previously reported. The formulation of Mller [11], which has been used in [5], [6], [17] can be also regarded as a particular case, with the additional assumption that the currents are related by a , where the current proportionality relationship: is independent of bias and temperature. The rather gain crude assumption relative to the constancy of has been not previously noted, and can limit the accuracy of the formulation of Mller [11]. The effect of the term can be analyzed by using the theoretical expression given in the Appendix, or . For using experimental results for the dependence example, using in (11) the relation found in [19] for the thermal resistance of AlGaAs/GaAs HBTs, we find for W, while using the relation found in [25] we for mW. These results indicate that find the dependance of the thermal resistance on dissipated power should be taken into account, particularly at high levels of dissipated power (see also the discussion in Appendix).

III. DISCUSSION As noted in Section I, (4)(9) can be used can be used in compact modeling [6] or for thermal characterization. In order to verify the accuracy of these relations, ac thermal effects were studied by means of the two-dimensional device simulation program MEDICI [26]. While previous papers were based on valuable experimental data, here we perform simulated experiments, which avoid the uncertainties encountered in experimental investigations, and allow an unambiguous verification of model predictions. A bipolar transistor is used as a test device. Fig. 1(a) shows the equivalent thermal network of the simulated structure. The overall device-to-ambient thermal includes two components: a thermal spreading resistance related to heat flow into the substrate, and a resistance , related to the substrate-to-ambient thermal resistance thermal characteristics of the package. Although two-dimensional (2-D) simulation codes are being widely used for studying the impact of thermal feedback effects on device behavior [7], [14], a word of caution is in order. Two factors may limit the accuracy of 2-D simulations. i) Due to the limited number of grid points available, the simulation domain represents only a small portion of the substrate [see Fig. 1(a)]. While a reduced simulation domain does not affect the electrical behavior, the temperature distribution may be markedly modified. ii) A 2-D simulation code cannot accurately model the heat flow into the substrate which is three-dimensional (3-D) in nature. These problems were overcome by means of a simple trick. 30 m First, the thermal spreading resistance of a 1 m emitter was calculated using a 3-D analytical model [10]. Then, a thin oxide layer was added at the bottom of the simulation domain, as shown in Fig. 1(b). The oxide layer thickness was obtained chosen so as the thermal spreading resistance

Fig. 1. (a) Schematic cross section of an integrated bipolar transistor with equivalent thermal network. The portion of the substrate considered for the simulations is indicated. (b) Schematic cross section of the simulated structure. An oxide layer is added at the bottom of the simulated structure, so as the value of the thermal spreading resistance R for the simulated structure (b) coincides with that of the real structure (a).

by MEDICI simulations agreed with the calculated value1 . To include the effect of the package a thermal resistance C m W was added to the simulated plot. As is structure (see Fig. 1). Fig. 2 shows the well known, the region with negative differential resistance is due to self-heating [1][4]. The snapback point (marked by full circle) is related to the onset of thermal instability, and limit the safe operating area of the device [1][4]. It should be noted that the snapback in the output characteristics can be also caused by impact ionization. In the simulation results shown in Fig. 2 avalanche effects were not included since the snapback voltage is only marginally influenced by impact ionization. However, depending on bias conditions and device structure, the snapback voltage related to avalanche effects may be lower than the snapback voltage associated with thermal effects alone [1]. In such cases impact ionization should be included into analysis.
1It should be noted that using this approach we can fit the value of the thermal resistance but not that of the thermal capacitance. As a consequence, the value of the thermal cutoff frequency extracted from 2-D simulations should be taken with caution.

326

IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 48, NO. 2, FEBRUARY 2001

Fig. 2. Output characteristics (MEDICI simulations). Also shown are the snapback points (full circles), and the quiescent operating point considered for the ac analysis (point ).

