IEEE Copyright

POWER PERFORMANCE OF InGaAs/InP SINGLE HBTs

D. Sawdai, J-O Plouchart, D. Pavlidis, A. Samelis, and K. Hong

Solid State Electronics Laboratory, Department of Electrical Engineering and Computer Science

The University of Michigan, Ann Arbor, Michigan 48109-2122, U.S.A.

Phone: (313) 747-1778, Fax: (313) 763-9324, e-mail: dsawdai@engin.umich.edu

Impressive microwave results have been published for both Single- (SHBT) and Double- (DHBT) Heterostructure Bipolar Transistors based upon the InP material system. InP-based SHBTs have been reported to have excellent high-frequency performances such as unity current-gain frequency (f_T) of 200 GHz (1) and maximum oscillation frequency (f_max) of 236 GHz (2). The best results for InP-based DHBTs include f_T of 160 GHz (3) and f_max of 267 GHz (4).

Most data published on the power performance of InP-based HBTs so far has concentrated on DHBTs. Due to the wide-bandgap collector, InP DHBTs typically have a higher breakdown voltage and hence can produce higher levels of output power. Recent reports on InP-based DHBTs include 5.3 W/mm of output power at a power-added efficiency (PAE) of 62% from ten 2×30-m m² emitters under class B operation (5).

Typically, the microwave power characteristics of InP/InGaAs SHBTs has not been addressed due to their relatively inferior DC characteristics when compared to DHBTs, which implies early breakdown and thus limited power performance. On the other hand, SHBTs are very attractive for higher frequency applications due to the absence of the heterojunction spike at the base-collector (B-C) interface. Moreover, the homojunction B-C structure offers direct compatibility for HBT integration with PIN diodes, since the latter can be realized by using the B-C-subcollector region. Such integration is needed not only for OEICs but also for MMICs with switching capabilities, as recently reported by the authors (6). This paper reports for the first time a systematic investigation of InP-based SHBT characteristics and demonstrates their suitability for power applications.

1. Device Design and Fabrication

The layer structure for InP/InGaAs SHBTs used in this investigation is shown in Table 1. Based on separate studies of similar designs, we employed here an abrupt InP/InGaAs emitter-base junction, which shows good gain and performance. The 600-Å base provided a good trade-off between base transit time and base resistance. Finally, the moderate 5000-Å n^- collector allowed for reduced electric fields in the collector, which increased the device breakdown voltage, without seriously degrading the collector transit time. The moderately large collector size also helped to reduce the C-B capacitance, further enhancing power performance.

Layer Name	Material	Thick (Å)	Type	Doping (cm-3)
Emitter Cap	InGaAs	2000	n+	2 × 1019
	InP	700	N+	2 × 1019
Emitter	InP	1500	N	5 × 1017
Spacer	InGaAs	100	i	-
Base	InGaAs	600	p+	1.5 × 1019
Collector	InGaAs	5000	n-	5 × 1016
Subcollector	InGaAs	5000	n+	2 × 1019
Buffer	InP	2000	i	-
Substrate	InP	-	LEC InP:Fe

Table 1: InP/InGaAs single HBT layer structure. All InGaAs layers are In_0.53Ga_0.47As.

The epitaxial layers used here were grown at the University of Michigan using an EMCORE low-pressure Metalorganic Chemical Vapor Deposition system. 1% disilane and DEZn were used for n and p doping, respectively. The purging time between base and emitter growth was optimized to minimize quaternary formations, which can be detrimental to performance and can pose fabrication difficulties. A 100-Å spacer was used in order to minimize Zn diffusion-related difficulties. The base was grown at 530 °C, and all other layers were grown at 570 °C. These temperatures provided the best compromise of layer electrical characteristics, layer morphology, and reduced growth interruption at critical heterointerfaces.

