Behavior of GaAs FET Pulsed IV Characteristics
The IV characteristics of GaAs MESFET devices acquired under a variety of pulsed conditions are compared. The results show that the differences between the static and pulsed characteristics are not due to thermal effects alone as is sometimes assumed.
University of Bradford, Department of Electronic & Electrical Engineering
West Yorkshire, UK
A comparison of the IV output characteristics of GaAs MESFET devices measured under static and pulsed conditions is often used as a means of observing and modeling heating effects in the device.1 However, results presented in previous papers have cast doubt on the assumption that the differences in the characteristics are due solely to temperature effects. The results have indicated that frequency dispersion effects contribute to such differences and that this phenomenon is linked intrinsically to thermal effects.2 In this article, data under a variety of test conditions are presented to highlight the deficiencies of the pulsed measurement technique for modeling purposes.
Measurements and Test Devices
The measurement of the device IV output characteristics under pulsed conditions is normally performed using the scheme shown in Figure 1 .3 In such systems, the gate and drain terminals are pulsed and the drain current is measured during the on period of the pulse. The width of the drain pulse normally is smaller than the gate pulse and is applied after the gate pulse to prevent the flow of excessive drain current. Providing the width of the applied pulses is short and the period of the pulses is long, the effects of the device self-heating can be minimized. The differences between the characteristics measured in this fashion and those measured under static conditions are often assumed to be due to thermal effects.
Although systems able to measure the characteristics of the device using narrow pulses (< 300 ns width) with long repetition rates (> 1 kHz) are available, they tend to be expensive and cumbersome to use. These problems worsen as the device size increases (due to the higher drain currents required) and as the pulse width decreases (due to the requirements from the drain terminal pulse generator). These drawbacks can be eliminated by removing the need for the drain pulse generator.
Recognizing that, for the majority of devices, the drain-source resistance is high when the device is pinched off makes this elimination possible. Under this condition, the drain current will be insufficient to cause any significant heating when the drain-source voltage is maintained at a constant DC potential. Therefore, the pulsed IV characteristics of the device can be measured using the system shown in Figure 2 . For this case, only the gate terminal is pulsed and, since the power requirements from this generator are low, a variety of commercial units can be employed. In fact, the results presented in this article were acquired using a system implemented with readily available commercial equipment. This equipment includes a model HP8118A programmable pulse generator with DC offset control to define the gate-source terminal, a model HP6632A programmable DC power supply to define the drain-source terminal voltage and a model HP54510B programmable oscilloscope to measure the drain current and drain-source device terminal voltage VDS '. Clearly, the DC level on which the gate pulse is superimposed is kept sufficiently low to ensure that the device is off when the pulse is not present. The gate-source voltage is controlled by adjusting the pulse amplitude and drain current measured when the device is on.
The 10 W resistance in series with the drain terminal is used to measure the drain-source current. The measured voltage across the device drain-source terminal is used as the true VDS device voltage. Since all of the test conditions (pulse width, rise time, period and voltage) can be varied easily under program control, the system is considered a flexible research tool. Broadband matching networks also were determined to be necessary for narrow pulse width measurements.
The system described previously was employed on a moderately large sample of commercially available GaAs MESFET devices of varying number of fingers and gate widths employing a p -gate design. In all cases, the devices were of 0.5 m m gate length with a pinch-off voltage of –2 V. These devices were from the same wafer and were produced using an ion-implanted process on a liquid-encapsulated Czochralski substrate. Ti/Pt/Au gate metallization with silicon nitride passivation was used. Through-GaAs vias minimized the source inductance. Similar results to those presented here were observed with MESFET devices from other sources.
Since one of the tests consisted of measuring the device characteristics under different ambient temperatures, the devices were cut from the wafer and bonded onto a thick-film alumina carrier to make the handling easier. The devices in these carriers were used for all the described measurements. In addition, the measurement of the device temperature using liquid crystals4 under different bias and pulse conditions was performed. These two additional tests provided further information and enabled the effects of temperature and frequency to be observed independently. The measurement system (thick-film carrier and cabling) was kept the same for the various measurements on all the devices to enable accurate comparisons to be made. High frequency cabling and probes together with suitable calibration procedures ensured consistent results.
Since it is not possible to show the results obtained for all the devices, a four-finger, 225 m m finger gate-width device (900 mm total gate width) is used as the demonstrator. Figure 3 shows the output characteristics of this device when measured with a 10 ms-wide pulse using a variety of pulse repetition rates covering the 10 Hz to 5 kHz frequency range (where frequency dispersion effects are most relevant). It is also the frequency range (» 1 kHz) most commonly selected for pulsed measurements by authors to minimize the self-heating effect. The bias conditions have been chosen to illustrate the behavior of the device in the linear and saturation regions under low and high drain current conditions. The VGS + 0.8 V case, while not representative of normal FET behavior, was chosen to increase the likelihood of heating. For this case, the gate current was insufficient (38 mA) to contribute significantly to the heating of the device.
As can be seen, a large step change exists in the drain current between the 20 Hz and 1 kHz curves. This result is consistent with other independent experimental work that shows the device to be strongly dependent on frequency around the 100 to 500 Hz region,2 the frequency dependency of the traps resembling the characteristics of a lowpass filter. These data also confirm other experimental work, which shows the device to be frequency independent in the linear region. Although some self-heating may be occurring, the main reason for the large step change in ID in this frequency range is due to frequency dispersion. The liquid crystal data also showed no significant average self-heating for any of the test frequencies. The data presented also show little change in the linear region, which would not be the case if there was significant heating. Devices of different gate widths and number of fingers also exhibited an abrupt change in ID when the period was varied around this frequency area. A similar picture emerged when the 900 m m device was measured under the same conditions but with a 100 m s-wide pulse, as shown in Figure 4 . It is interesting to observe that a tenfold increase in the pulse width has little effect on the value of the drain current or its frequency dependency.
