Diamond Cools High-power Transistors
The vacuum tube amplifier is a technology threatened with extinction. Silicon LDMOS is already preferred for low-frequency, low-power applications and the hold that the vacuum tube has on the higher-frequency, higher-power applications could soon slip away. Although these tubes are still employed in amplifiers used in radar systems and satellite communications, there is an ever increasing threat from wide-bandgap technologies that are delivering impressive, improved performance characteristics.
Next-generation SiC and GaN amplifiers are expected to deliver output powers of hundreds of watts at frequencies beyond 10 GHz. This will make these devices an attractive option in various markets, including amplification for mobile-phone base-stations, satellite communications and high-power phased array radar systems. The primary advantages of these solid-state devices, much like their silicon brethren, are smaller sizes, cooler operation and suitability for integration into complex systems.
Transistors are already being deployed in increasing numbers in smaller areas and ever more complex systems. Radar systems, for example, can now employ more than 50,000 distinct amplifiers. This miniaturization requires modules that combine high performance with compactness and creates significant heat-management issues at the transistor and system level.
Smaller devices, such as FETs, present the greatest challenge for heat removal. These transistors, which are the primary workhorse of the microwave power industry, contain sub-micron heat sources that produce large temperature drops in the immediate proximity of the heat source – the regions around the gates and contacts. Traditional approaches for removing heat from these devices are based on external heat sinks. However, these only have a marginal impact on the temperature drops within 1 µm of the heat source, so there is a lot of room for improvement in thermal management.
The only practical way to improve heat flow from a sub-micron source is to spread it into the material surrounding the source. Placing materials with very high thermal conductivities close to the active region can achieve just this, and it occurs when SiC, a good heat conductor, is used as the substrate for GaN device growth. However, GaN-on-SiC devices are very expensive, while even better thermal management is possible by turning to a more powerful thermal conductor – synthetic diamond.
CVD-grown synthetic polycrystalline diamond has the best thermal conductivity of any naturally occurring substance known, with the exception of prohibitively expensive single-crystal diamond. Although the value for thermal conductivity depends on the quality of the diamond, it is in the range 12–15 Wcm–1K,–1, which is up to four times as high as that of SiC. This means that synthetic diamond can act as an exceptionally efficient heat spreader, by itself or in combination with other materials.
In many applications, a thin diamond layer can produce an adequate improvement in transistor thermal management. For example, calculations show that a 100 µm thick diamond layer can cut a SiC device s peak temperature by up to 50% (see results in figure 1a), or commensurately produce 50% more power than a SiC substrate device. Measurements on real devices back up this claim of improved thermal management, and those made by Lester Eastman s group at Cornell University show a substantial reduction in device operating temperature for a given power output.
Substrates with a CVD diamond layer are also competitive from a cost perspective – the price of a GaN-on-diamond FET epiwafer is very similar to a GaN-on-SiC equivalent. This cost, coupled with excellent conductivity, makes this substrate a very attractive platform for wide-bandgap transistors.
CVD diamond is polycrystalline, so it is unsuitable for epitaxial growth and GaN epilayers must be transferred onto this material from another wafer. At Group4 Labs, which is based in Menlo Park, California, we have a proprietary process that can do just this. Our process removes GaN epilayers from their original substrates and atomically attaches them to smooth CVD diamond wafers (see box "How to make GaN-on-diamond wafers"). The most important and innovative feature of this process is that the heat-conductive substrate ends up just hundreds of nanometers below the transistor s active area. This results in exceptional heat spreading and conduction away from the heat source, while maintaining the substrate s electrical isolation and low microwave losses.
We use three key metrics to assess the quality of our composite GaN-on-diamond wafers: bonding yield; wafer bow; and the mobility of the HEMT s 2D electron gas (2DEG).
The bonding yield is defined as the proportion of the wafer s surface that contains GaN joined to synthetic diamond free from voids. This is currently evaluated by visual inspection under a microscope. The quality of the materials surface preparation governs the bonding yield, which is typically 95%.
Minimal wafer bow is essential for the high-volume manufacture of GaN-on-diamond devices. We determined this bow by measuring the total-thickness variation (TTV) across our wafers with a surface profilometer. If optical lithography is used for device processing, then the TTV should be less than 50 µm over a 4 inch wafer, or less than 20 µm across a 2 inch wafer. Our state-of-the-art wafers currently exhibit TTV greater than 300 and 50 µm for 4 and 2 inch wafers, respectively, but these values can be driven down by optimizing our growth conditions.
Reductions in the HEMT s 2DEG mobility could be caused by dopant or dislocation diffusion. However, no changes in mobility have been observed from the transfer of the transistor to the diamond platform, according to Hall and photoluminescence measurements at various temperatures and resistivity maps. Photoluminescence mapping revealed that the aluminum composition remains unchanged to within the resolution of the measurement (