Double duty diamond

Equipping a GaN HEMT with a topside diamond film allows it runs cooler and more reliably, thanks to superior heat spreading and effective passivation.

BY JEONG-KYU KIM AND SRABANTI CHOWDHURY FROM STANFORD UNIVERSITY

At the heart of many RF and microwave power systems are GaN HEMTs. These devices offer a rare combination of high output power, strong efficiency, and operation at very high frequency. It’s a set of attributes that ensures that this class of transistor is essential for a number of applications, ranging from radar and satellite communications to advanced defence systems and next-generation wireless infrastructure. Yet even these high-performance HEMTs face a stubborn internal limit. As operating power increases, self-heating inside the channel rises rapidly to become a critical factor that restricts the maximum achievable output power, as well as long-term device reliability and operational stability.

An increasing channel temperature introduces several penalties, all appearing at once. They include a degradation in carrier transport as mobility drops, as well as an increasing on-resistance and a falling in efficiency. Due to this, performance that looks promising on paper maybe much harder to sustain in real operating conditions. At the same time, elevated temperature accelerates heat-driven degradation processes in contacts, barrier layers, and gate regions, causing devices to wear out faster and shorten their usable lifetime. The upshot is that designers of practical RF power amplifiers are often forced to step back from the theoretical performance limit, simply to maintain the junction temperature within a safe operating window. This means that thermal constraints frequently end up defining the real ceiling on output power, rather than the intrinsic electrical potential of GaN.

A second limitation appears at the device surface. Here trap states may introduce charge lag and distort local electric fields during high-frequency, large-signal operation. This quashes the maximum available drive current and makes the transistor behave less predictably. Often described as DC-RF dispersion, this phenomenon holds back the usable RF output and complicates device optimisation.

Unfortunately, thermal and trapping effects are closely coupled. Higher temperatures intensify trapping behaviour, and trapping can reshape local electric fields, while altering where power is dissipated. Working in menacing unison, these effects create an electro-thermal reliability challenge that’s more severe with increases in power density and operating frequency.

Engineers have traditionally treated this issue as two separate problems. They addressed thermal limits with superior substrates, better packaging, and external heat spreaders, and improved electrical stability at the device-unit level through the introduction of dielectric passivation layers. But that disparate methodology, shaping GaN design for many years, prevents these devices from delivering their full potential. It would be more effective to address heat and surface effects together, close to where they originate.

At Stanford University, we pursue this unified approach, leveraging diamond, a material that combines exceptionally high thermal conductivity with strong electrical insulation. By integrating a low-temperature-grown polycrystalline diamond film directly on top of fully fabricated AlGaN/GaN HEMTs, we are placing the world’s most conductive material incredibly close to the channel hot-spot region, where heat is generated most intensely. In this topside position, diamond spreads heat efficiently away from the active region without creating an electrical conduction path, while serving as a surface passivation layer. This dual role improves thermal and electrical performance.

Heat bottleneck
To suppress surface trapping and mitigate DC-RF dispersion, Ga-polar GaN HEMTs feature surface passivation. The most widely adopted solution is deposition of a SiN_x layer – providing effective electrical passivation, this is now standard practice across industry. However, there’s a critical limitation: SiN_x has a relatively poor thermal conductivity, so it is typically used at a thickness sufficient to ensure reliable trapping suppression.

Due to this weakness, it’s problematic to pair SiN_x with a topside diamond heat spreader. As the SiN_x layer sits directly above the active region, it’s a thermal barrier to heat flow from the channel hotspot to the diamond layer. It’s possible to thin the SiN_x layer and improve thermal coupling, but that comes at the cost of weaker surface control and the return of DC-RF dispersion. Due to this, designers seek the optimum trade-off when considering electrical stability and thermal performance.

But could compromise be avoided by employing a topside diamond layer, integrated close to the channel hotspot? Would this provide passivation-like surface stabilisation, while retaining excellent thermal spreading capability, and reducing or even removing the need for a thick SiN_x layer? If all this is possible, diamond would no longer function as just a heat spreader. Instead, it would act as a new kind of dual-function passivation layer – one that simultaneously improves heat extraction and suppresses dispersion. It’s a combination that’s especially attractive for high-power GaN transistors.

Close coupling
To put theory to the test, we have produced diamond-coated GaN HEMTs (see Figure 1), adding diamond with a microwave plasma CVD growth process that operates below 500 °C. As this growth temperature remains within a device-compatible range, we deposit diamond after full transistor fabrication, including gate formation and metallisation. This enables a device-first, diamond-last integration sequence, and avoids exposure of sensitive features to harsh thermal cycles that threaten to degrade performance.

Our polycrystalline diamond layer, approximately 1-1.5 μm thick, is grown on top of an ultrathin SiN_x layer. With a thickness of roughly 10 nm, it’s much thinner than would normally be used for passivation alone. According to electron microscopy and cross-sectional analysis (see Figure 2), we realise uniform, void-free diamond coverage across the device surface, including the narrow gate region. Our grain structure supports effective lateral heat spreading, and our geometry ensures that the thermal pathway is extremely close to our channel. In this configuration, an ultrathin dielectric separates the diamond layer from the active region. This architecture provides strong thermal coupling, while modifying the trapping environment at the surface.