The devices were fabricated using our standard self-aligned HBT process, which produces a 0.2-µm base contact-to-emitter separation. The nominal emitter dimension was 2×10 m m², which allowed for a moderate-sized active area while minimizing current-crowding effects. The base-collector capacitance was minimized by using the base contact as an etch mask for the base mesa. Typical total parasitic-to-active B-C junction areas for these devices were 3:1, allowing reasonably good high-frequency performance. Electroplated gold airbridges were used to connect the HBTs to the coplanar interconnect pads used for on-wafer probing. The semiconductor junctions under the airbridge pads were isolated from the active device and from each other by isolation etches to reduce parasitic capacitances. All reported data are for unthinned wafers without any device mounting. The reported performance can consequently be improved by using individual, discrete HBTs of the same technology.

2. Small-Signal and Power Characterization

First results of the characteristics of these HBTs were reported recently by the authors (7). The devices had a base sheet resistance of 1150 W /q and a DC current gain of 34. The 5000-Å/5×10¹⁶-cm^-3 InGaAs collector produced high breakdown voltages BV_ce exceeding 7.0 V at I_c=10 µA. High collector current densities of more than 1.4×10⁵ A/cm² were also achieved with this design. The devices had f_T of 95 to 98 GHz and f_max of 55 to 67 GHz. The latter were limited by the B-C capacitance due to incomplete depletion of the collector region. This paper provides further details of their electrical performance with emphasis on their power features.

On-wafer power characterization was performed using a system developed in-house using FOCUS electromechanical tuners. The input RF signal was monitored during actual device measurement with a -20 dB coupler and a power meter. A bias tee was used between the input coupler and the source tuner. Coplanar Cascade probes were employed to contact the wafer. The output from the HBT fed into the load tuner, through another bias tee, and then to the output power meter. Due to various losses in the test setup and to limitations of the tuners, the effective matching range of the tuners was limited to approximately |G | < 0.75. Although limiting the measurement range of the HBTs, this test setup was suitable for evaluating the InP/InGaAs HBTs of this study. The entire system was computer controlled for measurement automation, and special software was developed for monitoring power-related characteristics such as power-added efficiency and intermodulation.

To extract their maximum power potential, we investigated possible ways of biasing the devices (Fig. 1). The 4f×(2×20) HBT was biased at the same DC operating point before applying either RF input signal. Then, the RF input was applied at various power levels, maintaining either V_be or I_b constant for the DC bias point. The source and load were matched to the HBT at each power level. By biasing the HBT with constant V_be rather than constant I_b, stronger self-bias and considerable I_c increase occurred as P_in increased, which led to a 75% improvement in output power at higher input power levels without any consequence on device stability.



Fig. 1: Power characteristics of two 4-finger HBTs at 10 GHz. The load and source impedances were optimized at each point. The 4f×(2×10) HBT produced an output power of 2.74 W/mm. The 4f×(2×20) HBT demonstrated the reduction of gain compression by using a constant V_be rather than a constant I_b bias.

Several load-pull measurements were performed on a 4-finger (2×20 m m² each) HBT at 10 GHz. The source impedance was matched for small signals before the measurement and held constant throughout. For small input signals, the contours of constant power and gain were concentric circles and were relatively independent of constant V_be or constant I_b bias. Another load-pull measurement was made with constant V_be at higher input power levels (Fig. 2). For this plot, at any given load impedance, the gain was initially measured at P_in = -10 dBm. Then P_in was increased up to the point where the gain at that impedance compressed (degraded) by 2.5 dB from the small-signal gain at that impedance. The plotted data was taken at this 2.5-dB compression point. The input power causing this 2.5-dB compression varied with the load impedance due the impedance dependence of the self-biasing effect. This variation of input power caused the contours of constant gain and constant output power to not coincide, distorting the contours of constant output power to ellipse-like shapes. The maximum output power was 16.9 dBm at an input power of 8.9 dBm.



Fig. 2: Load-pull characteristics (dashed lines = output power; solid lines = associated gain) of 4f×(2×20) HBT biased at 2.5 dB of gain compression at each load impedance. The DC bias point is V_be = 0.83 V and V_ce = 1.7 V. The source impedance was held constant at the optimized value Z_S = 18.6 - j2.3 W . Maximized output power (16.9 dBm) and associated gain (8.0 dB) were at Z_L = 13.9 + j16.3 W .

