GaN technology, widely utilized in terrestrial power amps, is gaining traction for space applications due to its high power densities and low losses at relevant frequencies. GaN epitaxy on SiC offers excellent thermal behavior but at a high cost, limiting its scalability for mass production. Conversely, GaN on Si substrate, though cost-effective, suffers from thermal issues, particularly critical in space applications where controlled temperature and voltage parameters are mandatory. Consequently, traditional power amplifier design paradigms must be re-evaluated to strike a balance between output power and maximum allowable channel temperature, emphasizing the need for reliable and accurate thermal models in this field.

Technology and Models

The D01GH 100 nm gate-length GaN/Si HEMT process design kit (PDK) by OMMIC offers high-frequency capability suitable for Ka-band applications. With a high cutoff frequency, careful load selection was crucial due to derating, especially regarding maximum junction temperature (Tj). ADS simulations indicate limited achievable output power, with 8x50 µm and 8x75 µm device sizes exceeding nominal power but suffering from elevated junction temperatures. At Tj = 200 ◦C, a maximum output power density of 4 W/mm was achievable, necessitating a minimum periphery of 3.2 mm for 12.6 W output. At derated Tj = 160 ◦C, the achievable output power density decreases to 3 W/mm, requiring a larger periphery of 4.2 mm for the same output power.

The OMMIC PDK offers two thermal models for its active devices: a simplified 1D heat flow model (as shown in the figure) and a more complex 3D thermal stack for FEM-based simulations. When it comes to accuracy, the thermal stack model was more reliable, but its computational demands limit its practical use in post-design verification. The simplified model, based on thermal resistance, was preferred for design and optimization due to its computational efficiency.

Figure 1: Simplified thermal model

It uses a non-linear thermal conductivity with hyperbolic dependence on temperature, facilitating adiabatic heat flow from device junctions to the heat sink. This model accommodates substrates with varying thermal properties, provided they share the same functional temperature dependence. Kirchhoff transformation remains valid across layers in such cases.

· Thermal Model Verification: Validation of the simplified thermal model was achieved through both 3D electro-thermal simulations and experimental Raman characterization. The Raman analysis focused on monitoring the Si layer beneath the GaN layer in the channel area to determine temperature changes. Initially, a calibration law was established without device bias to correlate Raman shift with temperature. Subsequently, varying the bias voltage altered the power dissipation while maintaining the base-plate temperature at 80°C during the dissipation phase. By using 3D thermal simulation results, the measured Si layer temperature was linked to the GaN channel's junction temperature.

Figure 2: Comparing electro-thermal simulations, simplified model predictions, and Raman measurement results

As shown in above figure 2, a strong agreement despite slightly higher temperatures was predicted by the simplified model, demonstrating its efficiency for rapid junction temperature estimation in MMIC design.

· Crosstalk Effect: In Ka-band power amplifiers (PAs), combining more devices in a small chip area increases device heating from neighboring devices (crosstalk). Adjusting R_th,REF compensates for this, typically requiring 3D electro-thermal simulations for accuracy.

Figure 3: Simulations for 4 parallel combined 8x75 µm devices

For instance, simulations for 4 parallel 8x75 µm devices show minimal heat spread and via holes cooling the source terminals, nearly isolating devices thermally. Only a 2.2% increase in R_th,REF effectively addresses thermal crosstalk. These findings are illustrated in Figure 3.

Designing the Thermal Aware for 10W KA-BAND PA

Figure 2 illustrates the junction temperature's significance, stopping dissipated power at roughly 2.2 W/mm at TBP = 80 ◦C. In space-based and related applications, load termination selection balances output power and junction temperature, diverging from traditional PA designs. Load-pull maps help find the optimal compromise among power gain, efficiency, and output mismatch.

Figure 4: Load-pull results for VDS = 11.25 V, with a 1 dB compression.

Figure 4 presents a comparison between the load-pull simulation results for the 8x50 µm and 8x75 µm devices. It shows how the 8x75 µm was necessary to meet the Tj = 160 ℃ limit for space applications, while 8x50 µm suits classical PA designs. Junction temperature contours are calculated from CW dissipated power, factoring in 8 parallel devices. These observations guide load selection to maximize performance despite thermal constraints.

Conclusion

An experiment was conducted to explore the thermal-aware design approach for MMIC power amplifiers (PAs) targeting space applications, particularly focusing on GaN/Si technology. The study revealed that space thermal requirements significantly constrain device performance in GaN/Si amplifiers, necessitating larger peripheries to achieve the same targeted output power. However, through a thermal load-pull approach, it was demonstrated that GaN/Si technology can still achieve competitive results at Ka-band frequencies even when it comes to the space-related technical rules.