To investigate dynamic self-heating effects, the device was biased at point in Fig. 2, and ac analysis was performed [26]. Re relative to Fig. 3(a) shows the conductances the common-emitter configuration as a function of frequency. MEDICI simulation (full lines) are compared to analytical results (symbols) obtained from (4), (5), (7), (8). In these relations the thermal impedance was approximated using a single-pole function [8], [11], [13]

(12) All parameters on the rhs of (4), (5), (7), (8) were evaluated from MEDICI simulations. The thermal coefficients , were calculated by ad hoc simulations, while the dc currents and voltages are directly given by the program. The evaluation of the is rather straightforward, since the prothermal resistance gram also includes the junction temperature among the output , and variables. The extraction of the thermal time constant will be discussed below. The of the isothermal parameters plots in Fig. 3(a) evidence the existence of three regions. In the first region the frequency of the sinusoidal signal is lower . In this rethan the thermal cutoff frequency are difgion dynamic self-heating is active, and parameters . The ferent from the corresponding isothermal parameters and parameters difference is more pronounced for the , ac self-heating also causes a change [5]. In the case of of sign, as evidenced in Fig. 3(a). These effects may significantly impact the performance of analog circuits [5], [6], [8]. Note that the difference from the isothermal case increases as moves toward the snapback the quiescent operating point point, since the term at the denominator of (4)(9) approaches unity (see Section IV). As the frequency becomes higher than (0.25 MHz in Fig. 3), the dethe thermal cutoff frequency vice temperature cannot follow the ac signal. Therefore, in reapgion II ac shelf-heating disappear, so that the parameters , as can be proach the corresponding isothermal parameters . In easily deduced from (4), (5), (7), (8) by letting
Fig. 3. Magnitude of common-emitter conductances (a) and capacitances (b) as a function of frequency; full lines: MEDICI simulations with self-heating included; dashed lines: MEDICI simulations without ac self-heating; symbols: analytical results. The signs and indicate the change of sign in the (a) parameter and in the (b) capacitances and .

the high-frequency region (denoted as region III) the conductances become frequency-dependent due to distributed effects, and the representation of electrical behavior using a lumped-parameter equivalent circuit becomes inaccurate. To further confirm this interpretation, isothermal MEDICI simulations were performed by excluding the self-heating option from the command file [26] (dashed lines). In order to include dc self-heating, the temperature was set equal to the value determined at point in the nonisothermal simulation. Since point is far from the snapback point, the device temperature is only 303 K. As shown in Fig. 3(a), the isothermal curves (ac heating excluded) coincide with the nonisothermal curves (ac heating included) in region II and III. Therefore, the value of the isothermal parameRe to be inserted in (4), (5), (7), (8) could be ters extracted either from the nonisothermal simulations (region II) or from the isothermal simulations. The last parameter to be determined is the thermal time constant. A common technique is to fit the frequency response of the [23], or another small-signal parameter output conductance [13]. Here a different approach is adopted. Fig. 3(b) shows the Im as a function of frequency. This capacitances behavior has not previously studied. Again, the nonisothermal

