High-Voltage GaN-on-Si Devices Deliver High Power

Migrating existing +28-VDC processes to allow for +48-VDC operation requires redesigning the device structure to improve reliability due to the increased electric field. In the X-Y direction, the underlying substrate, epitaxial structure, and channel characteristics remain largely the same for higher-voltage operation, with increased gate-to-drain spacing needed to raise the breakdown voltage. In the Z-direction (vertically), the gate oxide is thickened and changes to the gate metal allow for both higher reliability and increased standoff voltages.

Of greater importance to RF/microwave amplifier designers, Table 1 shows that device output capacitance (C_OUT) scales by device size. Because of the increased power density, the resulting C_OUT) for a given power level is reduced. This reduction in output capacitance is the largest differentiator for higher-voltage operation since the resulting change in load impedance offers a major advantage in device operation.

Perhaps the biggest challenge that designers face at higher power levels and frequencies is matching the low output impedances of larger devices. A parallel resistor-capacitor (R-C) circuit is a good model for the transistor output, corresponding to the equivalent R with C_OUT as specified in Table 1. Given the supply voltage (V) and desired power (P) of an ideal amplifier, there is a simple equation that estimates the output resistance (R_OUT):

P ≈ V²/2R_OUT   (1)                                                                                                                        

Figure 1 plots R_OUT for both +28- and +48-VDC device operation across a typical output-power range. It shows how an amplifier designed for +28-VDC operation is well optimized for output-power levels of 8 to 10 W, requiring minimal impedance transformation for 50-Ω systems. It becomes obvious from Fig. 1 how +48-VDC operation is capable of nearly 25 W output power with minimal impedance matching needed for the same 50-Ω system compared to +28-VDC operation.

In addition to increased R_OUT, +48-VDC operation also provides further benefits that can ease matching complexity. All things being equal, the usable bandwidth for an amplifier is fundamentally limited by the metrics of impedance transformation ratio and device quality factor (Q):

Impedance transformation ratio = Re(Z_DEVICE)/50 Ω Device Q, where Q ≈ Im(Z_DEVICE)/Re(Z_DEVICE) and Z_DEVICE ≈ R_OUT + 1/jωC_OUT   (2)

The impact of these limitations varies with frequency, but there are general trends. With +28-VDC devices, the transformation ratio becomes the limiting factor—i.e., low R_OUT tends to limit the available bandwidth. With +48-VDC devices, smaller C_OUT and larger R_OUT mitigate both factors, and device Q tends to set the limit. Parameter C_OUT limits the usable maximum frequency in all cases.

For a realistic comparison, Table 2 shows fundamental properties for two similar 100-W GaN-on-Si devices, the +28-VDC model NPT1010 and +48-VDC model NPT2010 devices. Both devices use a 100-W die with no output impedance prematching and the same substrate thickness, die attach technique, and metal air cavity package. Among these measures, two differences stand out. The optimum load impedance of the +48-VDC device is significantly higher with the ratio of the real part at 3:1, consistent with theory.

The second obvious difference is thermal resistance (R_θJC). This is the exception in the comparison of devices, whereby the smaller +48-VDC die can push performance in the wrong direction on account of more power being produced in a smaller area. Both devices are designed to operate with adequate margin to achieve rated output power levels below maximum rated junction temperatures for a given ambient condition. A significant thermal limitation for these high-power devices is the AC360 air-cavity flange package. The limitations of the package serve as a reminder of the need for more enhanced thermal packaging techniques to further enhance +48-VDC device operation.

In addition to enabling broader bandwidths, higher output impedances also enable alternative matching topologies, some of which were previously impractical. One of these, for example, is a broadband amplifier targeting 90 W output power from 100 MHz to 1 GHz. A ferrite-based impedance transformer is the best choice for a decade-plus bandwidth in the VHF/UHF bands; these topologies are ideally suited for N²:1 impedance ratios such as 1:1, 4:1, and 9:1.

From Eq. 1, an amplifier based on +28-VDC devices would require a load resistance of about 4.5 Ω, and 9:1 would be the ideal ratio for a transformation to 50 Ω. But a 9:1 impedance transformer is difficult to realize. The higher permeability materials needed to enable the low-frequency transformation become lossy at high frequencies, preventing full band coverage.¹ Realistically, a high-power ferrite transformer with 9:1 transformation ratio will suffer from high loss and struggle to work above a few hundred MHz.