A study of the influence of both emitter geometry and number of emitter fingers on power and overall device characteristics was performed. Multiple emitter fingers were used to generate comparable total emitter areas and larger output power. The geometry-related power measurements were all made at 8 GHz, a DC bias point of J_c »  2.5×10⁴ A/cm², and a constant V_be bias as the input power was varied. The DC large-signal gain b and the small-signal impedance-matched gain at 8 GHz are shown in Fig. 3. While the DC gain is relatively invariant to the emitter area, the microwave gain significantly decreases as the total emitter area increases. Details of these characteristics are discussed in the subsequent section.



Fig. 3: (a) Large-signal DC gain and (b) small-signal microwave gain at 8 GHz as a function of emitter geometry and area. The source and load impedances were matched for maximum microwave gain.

Subsequently, each device was impedance-matched at the source and load for high-power operation using constant V_be bias. In general, as more fingers were added in parallel, the total HBT input and output impedances were reduced. Unfortunately, the input impedance for the 10-finger HBTs was too small to be matched by our tuners, so these devices were not compared in this part of the study. Using these fixed impedances matched at high-power for each device, measurements were taken at 1-dB of gain compression from the small-signal gain at the same impedances (Figs. 4-6).

3. DC and Microwave Power Results

The DC, small-signal, and power characteristics were studied as a function of device geometry. Devices with variable emitter-finger size (2×10, 2×20, and 3×10 µm²) and number of fingers (1, 2, 4, and 10) were analyzed. Considering the single heterojunction design, these devices offered a high output power level of 2.74 W/mm for a 4-finger (2×10 µm² each) HBT and a high power-added efficiency (PAE) of 43% for a 4-finger (2×20 µm² each) HBT at 10 GHz, both biased by constant V_be under class A operation with optimally-matched source and load impedances.

As previously discussed from Fig. 3, the microwave small-signal gain showed a larger geometry dependence and degraded much more quickly than the DC large-signal gain did as the emitter area increased. This indicated greater microwave parasitics for the larger HBTs. The parasitics can be attributed to the non-uniform current flow and possibly distributed effects in very large devices.

The output power at 1-dB of gain compression is shown in Fig. 4 to scale linearly from 1 to 2 emitter fingers; however, it started to degrade from this linear scaling at 4 emitter fingers. This degradation of power density at the 1-dB compression point is most likely due to thermal and/or current crowding effects. Future work is intended to explain these trends. Despite the observed degradation, devices with 4 emitter fingers still showed excellent power performance, as commented above.



Fig. 4: Output power at the 1-dB gain compression point at 8 GHz for varying emitter geometries. The source and load impedances were matched for maximum output power.

For a fixed finger length of 10 µm, Fig. 5 indicates that an increase of width from 2 to 3 µm resulted in reduced output power and reduced associated gain at the 1dB-gain-compression point for the same total emitter areas (about -1dB gain for A_E = 40 µm²). However, this also reduced self-biasing and thus increased the associated PAE from 25% to 30% at the same A_E (Fig. 6). Thus, small finger widths are more suitable for maximum power and gain but offer smaller PAE. However, when the comparison was made at larger emitter areas, i.e. 80 µm², smaller finger widths resulted in no power advantage but caused increased self-biasing and reduced PAE as before. This effect was probably due to uneven current distribution as the number of emitter fingers increased.



Fig. 5: Output power and associated gain at the 1-dB gain compression point at 8 GHz for varying emitter geometries. The source and load impedances were matched for maximum output power.



Fig. 6: Self-biased collector current and power-added efficiency at the 1-dB gain compression point at 8 GHz for varying emitter geometries. The source and load impedances were matched for maximum output power.

For a fixed finger width of 2 µm, an increase of the length from 10 to 20 µm resulted in larger power, reduced gain, increased self-bias, and almost equivalent PAE at the same large A_E of 80 µm². These results indicate that longer emitter fingers can be used for increased power-handling capability.

4. Conclusion