RINALDI: SMALL-SIGNAL OPERATION OF SEMICONDUCTOR DEVICES

327

and isothermal MEDICI simulations (full and dashed lines, respectively) are compared to model results (symbols). Here the effect of ac self-heating is even more dramatic, causing a variation of several orders of magnitude in all capacitances, and a and which become negative in the low change of sign in frequency region. The analytical relations (4), (5), (7), (8) indicate that the magnitude and sign of all capacitances at low frequencies depend on the value of the thermal time constant. can be set so as to obtain a good Therefore, the value of agreement with MEDICI data at low frequencies. By fitting the obtained from (4) to the MEDICI low-frequency value of MHz. Using this value, all caldata yields the value and curves were found in good agreement culated with simulation results (Fig. 3). The error sensitivity of the proposed method does not appear to be critical. Since the extracted thermal cutoff frequency is extracted from the value of the capacitances measured at low-frequencies, the dependence of the was investigated. A rather tedious analysis capacitances on the capacitances has shown that at low frequencies . As a consequence, the are approximately proportional to is approximately equal relative error in the extracted value of to the relative error of the measured capacitance: . Note that, due to the small temperature increase, the effect of dependence of the thermal resistance on dissipated power can be neglected for the case under study ). Additional simulations to investigate this effect were ( not performed, since 2-D simulation programs cannot correctly describe thermal spreading effects, as noted above. The frequency behavior of the thermal impedance in practical cases is usually more complex than a first-order system [13], [23]. This is due to distributed thermal spreading effects in the substrate [17] and to the thermal behavior of the package [5]. To nJ investigate the latter effect, a thermal capacitance C m was placed in parallel to , and the ac analysis was repeated. The MEDICI results are shown in Fig. 4 (full lines). A two-pole system is now necessary to accurately describe in the analytical model

Fig. 4. Magnitude of common-emitter conductances (a) and capacitances (b) as a function of frequency; full lines: MEDICI simulations with self-heating included; symbols: analytical results. These results differ from those shown in Fig. 3 in that a thermal capacitance C has been added in parallel to R .

all parameters, the analytical and traced, as indicated by symbols in Fig. 4. (13) where the first term accounts for the contribution of the package, while the second term is related to thermal spreading effects in related the substrate. Since the time constant to the package is usually much higher than [5], region I is further decomposed into two subregions. For (subregion IA) both terms in (13) contribute to the (subregion IB), thermal response. For is shunted by , and only the second term in (13) is significant. The procedure for extracting the thermal time constants is was found by fitting similar to that outlined above. First, obtained from (4) to MEDICI data in subthe value of . region IB, and by approximating was easily found from The thermal spreading resistance MEDICI results, while the isothermal parameters were deterwas evalmined from MEDICI ac data in region II. Finally, at Hz (subregion uated by fitting the capacitance IA) and by considering both terms in (13). Having determined

plots can be thus

The above analysis has shown that the analytical relations (4)(9) give an accurate description of ac self-heating, provided that an appropriate model for the thermal impedance is adopted. With the exception of the thermal time constants, no fitting parameters were used in the calculated admittance parameters. Conversely, the analytical relations for the conductances can be also used for the measurement of the thermal resistance(s) if all other parameters are known or measurable. The expression can be easily obtained by taking the of the conductances at low frequencies (region I), real part of the admittances i.e., by performing the following substitutions in (4), (5), (7), , , and Re . Depending (8) (region IB) or on the chosen frequency, one can extract (region IA). In principle, it is therefore possible from the package resistance. Of course, for the to separate thermal resistance extraction we can confine the analysis to only [8], [27], one parameter, usually the output conductance can be obtained form [28]. The nonisothermal parameter the dc characteristics, or using an impedance analyzer. Various

328

IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 48, NO. 2, FEBRUARY 2001

techniques have been adopted to determine the isothermal pa. As shown above, it can be obtained by measuring rameter at frequencies above the thermal cutoff frequency [8], [22], for [23] (region II). Caviglia and Iliadis [28] determined a SOI MOSFET by fitting the measured -parameters with the negligible small-signal model. Jomaah et al. [21] assumed for a SOI MOSFET biased in the saturation region. Another problem is the evaluation of the temperature coefficient , since the device temperature is not directly measurable. In some papers, analytical models are employed (e.g. [23] for SOI MOSFETs and [27], [29] for bipolar transistors). This approach, however, links the accuracy of the thermal resistance extraction method to the accuracy of the model [28]. A more elegant approach is based on the analysis of the response to variations of ambient temperature. As noted in Section II, the ambient can be regarded as an independent variable. Diftemperature yields ferentiating , and with respect to (14) (15) (16)