With a +48-VDC drain supply, the load resistance is much higher, nearly 12.5 Ω, and a lower-ratio 4:1 transformer is better suited to make the impedance transformation to 50 Ω. This lower transformer ratio is ideal for frequency coverage from 100 MHz to 1 GHz. With a nominal 100-W device, the simple and low-cost design shown in Fig. 2 delivers more than 80 W output power over the entire frequency band, achieving 50 to 70% efficiency with a single device. Figure 3 shows the efficiency performance for this design.

Power amplifiers (PAs) for wireless infrastructure applications with wideband-code-division-multiple-access (WCDMA), Long Term Evolution (LTE), orthogonal-frequency-division-multiplex OFDM, and other high-peak-to-average (PAR) waveforms face a real challenge. Achieving even modest amplifier efficiency is difficult when the RF amplifier is appropriately sized for peak power levels that are 8 to 10 dB higher than the average power level.

Envelope tracking (ET) is one of several effective techniques to address this issue. ET raises and lowers the drain voltage to follow the instantaneous peak amplitude of the device, effectively adjusting the compression point of the amplifier in real time to increase amplifier efficiency (Fig. 5). A peak-to-average ratio (PAR) of 10 dB, or 10×, implies a voltage ratio of approximately 3:1. This is a typical ratio seen in commercial systems.

For ET with +28-VDC devices, the drain supply typically varies between +10 and +30 VDC. This presents a problem for both GaN and LDMOS devices at the low end of this range because small-signal gain becomes nonlinear as a function of supply voltage. Above about +12 VDC, the gain increases linearly with voltage, but below +10 to +12 VDC, the gain enters a nonlinear region and can drop several dB below its nominal level at higher voltages. The statistical nature of a high-PAR signal implies an ET PA operates in this low-voltage region most of its time, thus encountering this unwelcome nonlinear gain condition. With a +48-VDC device, a typical ET system varies the drain between +20 and +60 VDC and the minimum voltage remains well above the onset of the nonlinear gain condition. 

A +48-VDC GaN device provides an additional benefit over an LDMOS device because of the reduced C_OUT of the device. As mentioned previously, the higher C_OUT of LDMOS devices mandates internal pre-matching on the output to provide more friendly terminal impedances. The chosen topology for this output pre-match is usually a shunt-L match in the RF path. The shunt-L is designed to resonate with C_OUT to improve the terminal impedance.

All active transistor devices have terminal capacitances that vary versus applied voltage. When the drain supply voltage is modulated to follow the signal envelope, the drain-to-source capacitance (C_DS) also varies and the shunt inductance (L) resonance is impacted, changing the terminal impedance of the device in real-time with the drain voltage. This leads to a nonoptimal impedance match for ET applications.⁵

In addition, because the shunt-L match requires a large shunt DC-blocking capacitor, there is additional strain on the envelope modulator design (Fig. 5), which is already challenged to swing large voltages with high current capability at high data bandwidths. The large blocking capacitor prevents the envelope modulator from being able to raise and lower the supply voltage in time with the modulation bandwidth. The low C_OUT of GaN devices means this matching topology is not needed, making +48-VDC GaN devices inherently optimized for ET applications.

A transition to a +48-VDC drain supply may raise concerns given the higher voltage. Although there may be thoughts that a higher voltage causes additional stress and lower reliability, this has not been found to be the case. Between +28 and +48 VDC, no difference has been found in DC reliability studies, with both voltage optimized technologies achieving 1 × 10⁶ hours mean time before failure (MTBF) at a junction temperature of +200°C. It is important for RF system design engineers to realize the connection between junction temperature and device reliability. The key to reliable design is maintaining device junction temperature below specified limits. Designers need to monitor ambient temperature and operating conditions (power dissipation) to guarantee reliable operation.⁶

Compared to +28-VDC devices, +48-VDC devices are physically smaller for a given RF power rating and the advantages of performance come primarily from this difference. This flows down into every aspect of the device, from the difficulty and performance of the impedance match, to the usable bandwidth and relative achievable performance.

Legacy commercial and military RF systems are well rooted in the use of +28-VDC power-amplifying devices. While LDMOS and GaN device suppliers will continue to support and increase +28-VDC offerings, new systems with the flexibility to do so should consider higher-voltage devices. The +48-VDC GaN-on-Si devices offer compelling advantages for those willing to consider this option.

Raymond A. Baker, Field Applications Engineer

Walter H. Nagy, Principal Engineer

David W. Runton, Vice President, Engineering

Robert A. Sadler, Principal RF Applications Engineer

Nitronex Corp., 523 Davis Dr., Ste. 500, Morrisville, NC 27560; (919) 807-9100, FAX: (919) 472-0692.