Cooler channel

Figure 1. Cross-sectional schematics of the device (left) before and (right) after topside polycrystalline diamond integration. Diamond integration enhances thermal dissipation from the hotspot, enabling more efficient heat flow in both upward and downward directions.

To quantify the thermal benefit of our technology, we turned to gate resistance thermometry. After calibrating the temperature dependence of the gate resistance, we monitored resistance while the device dissipated power, and converted this signal into average channel temperature. This provides a direct electrical method for tracking internal heating under realistic high-power operating conditions.

Figure 2. Top-view scanning electron microscope (SEM) images of the fabricated AlGaN/GaN HEMT used in this work: (a) before polycrystalline diamond passivation, (b) zoomed-in view of the gate finger region showing the adjacent source and drain contact, (c) after polycrystalline diamond passivation, and (d) zoomed-in view of the gate finger region covered with polycrystalline diamond grains. Figures (c) and (d) confirm uniform polycrystalline diamond coverage across the entire device, including the gate finger region. (e) Focused ion beam (FIB) SEM cross-sectional image of the gate metal region, showing the topside polycrystalline diamond layer uniformly encapsulating the gate metal without voids.

Our measurements revealed a substantial reduction in device temperature (see Figure 3). When the HEMTs delivered a DC power density near 18 W mm^-1, the average channel temperature dropped by roughly 55 °C after diamond integration. At lower power densities that difference reduce, but it’s still 12 °C at a power density near 4 W mm^-1. These results confirmed that the addition of a topside diamond layer introduces an efficient thermal pathway and significantly lowers the effective thermal resistance.

Figure 3. Gate resistance thermometry results indicating strong channel temperature reduction after topside diamond integration.

As well as lowering the operating temperature, heat flows in different directions. With the introduction of our technology, as well as a downward flow through the GaN epilayer and into the substrate, heat spreads upward into the diamond layer. As the diamond, which sits very close to the hotspot, has high in-plane thermal conductivity, heat is redistributed laterally across the device surface. This spreading is particularly beneficial in multi-finger RF transistors, as it combats localised heating that can create strong temperature gradients and limit uniform current operation.

Trap control
Electrical measurements show that our diamond layer improves surface stability. We determined this by examining surface trapping under high-frequency operation – this is accomplished by comparing DC current-voltage characteristics with short-pulsed current-voltage characteristics, using electrically stressed quiescent bias conditions. When trapping is significant, there’s a reduction in pulsed current and a clear discrepancy from DC behaviour (see Figure 4). This undesirable phenomenon, known as current collapse, is a well-known signature of charge trapping at the device surface. Following diamond deposition, dispersion is dramatically reduced. On-current mismatch nearly disappears, and knee-voltage movement is minimal, even under more demanding bias conditions.

Figure 4. Conceptual illustration of DC–RF dispersion and the effect of diamond passivation in AlGaN/GaN HEMTs. Without passivation, surface traps have little impact under DC conditions, but under short-pulsed operation they capture charge, leading to channel depletion and current collapse. With diamond passivation, surface trapping is suppressed, preserving channel charge and maintaining stable current under dynamic operation.

An investigation of multiple devices determined that diamond integration reduced dispersion values from around 30-40 percent to roughly 5-8 percent (see Figure 5). That’s a level of suppression comparable to that realised with conventional SiN_x passivation. This shows that the diamond overlayer, together with the ultrathin dielectric underneath, is sufficient to control trapping, and indicates that diamond actively contributes to stabilising the surface. In other words, as well as increasing heat flow, the diamond layer improves electrical behaviour.

Scaling ahead
Our study provides an early example of how a low-temperature CVD-grown diamond layer improves a conventional GaN HEMT on two fronts: reducing surface trapping and enhancing heat removal.

Figure 5. Measured DC and short-pulsed I–V characteristics of GaN HEMTs without (left) and with (right) diamond passivation. Without diamond, a clear reduction in pulsed drain current compared with DC behavior indicates strong DC–RF dispersion due to surface trapping. With diamond passivation, the discrepancy between DC and pulsed currents is significantly reduced, demonstrating effective suppression of current collapse and improved dynamic performance.

In higher-voltage GaN HEMTs, featuring advanced field-plate designs, self-heating becomes much more severe as power density increases. This makes thermal resistance a key factor limiting performance. As our technology positions the diamond layer very close to the channel hotspot, with no thick SiN_x dielectric layer in between, cooling should become even more pronounced as bias voltage and output power increase during amplifier operation.

We expect applying our dual-function diamond layer to high-voltage GaN transistors will enable higher output powers, improved efficiency, and longer device lifetimes. Our initial results are just a baseline demonstration, and even larger gains are likely in more aggressively scaled, higher-power RF GaN devices.

Beyond GaN
Our work points to a broader shift in how thermal materials may be applied to compound semiconductor devices. Rather than serving as only external heat spreaders, materials such as diamond can become functional parts of the device. When integrated late in the fabrication process at low temperature, diamond is not just influencing heat flow – it is also smoothing peaks in surface electric fields and combatting trapping behaviour.

A topside diamond layer that performs a dual role – spreading heat and passivating the surface – provides a practical path toward the production of compound semiconductor transistors that run cooler, behave more predictably under large-signal operation, and scale more effectively to higher power densities. As integration techniques continue to advance, we anticipate dual-function layers, such as diamond, will become broadly useful across a wide range of next-generation compound semiconductor technologies.