measurable coefficients , instead of . A general formulation for the admittance parameters in terms of the thermal cowas still lacking. Similar relations for the parameefficients or obtained in previous papers [21], [28], [30] may ters regarded as particular cases. In [28] and [30], the difference beand was not considered (since the dependence of tween on and was neglected). In Appendix it is shown that neglecting this difference may result in an inaccurate value for the thermal resistance extracted using the differential technique. In [28] a relation for the output conductance of a MOSFET was derived, which can be obtained from our general relations by letand (i.e. ). Jomaah et al. ting in the case of a MOSFET by [21] derived an expression for on , so that the exconsidering only the dependence of derived in [21] is actually incorrect ( ). Fipression of nally, Zweidinger et al. [30] derived an analytical expression for parameter of a bipolar transistor which can be obtained the expression given by (19) by assuming from the and . has been extracted, the It is worthwhile noting that once factor can be easily evaluated from (17), and the terms , and can be evaluated from (14)(16). Therefore, this technique also allows a simple calculation of the thermal coefficients and . IV. THERMAL INSTABILITY ANALYSIS

is a term related to the depenwhere dance of the thermal impedance on ambient temperature. This effect has been taken into account only in the expression of the output conductance of SOI MOSFETs derived in [21], where it is shown that it can be particularly significant at low temperatures. An analytical expression for can be found in Appendix. and can Note that, contrary to and , the coefficients is a directly measurable quantity. be easily determined, since Substituting (14)(16), we obtain (17) where the term is defined as (18) Substituting (14)(16) into the expression of the admittance parameters (4), (5), (7), (8) yields (19) . By taking the real part of (19), the expreswhere , can be easily found. sions of the nonisothermal conductances These relations can be used to extract the thermal resistance Re and hence the actual thermal resistance . An analytical expression for the modified thermal resistance is derived in Appendix. The advantage of this approach is that the thermal resistance is expressed in terms of the more easily

The theoretical relations of the nonisothermal conductances may provide a deeper insight into the analysis of thermal instabilities. As lucidly evidenced by different authors [1][4], the snapback point in the output characteristics is related to the onset of thermal instability. The snapback point is defined by , which implies (see (8)), and the condition hence from (10) (20) If analytical models are available for all quantities in (20), this relation can be used to predict the snapback voltage and hence to construct the boundary of the safe operating area related to and (20) thermal effects [3]. Assuming (see, e.g., reduces to the well known condition [3]). Therefore, (20) represents the most general condition for the thermal instability for a voltage-controlled device ( ). A similar analysis yields the instability ). condition for a current-controlled device ( in the numerator of (4), in this case Neglecting the term the instability condition becomes . From relation (4), (5), (7), (8) it follows that in correspon) all conductances become dence of the snapback point ( infinite. In contrast, Fox et al. [5] deduced the existence of two instability conditions relative to the input and output conductance parameters. This obviously incorrect result is due to a naive application of Mllers model [11], and to the fact that the limitations of this models were not recognized. To further show the generality of this approach, let us find the is instability condition when an extrinsic resistance (ballast)

RINALDI: SMALL-SIGNAL OPERATION OF SEMICONDUCTOR DEVICES

329

solutions have been suggested. Thermal characterization of semiconductor devices using differential techniques has been also discussed. Finally, by means of differential analysis of thermal feedback, the conditions for the onset of thermally-induced instabilities for various circuit configurations have been determined. It has been shown that this approach represents a rigorous method for predicting the limits of the safe operating area related to thermal runaway. APPENDIX In this section, we derive an analytical model for the thermal resistance, which includes nonlinear thermal effects and the contribution of the package. The thermal resistance is defined by
Fig. 5. Factor @ T =@ T as a function of collector-base bias V for V 0:75 V; full lines: MEDICI simulations; symbols: analytical results (see text).

