InP-based HBTs have demonstrated excellent frequency performance and extremely good current-handling capability. Research on InP-based HBTs has almost exclusively focused on NPN HBTs, since the electron velocity is several orders of magnitude higher than the hole velocity in InGaAs. PNP HBTs are also of interest, primarily for integration in circuits with NPN HBTs. Together, NPN and PNP HBTs can form a simple, efficient, and linear Class B power amplifier or output buffer. Such an AlGaAs/GaAs integrated NPN/PNP push-pull amplifier has demonstrated power performance at 2.5 GHz [4]. Multiple-stage amplifiers could be designed with alternating NPN and PNP stages, allowing for simple designs with a single power supply and direct coupling between stages. PNP HBTs could also be used as active loads and current sources for NPN amplifier stages, which would provide higher gain per stage, reduced power consumption in the load, and reduced wafer area consumed by passive resistors. Overall, the integration of PNP with NPN HBTs offers simpler circuits with fewer components and reduced power consumption.
While not as impressive as their NPN counterparts, the current state-of-the-art frequency performance for PNP HBTs are sufficient for X-band and higher performance. The best-published InAlAs/InGaAs PNP HBTs demonstrated
b = 170, fT = 14 GHz, and fmax = 22 GHz [2], while the best published InP/InGaAs PNP HBTs demonstrated b = 20, fT = 11 GHz, and fmax = 25 GHz [5]. In the AlGaAs/GaAs material system, results have been presented with b = 11, fT = 33 GHz, and fmax = 66 GHz [6]. The apparent higher performance of the GaAs-based HBTs is mostly due to more aggressive base designs and higher attainable doping in the emitter cap and subcollector.Two sets of PNP epilayers were grown on Fe-doped semi-insulating (001) InP by solid-source molecular beam epitaxy at The University of Michigan. The growth rate was 0.7
m m/h at 490º C, and an InGaAs/InAlAs superlattice was grown on the substrate to improve the material quality. The background doping for undoped InGaAs layers was below 5´ 1015 cm-3, and undoped InAlAs layers were semi-insulating.Both layer structures were almost identical [1] except for the doping of the 500-Å base layer: Wafer A had a uniform 5
´ 1018 cm-3 base doping, which resulted in a sheet resistance of 76 W /q and a contact resistivity to Ti/Pt/Au of 1.0´ 10-6 W -cm2. Wafer B had a linearly-graded base doping from 5´ 1018 cm-3 at the emitter end to 1´ 1018 cm-3 at the collector end, which resulted in a sheet resistance of 165 W /q and a contact resistivity to Ti/Pt/Au of 1.9´ 10-6 W -cm2. Both wafers were fabricated with the same process, which was based upon a high-performance NPN HBT process described elsewhere [7]. Nominal HBT emitter dimensions varied from 1-finger 1´ 10 m m2 to 10-finger 5´ 40 m m2, with typical dimensions of 2´ 10 m m2 and 5´ 10 m m2.The DC characteristics of Wafer A are described in detail elsewhere [1]. For a 5
´ 10 m m2 HBT, the breakdown voltage BVECO (at 10 A/cm2) was 5.6 V. From the Gummel plot, the ideality factors nB and nC were 1.60 and 1.00, respectively, and the maximum b was 12 at JC = 34.8 kA/cm2. The gain increased significantly (b > 20 and hfe > 30) at higher VEC. However, the gain compressed rapidly for JC > 50 kA/cm2, which was attributed to base push-out through simulations using similar collector doping.Both |h21|2 and Gmax were calculated from on-wafer common-emitter measurements of the S-parameters up to 25.5 GHz and extrapolated (when necessary) at 20 dB/decade to find fT and fmax, respectively. The optimal frequency performance was fT = 11 GHz and fmax = 31 GHz at VEC = 4.0 V and IC = 11.69 mA. Due to the short diffusion length of the holes, base push-out had a very significant effect on both gain and high-frequency performance at higher current densities. At similar bias conditions for a 1
´ 10 m m2 HBT from Wafer A, the best frequency performance was fT = 12 GHz and fmax = 35 GHz. This is the highest reported fmax for any InP-based PNP HBT.The characteristics of the 5
´ 10 m m2 HBT on Wafer B (see Fig. 1) were very similar to those on wafer A. The breakdown voltage BVECO (at 10 A/cm2) was 6.8 V, and the maximum b with VBC = 0 V was 4.2 at JC = 14.2 kA/cm2. Similar to Wafer A, the gain increased significantly (b > 20) at higher VEC.; however, the Early effect was more pronounced for Wafer B due to the lower base doping near the collector. The best frequency performance was fT = 13 GHz and fmax = 26 GHz at VEC = 4.5 V and IC = 6.47 mA (see Fig. 2). Comparing to Wafer A, the 15% increase in fT is due to a 25% decrease in t b caused by the drift electric field in the base. However, the lower average doping in the graded profile that created this drift field also increased RB by 85%, resulting in a 13% decrease in fmax.