When the same measurements were repeated with a pulse width of 1 ms, the results were practically unchanged. This result suggests a rapid (< 1 m s) thermal time constant for the device so that the thermal differences between the three pulse width conditions are small and, therefore, the observed abrupt changes when the period is altered are due to frequency dispersion effects. This conclusion was further reinforced when the rise and fall times of the pulses were varied, producing a similar effect to a change in the pulse period due to the change in the frequency component of the signal.
It was determined for the majority of devices that by keeping the frequency component of the signal relatively constant as the pulse width was varied, pulse widths less than 100 ns are required to eliminate thermal heating. However, the thermal effect is relatively small compared to the frequency dispersion effect in this frequency range.
The same device was measured in a temperature chamber using identical test conditions to verify from yet a different angle that these measured changes in ID are not due to self-heating alone. Figure 5 shows the measured characteristics of the device at +80° and –75°C. These data show the same general behavior with frequency and confirm other independent work that shows the frequency dispersion problem is affected by temperature.2,5 At –75°C, the device is clearly less dispersive than at +80°C. A comparison of the characteristics at a repetition rate of 10 Hz for these two temperatures is shown in Figure 6 . In this case, the effects of temperature only are clearly seen. As expected, the characteristics in the linear and saturation regions are affected. Notice also the large temperature difference required to produce the changes in ID .
Figure 7 shows the pulsed and static characteristics of the device at normal ambient temperature. As expected, the DC curves closely follow the pulsed curves in the linear region. In the saturation region, the DC curves closely follow the pulsed curves measured at 10 Hz. This result makes sense considering the dispersion effect occurs at a higher frequency. The VGS + 0.8 V case, which is not a realistic bias mode for the device when operating as a FET, is the exception. This instance does illustrate the significant effect introduced by driving the gate-source junction diode hard into forward bias, but this aspect is beyond this scope of this article.
The results indicate that the differences between the static and pulsed characteristics of the device should not be interpreted as arising solely from thermal effects. Another important point to note is that the pulsed characteristics are obtained under large-signal conditions whereas the device model is based on the quasi-static assumption.
The same observations were determined to be true for the other devices considered. The number of fingers and gate width per finger were determined to affect the frequency area where the dispersion problem occurs and the magnitude of the changes. For example, the frequency area where the dispersion effect was most pronounced increased as the device size and number of fingers increased but the changes in ID decreased. This result is consistent with the better thermal behavior of the devices, which, in turn, influences their dispersion characteristics.
However, the interaction between frequency dispersion and temperature is difficult to quantify because it changes with the bias conditions, device design and processing conditions. A suitable general method for quantifying or modeling this phenomenon from a nonlinear point of view remains as elusive as ever. From a linear point of view, the modeling problem is not as severe where the bias point is fixed. However, this result is somewhat misleading since the variation in the device transconductance and output conductance due to frequency dispersion and processing tolerances makes it difficult to maintain the circuit’s performance consistently within specifications in many applications due to the inability to control or predict the device behavior with sufficient accuracy. As demonstrated previously (in the DC to 5 kHz range), the variation in the device performance is particularly difficult to predict.
The experimental results presented in this article indicate that the differences between the characteristics of the device acquired under static and pulsed conditions are due to thermal and frequency dispersion effects, and the frequency dependency of the device decreases as the ambient temperature of the device is reduced. However, the temperature change needs to be large to cause a significant effect. In addition, the number of fingers and the gate width size have a significant effect on the device’s thermal and, hence, dispersive characteristics.
The techniques described, while enabling the thermal and frequency dependency of the device to be observed, are based on large-signal measurements (a method that is incompatible with the quasi-static-based model). The derivation of the device IV model parameters from the pulsed characteristics also should take into account the fact that the parameters do not represent the true DC behavior of the device.
The author would like to thank Andreas Seiffert (final-year undergraduate student at the University of Rostock), Gianni Filippi (final-year undergraduate student at the University of Bologna) and Adrian Muscat (graduate student at the University of Bradford) for their help in performing the measurements described in this article. The GaAs MESFETs used for the measurements were manufactured using the GEC-Marconi Materials F20 process.
1. L. Selmi and B. Ricco, "Modeling Temperature Effects of the DC I-V Characteristics of GaAs MESFETs," IEEE Transactions, 1993, Edition 40, No. 2, pp. 273–277.
2. J. Rodriguez-Tellez, B.P. Stothard and M. Al-Daas, "Static-, Pulsed- and Frequency-dependent Current/Voltage Characteristics of GaAs FETs," IEE Proceedings of Circuits, Devices and Systems, 1996, Vol. 143, No. 3, pp. 129–133.
3. T. Fernandez, Y. Newport, J.M. Zamarillo, A. Mediavilla and A. Tazon, "High-speed Automated Pulsed Current/Voltage Measurement System," 23rd European Microwave Conference Proceedings, Madrid, Spain, 1993, pp. 494–496.
4. J. Rodriguez-Tellez, S. Loredo and R.W. Clarke, "Self-heating in GaAs FETs: A Problem?" Microwave Journal, September 1994, Vol. 37, No. 9, pp. 76–92.
J. Rodriguez-Tellez received his BSc, MPhil and PhD degrees in electrical engineering from Leeds and Bradford Universities in 1979, 1982 and 1985, respectively. From 1979 to 1983, he worked for Standard Telecommunications Laboratories, Harlow as a research engineer in high speed communication systems. Since 1983, Rodriguez-Tellez has worked at the University of Bradford as a senior lecturer and, from 1996, as a reader in solid-state devices. His research interests include device, interconnect and package modeling for microwave circuit applications.