Skip to content

Enhanced Thermal Management in GaN Superlattice Castellated Field Effect Transistors

FieldValue
JournalIEEE Electron Device Letters
Published2026-05-25
AuthorsPharyanshu Kachhawa, S. Y. Park, Jagori Raychaudhuri, Jonathen Malen, Tariq Jamil
DOI10.1109/led.2026.3696134

Combining a 2.2 µm AlN buffer with a 1.5 µm top-side polycrystalline diamond overcoat reduced the reported thermal resistance from 29 ± 1.5 to 11.9 ± 0.3 mm°C/W. That is a 2.44-fold reduction in the authors’ AlGaN/GaN multi-channel SLCFET. The authors project a matching 2.44-fold increase in RF power-density capability at fixed maximum channel temperature rise, but they did not report an RF output measurement for the modified devices.

Thermal resistance gains and power implications

Section titled “Thermal resistance gains and power implications”

Buffer replacement produced the first part of the gain. Device B reduced thermal resistance from 29 ± 1.5 to 17.2 ± 0.4 mm°C/W, which the authors report as a 1.66-fold reduction.

Adding the polycrystalline diamond overcoat reduced thermal resistance further to 11.9 ± 0.3 mm°C/W. The total reduction from Device A to Device C was 2.44-fold.

The measured average thermal conductivities were 223 W/m·K for AlN and 157 W/m·K for polycrystalline diamond. Despite its lower conductivity, the diamond created an additional near-junction lateral heat-spreading path.

Device architectures and integration sequence

Section titled “Device architectures and integration sequence”
DeviceBuffer layerTop-side layerThermal resistance, mm°C/W
A1.8 µm AlGaNNone29 ± 1.5
B2.2 µm AlNNone17.2 ± 0.4
C2.2 µm AlN1.5 µm polycrystalline diamond11.9 ± 0.3

The AlN buffer in Device B was deposited by MOCVD at 1250 °C. Device C received its diamond overcoat after device fabrication.

All three devices used six AlGaN/GaN superlattice channels and nominally identical layouts. Each device had six 100 µm gate fingers with a 50 µm gate-to-gate pitch. The wrap-around gate contacted the castellated channels through a SiN dielectric.

The authors measured individual layers using time-domain thermoreflectance and frequency-domain thermoreflectance. They analyzed separate epitaxial stacks one layer at a time to reduce parameter correlation.

LayerTTR, W/m·KFDTR, W/m·KWeighted average, W/m·K
AlGaN buffer38 (−5, +5.5)Not measured38 (−5, +5.5), TTR value used
AlN buffer217 (−21, +23)245 (−37.2, +49.2)223 (−18, +21)
GaN87.6 (−17.6, +19.1)91.4 (−16.8, +20.3)89 (−12, +14)
Polycrystalline diamond180 (−41, +44)153 (−17, +19)157 (−15, +18)

Time-domain thermoreflectance used a 532 nm pump laser with 3 ns pulses and a 491 nm continuous-wave probe. Frequency-domain thermoreflectance covered 100 kHz to 5 MHz with spot diameters of 6 ± 0.2 and 12 ± 0.4 µm.

Each fit treated the interface thermal boundary resistance and the top layer’s thermal conductivity as the two unknowns. The authors fitted an analytical heat-diffusion solution and used Monte Carlo analysis for 5% to 95% confidence bounds.

The measured AlN conductivity was five to six times that of the AlGaN buffer. The authors associate the high values with high crystalline quality and identify film thickness as the main limit from phonon boundary scattering.

Raman nanoparticle thermometry measured the gate-metal temperatures of Devices A and B. The authors drop-cast approximately 30 nm TiO₂ particles onto the gate surfaces. A 488 nm Ar+ laser and 0.5 NA objective produced a 0.5 µm focal spot.

For Device C, the transparent diamond itself served as the Raman thermometer. Its calibrated coefficient was 1.34 ± 0.04 × 10⁻² cm⁻¹/K. This measurement sampled a depth-averaged temperature through the 1.5 µm diamond layer.

The devices were measured on 15 mm × 5 mm coupons at 22 °C. The drain bias was fixed at 5 V, and power dissipation was controlled through the gate voltage.