Fig. 1: Forward I-V characteristics of 5
´ 10 m m2 HBT from Wafer B. IB = 0.3 mA/step.
Fig. 2: Frequency response of 5
´ 10 m m2 HBTs from Wafers A and BOn-wafer power characterization was performed at 10 GHz using a system developed in-house using FOCUS electromechanical tuners on both the source and the load. The 5
´ 10 m m2 HBT from Wafer A was biased using constant VEB and VEC while the source and load impedances were optimized for maximum gain at Pin = -20 dBm, resulting in G S = 0.740Ð -179° and G L = 0.596Ð 26° . The power characteristics at these impedances (Fig. 3) demonstrate a small-signal gain of 10.0 dBm, peak power-added efficiency of 24%, and a maximum output power density of 0.49 mW/m m2. These characteristics are very similar to InP-based NPN single HBTs fabricated with the same technology [8]: the NPN HBTs showed slightly higher gain (+1 dB) and efficiency (+5%), while the PNP HBTs produced more output power (+3 dBm) before saturating. Note that these NPN HBTs, which provide up to 1.37 mW/m m2, show the best reported power performance for InP-based NPN single HBTs and are only surpassed by double HBTs, which provide up to 3.6 mW/m m2 [9]. While power handling capability has not previously been reported for InP-based PNP HBTs, GaAs-based PNP HBTs have demonstrated output power up to 0.63 mW/m m2 [3], which is comparable to the HBTs presented in this work.
Fig. 3: Power characteristics of 5
´ 10 m m2 HBT from Wafer A under constant VEB bias with VEC = 4.0 V. Maximum Pout = 0.49 mW/m m2 and PAE = 24%.Load-pull characteristics of the PNP HBT at Pin = 3.17 dBm and at 1-dB of gain compression demonstrated that the optimal load impedances for linearity, gain, and efficiency are approximately the same, indicating that circuit designs need not choose impedances to trade off these characteristics. In contrast, previous studies of NPN HBTs [8] demonstrated that any single load impedance either reduces the gain by 4 dB, the maximum output power by 2 dB, or the efficiency by 14 % from the best-matched values.
Several power studies were performed at 10 GHz on the HBTs from both wafers to determine the influence of layer structure, emitter geometry, number of emitter fingers, and common-emitter versus common base configuration. All HBTs were biased under approximately the same conditions: constant VEC = 4.00 V (for common base, VBC = 3.20 V since VEB » 0.8 V), constant VEB adjusted for maximum small-signal microwave gain, and G S and G L optimized for maximum small-signal microwave gain. The only exception is one set of data which was taken at VBC = 4.00 V for a common-base 2-finger 5´ 10 m m2 HBT. Some of the resulting data is shown in Fig. 4, where the curves connect HBTs of the same emitter geometry but with 1, 2, 4, or 10 emitter fingers. Note that these results represent average measurements, while the previous graphs present the best measured results.
Fig. 4: Output power at 1-dB gain compression for HBTs biased at JCopt.
As mentioned previously, for each HBT VEB was adjusted for maximum small-signal microwave gain, resulting in the optimal collector current density JCopt. For Wafer A, JCopt decreased from 2.5´ 104 to 1.5´ 104 A/cm2 for emitter areas from 20 to 500 m m2, respectively. Other than this slight area dependence, JCopt did not vary with emitter geometry. This indicates that the variation is not due to current crowding or surface recombination but rather is due to self-heating generated from the total power dissipated in the HBT. Also, the JCopt for HBTs on Wafer B was approximately one-half that for HBTs on Wafer A, probably due to a slightly lower collector doping. Since gain decreases dramatically after the onset of base push-out, JCopt must occur at a lower current density than base push-out. Lower collector doping causes base push-out at lower current levels, which lowers JCopt. The decrease in gain above JCopt also limits the large-signal current swing for Wafer B, which leads to worse power performance.