(A1) is the ambient temperature, where is the dissipated power, is the device temperature at some reference point (e.g., and the emitter center). The temperature distribution caused by the power dissipated by a device placed on the surface of a chip is described by the nonlinear 3-D heat flow equation [4], [10] (A2) where the temperature-dependence of thermal conductivity is often expressed as (A3) where is a reference temperature and represents the value . Assuming K, quoted values are of for W cm K and for GaAs [9], while W cm K and for Si [1]. A different dependence was considered by Heasell [4], relation for the see also [20]. A well-known technique to remove the nonlinearity in (A2) is a change of dependent variable by means of the Kirchoff transformation [4], [9], [10]. A transformed temperature is defined by [9] (A4) where use has been made of (A3). By substituting (A4) into (A2) we find that the fictitious temperature satisfies the Laplace equation (A5) with linearized boundary conditions [4]. The linearized temperature may be simply regarded as the solution of the linear heat flow problem for the ideal case of a temperature-independent ). The solution of the linearized thermal conductance ( problem may be found by means of conventional techniques.

connected to the emitter terminal of a bipolar transistor (see [31, Fig. 1]). Using a common-base conductance representation for the bipolar transistor, the derivative of the emitter (input) current with respect to the external emitter-base voltage under shorted output conditions is (21) is given by (4). where the common-base input conductance . This conThe instability condition now reads dition reduces to that obtained by Adlerstein [31] ( ) if the following simplifications are made: and in (10), and , , , and in the numerator of (4). With (which is not verified for the exception of the assumption significant dc self-heating), all other approximations are reasonable for the common-base connection. Finally, it is interesting to observe that in correspondence of the snapback point defined by (20), the device becomes unstable also with respect to variations of ambient temperature (from (16) ). This behavior is it follows that is plotted as a shown in Fig. 5, where the factor function of collector-base voltage as the operating point moves V in Fig. 2. along the characteristic relative to MEDICI simulations (full line) are compared to analytical results obtained from (16) and (17). V. CONCLUSION A rigorous mathematical treatment of the dynamic response of semiconductor devices to variations of electrical signal and ambient temperature has been presented. Two formulations for the admittance parameters have been derived, which include all previous results as particular cases. The analysis accounts for nonlinear thermal effects, which have been further studied by deriving an analytical model for the thermal resistance. Dynamic self-heating in bipolar transistors has been analyzed in detail by means of numerical simulations. Limitations of 2-D simulation codes have been pointed out, and possible

330

IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 48, NO. 2, FEBRUARY 2001

Due to the linearity of transformed heat flow problem, the transmay be generally exformed device temperature pressed as (A6) denotes the transformed substrate temperature (see Fig. 1), and represents the thermal ). The spreading resistance for the linearized problem ( , fictitious temperature distribution, and hence the value of depend on the geometry of the heat source and of the substrate. have been obtained for particuAnalytical expressions for larly simple geometries [1], [4]. For the more realistic case of a rectangular heat source, analytical expressions were derived by means of approximate relations [5], [6] or in form of Fourier series [10]. is related to the The actual substrate temperature package resistance by where (A7) By combining (A1), (A4), (A6) and (A7) we obtain

Fig. 6. Ratio R =R as a function of normalized power p, with normalized ambient temperature t and normalized package resistance r as parameters.

(A8) where all quantities have been written in normalized , , , form: . It is to be noted that although the above analysis is similar to that presented in [1], [4], [9] two important additional factors have been accounted for: 1) the effect of the package thermal resistance, and 2) an ambient temperature different from the reference temperature K. Having obtained a general expression for the thermal resistance, we are in a position to evaluate the terms and related on dissipated power and ambient temto the dependence of perature, respectively, as follows:

for large dissipated power or for a high ambient temperaand , at low dissipated power ture. Assuming ) the term and the modified thermal resistance can ( and . be approximated as As a consequence, nonlinear thermal effects become signifi, since cant if the normalized power is greater than about and . It should be pointed out that the thermal resistance model presented above is accurate if hot spot phenomena do not occur. As well known, these effects occur in multifinger bipolar transistors (e.g., [19]), and are due to the fact that as the power dissipation is increased the innermost sections of the device tend to carry most of the current. The reduction in active area, in turn, causes the thermal resistance to increase. Therefore, in the case of hot spot occurrence, the rate of increase of the thermal resistance with dissipated power is actually greater than predicted by (A9). REFERENCES
[1] B. H. Krabbenborg, Modeling and simulation of electrothermal interactions in bipolar transistors, Ph.D. dissertation, Unive. Twente, Enschede, The Netherlands, 1996. [2] M. Latif and P. R. Bryant, Multiple equilibrium points and their significance in the second breakdown of bipolar transistors, IEEE J. SolidState Circuits, vol. 16, pp. 815, 1981. [3] P. L. Hower and P. K. Govil, Comparison of one- and two-dimensional models of transistor thermal instability, IEEE Trans. Electron Devices, vol. ED-21, pp. 617623, 1974. [4] E. L. Heasell, The heat-flow problem in silicon: An approach to analytical solution with application to the calculation of thermal instability in bipolar devices, IEEE Trans. Electron Devices, vol. ED-25, pp. 13821388, 1978. [5] R. M. Fox, S. Lee, and D. T. Zweidinger, The effect of BJT self-heating on circuit behavior, IEEE J. Solid-State Circuits, vol. 28, pp. 678685, 1993. [6] D. T. Zweidinger, S. Lee, and R. M. Fox, Compact modeling of BJT self-heating in SPICE, IEEE Trans. Computer-Aided Design, vol. 12, pp. 13681375, 1993. [7] H. T. Lim, F. Udrea, D. M. Garner, and W. I. Milne, Modeling of self-heating effect in thin SOI and partial SOI LDMOS power devices, Solid-State Electron., vol. 43, pp. 12671280, 1999. [8] B. M. Tenbroek, M. S. L. Lee, W. Redman-White, R. J. T. Bunyan, and M. J. Uren, Impact of self-heating and thermal coupling on analog circuits in SOI CMOS, IEEE J. Solid-State Circuits, vol. 33, pp. 10371046, 1998. [9] K. Poulton et al., Thermal design and simulation of bipolar integrated circuits, IEEE J. Solid-State Circuits, vol. 27, pp. 13791386, 1992.

(A9) (A10) Finally, the modified thermal resistance is given by (A11) ) (A11) reduces to Neglecting nonlinear thermal effects ( , as expected. Fig. 6 shows the ratio as a function of normalized power , with normalized amas bient temperature and normalized package resistance may significantly differ from parameters. As can be seen,