A Silvaco ATLAS model represented one castellation and was calibrated against the measured current-voltage curve of Device A. The simulation assigned approximately three quarters of Joule heating to the channel beneath the gate and one quarter to the drain access region.

An ANSYS finite-element model covered the complete multi-finger device. It used weighted thermal conductivities from the two thermoreflectance methods. Simulated temperatures agreed with Raman measurements at the same locations within the stated uncertainties. The validated model then determined the buried peak channel temperatures used for the reported thermal resistances.

The paper places Device C at 11.9 ± 0.3 mm°C/W. Its benchmark table lists 12.51 ± 0.34 mm°C/W for a GaN SLCFET with a thin GaN buffer and 19.1 ± 0.7 mm°C/W for a standard four-channel SLCFET.

A previously reported GaN SLCFET with a polycrystalline diamond overcoat has a listed thermal resistance of 6 mm°C/W. These reference results differ in architecture, channel count, buffer design, and measurement method. They therefore do not isolate the effect of one design change.

The table also mixes measurement bases. These include gate resistance thermometry, scanning thermal microscopy, and peak-channel temperatures obtained through Raman-based simulation.

AlN thickness was studied through finite-element simulations from 100 nm to 2.2 µm. Thermal resistance decreased linearly across that range. The authors describe 2.2 µm as reasonably optimal within the tested interval because its conductivity was already about 80% of the highest reported bulk AlN value.

Thicker epitaxial AlN would require strain management. The authors also warn that low-conductivity transition layers could remove the benefit of a thicker high-conductivity buffer.

Diamond deposition introduces cost and tensile stress. The paper recommends low-temperature growth to manage stress. Device C had a threshold voltage of −12.0 V, compared with −10.9 V for Device B and −10.7 V for Device A. The authors attribute the negative shift to tensile strain that increases the channel two-dimensional electron-gas density.

For Device C, Raman thermometry measured a depth average through the diamond rather than the buried channel peak directly. The peak temperature and thermal resistance therefore depended on the validated finite-element model. The 2.44-fold RF power-density benefit remains a projection from the thermal result.

Implications for diamond material selection

Section titled “Implications for diamond material selection”

The result supports a two-path thermal design. The AlN buffer improves heat transport below the channels, while the diamond overcoat spreads heat laterally near the junction.

Diamond selection should use the measured thin-film conductivity rather than a bulk value. The paper reports an effective conductivity of 157 W/m·K for the 1.5 µm polycrystalline film and notes that boundary scattering lowers conductivity in thin layers.

The process must also control deposition stress and temperature to limit the threshold-voltage shift. Transparent diamond is required if the Raman depth-average thermometry method will be repeated. The paper does not report diamond grain size, surface roughness, or interface bond characterization, so those specifications cannot be inferred from the study.

6CCVD’s diamond materials catalog includes MPCVD-grown polycrystalline diamond from 0.1 to 500 µm thick, a range that includes the 1.5 µm top-side SLCFET overcoat. The paper identifies CVD but not the growth route for its overcoat, so teams should confirm process and strain compatibility; engineering support can support that review.

View Original Abstract

AlGaN/GaN multi-channel Super Lattice Castellated Field Effect Transistors (SLCFETs) have shown great potential for next generation RF applications. We demonstrate a 2.44× reduction in thermal resistance reduction in AlGaN/GaN multi-channel SLCFETs by replacing the standard AlGaN buffer with a high thermal conductivity AlN buffer layer and adding a poly-crystalline diamond overcoat layer. The thermal conductivity of each material in the device stack was measured using time-domain and frequency-domain thermoreflectance techniques, including 223+21−18W/m·K for 2.2 μm-thick AlN buffer versus 38+5.5−5W/m·K for the reference 1.8 μm AlGaN buffer. Device thermal resistance was reduced from 29 mm°C/W to 11.9 mm°C/W by combining the AlN buffer and poly-crystalline diamond overcoat layer, determined using Raman thermography together with 3-D thermal simulation. The corresponding 2.44× increase RF power density capability enabled by improved thermal management can unlock the next-generation ultra-high-power RF devices.