The DC large-signal current gain b for HBTs biased at JCopt was close to the maximum current gain for these HBTs, and it varied from 19.5 to 15.5. In general, b decreased with increasing emitter area but was independent of the emitter geometry, which can again be attributed to self-heating. For a fixed HBT size, Wafer B demonstrated greater b than Wafer A (between 5% and 45% greater, depending on bias). In Wafer B, the electric field generated by the doping gradient in the base pushes the injected holes towards the collector more rapidly, which reduces the number of holes that recombine. Simulations indicated that the dominant component of the base current is Auger recombination in the neutral base [1], so any decrease in base recombination results in significantly increased gain.
Variations in power performance can be seen in the output power at 1-dB of gain compression (Po-1dB, in Fig. 4) and in the maximum power-added efficiency (PAE) for HBTs biased at JCopt. Note that the tuning range of the load and source tuners limited the gain of the small and medium-to-large HBTs, respectively, and the maximum PAE typically occurred between the 2.5 and 3 dB gain compression points. In general, the Po-1dB scaled with the area of the HBT. Since the collector current also scaled with the HBT area, the PAE versus area is dominated by the HBT gain characteristics. The maximum PAE for common-emitter HBTs was 20% for the 2-finger 5´ 20 m m2 HBT from Wafer A. The smaller bias current density and the 1- to 3-dB lower microwave gain of Wafer B contributed to its 4-dB lower Po-1dB and 7% lower PAE.
At the same VEC bias, the Po-1dB is significantly (2 to 5 dB) higher for common-emitter HBTs than for common-base HBTs. This effect is primarily due to the limited output voltage swing imposed by VBC = 3.2 V for the common-base HBTs. When VBC was increased to 4.0 V (labeled "4 V" in the figure’s legend), the Po-1dB increased to match the value for common-emitter HBTs. However, at either bias point, the microwave gain of the common-base HBTs is 3 to 6 dB higher than for the common-emitter HBTs. Since the PAE is essentially determined by the microwave gain and the gain compression characteristics, this leads to similar efficiency (17-18%) for both configurations when VBC = 3.2 V and higher efficiency (22%) for common-base when VBC = 4.0 V. In general, the PAE of the common-emitter HBTs was limited by low gain, and the PAE of the common-base HBTs was limited by early output power saturation. The PAE should be able to be increased by increasing VBC in the common-base configuration. The values presented here are close to the PAE reported for GaAs-based PNP HBTs – 25% for common-base and 33% for common-emitter [3].
The characteristics presented above give some insight to the limitations of power performance in PNP InP-based HBTs. In general, the power and efficiency performance is dominated by the microwave gain (or equivalently fmax) and the bias conditions. For HBT designs that have their optimal fmax at higher JC, the available output signal swing is larger, which increases Po-1dB. This implies that an increased collector doping (to delay base push-out) and an increased breakdown voltage are key elements to power performance. The higher gain and breakdown voltage for common-base HBTs clearly give a performance benefit over common-emitter HBTs. Typically, BVBC0 for common-base HBTs is twice that of BVECO for common-emitter HBTs, allowing for much higher bias voltages in the common-base configuration. When combining higher bias VBC with its higher gain at frequencies close to fT, common-base HBTs should be able to produce more output power at higher efficiencies than was presented.
Although the doping gradient in the base of Wafer B decreased the base transit time, the base resistance also increased, resulting in lower fmax, less high-frequency gain, and worse power performance than Wafer A. Increasing the base doping in Wafer B should decrease the base resistance while maintaining the improved fT, which would increase fmax and the power performance.
Finally, since the high-frequency characteristics of these PNP HBTs were dominated by transit times rather than by parasitic charging times, their intrinsic power performance was only slightly dependent on device geometry or area. The lack of any significant gain or power dependence on the perimeter-to-area ratio indicates that surface recombination is not a major factor in these HBTs. However, the total device area still had a large influence on the input and output impedance of the HBT, making it difficult to fully match very small or very large HBTs.
Device modeling and analysis also suggested several potential improvements to the PNP HBT design for increased high-frequency power performance. In general, fT was dominated by the base transit time, so either a doping gradient or a compositional gradient in the base can create an electric field to reduce the base transit time. Drift-diffusion simulations [1] indicated that removing the emitter-base spacer would reduce the base transit time while increasing the DC gain, since a significant number of holes recombine in the spacer. These modifications will cause an associated increase in fmax and in the gain at 10 GHz, which should significantly improve the X-band power performance and efficiency. Slightly increasing the base doping and thickening the collector will reduce RBCBC, which in turn will also increase fmax. Finally, the power-added efficiency was partially limited by the parasitic emitter and collector resistances, indicating that a thinner emitter cap and a thicker subcollector should be used.