RINALDI: SMALL-SIGNAL OPERATION OF SEMICONDUCTOR DEVICES

331

[10] E. Koenig et al., Current-temperature feedback effects in III-V heterojunction bipolar transistors, IEEE J. Solid-State Circuits, vol. 31, pp. 122127, 1996. [11] O. Mller, Internal thermal feedback effects in four-poles especially in transistors, Proc. IEEE, vol. 52, pp. 924930, 1964. [12] O. Mller and J. Pest, Thermal feedback in power semiconductor devices, IEEE Trans. Electron Devices, vol. ED-17, pp. 770782, 1970. [13] A. L. Caviglia and A. A. Iliadis, Linear dynamic self-heating in SOI MOSFETs, IEEE Electron Device Lett., vol. 14, pp. 133135, 1993. [14] R. H. Tu et al., An ac conductance technique for measuring self-heating in SOI MOSFETs, IEEE Electron Device Lett., vol. 16, pp. 6769, 1995. [15] P. M. McIntosh and C. M. Snowden, Measurement of heterojunction bipolar transistor thermal resistance based on a pulsed I-V system, Electron. Lett., vol. 33, pp. 100101, 1997. [16] P. S. Barlow, R. G. Davis, and M. J. Lazarus, Negative output conductance of self heated power MOSFETs, Proc. Inst. Elect. Eng. I, vol. 133, pp. 177179, 1986. [17] R. M. Fox and S. Lee, Scalable small-signal model for BJT self-heating, IEEE Electron Device Lett., vol. 12, pp. 649651, 1991. [18] W. B. Joyce, Thermal resistance of heat sinks with temperature-dependent conductivity, Solid-State Electron., vol. 18, pp. 321322, 1975. [19] W. Liu, A. Khatibzadeh, J. Sweder, and H. Chau, The use of base ballasting to prevent the collapse of current gain in AlGaAs/GaAs heterojunction bipolar transistors, IEEE Trans. Electron Devices, vol. 43, pp. 245251, 1996. [20] F. Bonani, G. Ghione, and C. U. Naldi, A new, efficient approach to the large-scale thermal modeling of III-V devices and integrated circuits, in IEDM Tech. Dig., 1993, pp. 101104. [21] J. Jomaah, G. Gibaudo, and F. Balestra, Analysis and modeling of selfheating effects in thin-film SOI MOSFETs as a function of temperature, Solid-State Electron., vol. 38, pp. 615618, 1995. [22] W. Redman-White et al., Direct extraction of MOSFET dynamic thermal characteristics from standard transistor structures using small-signal measurements, Electron. Lett., vol. 29, pp. 11801181, 1993. [23] B. M. Tenbroek et al., Self-heating effects in SOI MOSFETs and their measurement by small signal conductance techniques, IEEE Trans. Electron Devices, vol. 43, pp. 22402248, 1996.

[24] M. Krasnov, A. Kiselev, G. Makarenko, and E. Shikin, Mathematical Analysis for Engineers. Moscow, Russia: Mir, 1989. [25] P. Baureis and D. Seitzer, Parameter extraction for HBTs temperature dependent large signal equivalent circuit model, in GaAs IC Symp., 1993, pp. 263266. [26] MEDICI User Guide, Technol. Model. Assoc., Palo Alto, CA, 1993. [27] R. M. Fox and S. G. Lee, Thermal parameter extraction for bipolar circuit modeling, Electron. Lett., vol. 27, pp. 17191720, 1991. [28] A. L. Caviglia and A. A. Iliadis, A large-signal SOI MOSFET model including dynamic self-heating based on small-signal model parameters, IEEE Trans. Electron Devices, vol. 46, pp. 762768, 1999. [29] G. Meijer, The current dependency of the output conductance of voltage-drive bipolar transistors, IEEE J. Solid-State Circuits, vol. 12, pp. 428429, 1977. [30] D. T. Zweidinger et al., Thermal impedance extraction for bipolar transistors, IEEE Trans. Electron Devices, vol. 43, pp. 342346, 1996. [31] M. G. Adlerstein, Thermal stability of emitter ballasted HBTs, IEEE Trans. Electron Devices, vol. 45, pp. 16531655, 1998. [32] V. Kadambi and N. Abuaf, An analysis of the thermal response of power chip packages, IEEE Trans. Electron Devices, vol. ED-32, pp. 10241033, 1985.

Niccol Rinaldi (M95) graduated (cum laude) from the University of Naples Federico II, Italy, in 1990, and received the Ph.D. degree in 1994. In February 1994, he became a Research Assistant at the University of Naples Federico II. In 1995, he was charged by the Second University of Naples to teach a course on electron devices. From July 1996 to December 1996, he was a Research Fellow at the University of Delft, The Netherlands, working on the modeling of high-speed bipolar devices. Since 1998, he has been Associate Professor at the University of Naples Federico II. Dr. Rinaldi has been a reviewer for IEEE TRANSACTIONS ON ELECTRON DEVICES, Solid-State Electronics, and Microelectronics Journal. He currently is vice-chairman of the IEEE Electron Device Chapter (Central & South Italy Section).

Вам также может понравиться