Silicon-on-diamond substrates: a GaN HEMT maker's best friend?
The substrate is the Achilles heel of today s GaN HEMTs. The high power densities promised by these devices, which are 10 times greater than their GaAs cousins, expose the wafer s poor heat handling capabilities, while the typical substrate sizes impact on economies of scale. These drawbacks limit the potential performance of the devices and hinder their deployment in cell phone base stations and other forms of wireless communications.
The development of substrates for these RF devices has a checkered history. GaN is the ideal platform from a growth perspective because it produces very low defect density epilayers. However, this type of substrate is not used because it is too expensive, difficult to make and limited to a few centimeters in diameter. GaN HEMT developers instead turn to other platforms for GaN epitaxy, including sapphire, AlN, SiC and several silicon-based wafers. They are much cheaper to produce, but aside from the various silicon types, are limited to sizes of 100 mm or less.
Unfortunately, silicon s larger wafer sizes are balanced out by a low thermal conductivity that keeps device performance within limits (see table 1). Poor thermal management caps the transistor s maximum output power and reduces the device s lifetime through high operating temperatures.
Hybrid structuresHowever, it is possible to combine the attributes of silicon s larger diameters and higher thermal conductivities by employing a multi-layer substrate.
At sp3 Diamond Technologies, based in Santa Clara, CA, we have produced silicon-on-diamond (SOD) substrates that can be used for GaN growth and feature an integrated thermal spreading layer of diamond (see figure 1).
These hybrid structures have a one-to-several-microns-thick silicon layer on top of a diamond layer that is tens of microns thick. Both of these layers, which are grown on a silicon substrate by CVD, have thicknesses that are fine-tuned to specific applications. The silicon wafer also serves as a handle for processing and it can be either thinned or completely removed before packaging. The addition of diamond boosts the thermal handling properties of the wafer. At this thickness the SOD wafer s thermal conductivity is influenced by the grain size, and is 2–3 times greater than copper and 4–5 times higher than SiC.
Our SOD substrates – which are manufactured with a hot filament CVD process in sizes of 50 to 300 mm and can be doped with boron – could increase the output power of HEMTs from the current typical value of 6 W/mm to more than 15 W/mm. This increase in performance results from a cooler channel temperature that reduces the device s running temperature and boosts transistor lifetime by several orders of magnitude. This improvement will also lead to an increase in the reliability of systems using these devices.
The performance gains produced by this type of substrate are illustrated by simulations from the Naval Postgraduate School in Monterey, CA, that compare AlGaN/GaN HEMTs on sapphire, silicon and SOD substrates. This approach, which uses commercial software to model the device with a matrix-mesh of thousands of grid points, can simulate transport properties using differential equations derived from Maxwell s equations. A lattice heat flow equation is also included in the calculations, which break down the HEMT into nine regions, such as the oxide region, an AlGaN layer and different electrical contacts. Electrical results from a two-dimensional model produced similar predictions to a three-dimensional one, but with greater speed. The results from this simpler model are shown in table 2.
Combating hot spotsThese calculations reveal that switching from SiC to the SOD platform should boost HEMT current density by 28%. Drain current would be increased by 43%, while maintaining an overall transistor temperature of 300 K.This operating temperature, which is 80 K lowers than that for GaN HEMTs built on either sapphire and silicon, boosts device lifetime. Calculations, which rely on the activation energy associated with the materials in the epilayers, predict that the reduction in electromigration within the device could increase the transistor lifetime by a factor of between 16 and 1700.
SOD wafers are also better at addressing issues associated with non-uniform heat generation across the chip – a common characteristic of many devices. The problem arises because GaN is unable to effectively remove the heat generated during device operation. Heat spreaders can be attached directly to the device to distribute heat away from the hot spots to colder regions of the chip. However, heat still has to be extracted through the substrate s entire thickness before it can be quenched by a heat spreader or a heat sink.
A more effective solution is to place the heat spreader as close as possible to the heat generating areas. Our SOD wafers do just this because they contain a heat-spreading diamond layer that is within a few microns of the chip s surface.
This diamond layer reduces the local junction temperature and equalizes the temperature across an entire chip at the device junction level. The improvements in temperature uniformity are predicted by calculations from Nitronex, which compare multi-gate RF power transistors on a standard silicon substrates and a SOD platform (see figure 2).The Rayleigh, NC, firm is working with us as part of a $0.75 million contract from the Missile Defense Agency to build active devices on SOD.
The advantages of SOD substrates over those manufactured from silicon and SiC for a range of output powers are also illustrated by calculations carried out by TriQuint (see figure 3). These simulations take a very conservative approach by using a thermal conductivity for diamond of only 6 W/cm–K and may underestimate the true advantage of SOD. Even so, the predicted improvements are dramatic.
Making the switchAlthough we expect GaN HEMTs to benefit most from SOD, we believe that our hybrid substrates with tailored silicon and diamond layer thicknesses could also aid optoelectronic devices. For example, there were reports late last year that Shimei Semiconductor, a start-up company based in Kyoto, Japan, is developing a blue LED using GaN-on-silicon epitaxy. Switching from silicon to SOD substrates could not only reduce the device s operational temperature, it could also increase its temperature stability and lead to lower variations in color of LEDs with device aging.
The advantages could also be felt in more exotic devices, such as GaN-based VCSELs, by increasing the emitter s output power. In fact, we believe that in the future all high-performance efficient devices will require the use of a heat spreading material like diamond in very close proximity to the junction.