Diamond: driving down the heat - synthetic supercapacitors' thermal management

Materials World magazine
24 Sep 2013

Diamond has long been heralded for its high mechanical strength and thermal conductivity. Felix Ejeckam, Bruce Bolliger and Dan Twitchen from global supermaterial company Element Six discuss how synthetic diamond is enhancing high-power radio frequency semiconductor devices.

Heat is known to be the source of most electronic device failures, performance limitations and size/weight excesses. The combination of these issues leads to tremendous system costs, particularly for defence applications. Jet engines are being pushed to do more work with less energy, radar systems are increasingly required to sense more and reach farther while weighing less, and satellite communications equipment designers are asked to transmit more information faster. All of these challenges require better thermal management of electronic devices.

Diamond has the highest room-temperature thermal conductivity of any commercially available thermal material, which makes it the superlative heat spreader. Typically, synthetic diamond can be engineered to have a thermal conductivity in the range of 1,000–2,000W/mK – a factor of up to 10 over current alternative materials such as beryllium oxide, aluminium nitride and aluminium oxide. The benefits of using diamond include lowering a device’s maximum temperature, and this can equate to longer lifetime, increased reliability, device miniaturisation and more power at a constant temperature. Combined with its high mechanical strength, synthetic diamond can also be an excellent semiconductor wafer substrate for thermal management of high-power radio frequency (RF) gallium nitride (GaN) devices.

R&D is currently underway at Element Six, in part under the auspices of the US Government’s Defense Advanced Research Projects Agency [DARPA], on the application of diamond as a heat spreader and semiconductor wafer substrate. The aim is to determine how the defence industry might most effectively integrate diamond to exploit its benefits for RF semiconductor applications.

Tailoring diamond properties
Diamond can be synthesised either by a high-pressure, high-temperature (HPHT) method or by chemical vapour deposition (CVD). The latter technique enables higher purities and larger deposition areas, so is most often used for applications that focus on diamond properties other than hardness – for example, heat spreading and optical windows. CVD diamond synthesis normally uses a small fraction of carbon (typically less than 5%) in a background of hydrogen. The carbon fraction can be supplied in the form of CH4. Growing diamond without large graphitic phases requires CVD gas temperatures above 2,000K to break down the hydrogen molecules into atomic hydrogen, which subsequently leads to stabilisation of the diamond growth surface and preferential etching of defects. One gas heating technique that enables higher levels of purity as well as greater diamond growth thicknesses is the use of a microwave plasma. Factors such as impurity concentration and grain size strongly influence the thermal conductivity and optical absorption properties of CVD diamond. The CH4 and H2 source gases are now commercially available with impurity levels of less than 1ppm, enabling CVD diamond growth with exceptionally high purity and with thermal conductivity as high as 2,200W/mK. For example, incorporating impurities into the gas phase is known to modify the morphology and texture of polycrystalline CVD diamond, and can influence growth rate. These variables allow the CVD diamond growers to tailor properties such as thermal conductivity, to optimise the cost/performance ratio for specific applications.

In addition to controlling purity levels, the CVD process can also grow polycrystalline diamond as freestanding wafers with diameters of up to 138mm, or conformally deposited on shaped substrates, enabling many different thermal management, optical and acoustic applications. These CVD diamond techniques have also been successfully applied to produce single crystal with typical sizes of 5mm.

Much of the R&D has been focused on refining microwave plasma-assisted CVD diamond properties, for applications such as:

  • thermal management of radar RF devices and laser diode arrays
  • optical windows for high-power lasers
  • high energy, electrochemical and magnetic sensors for robust environments
  • growing larger single crystal diamond to reach the ultimate in thermal conductivity, flatness, and particularly smoothness

Minimising roughness is required to bond diamond with minimal thermal barrier resistance to, for instance, semiconductor die. Researchers are currently developing polishing techniques to minimise roughness beyond what is commercially available today (average roughness [Ra] <20nm for polycrystalline diamond and Ra <5nm for singlecrystal diamond), and work on feature development is also ongoing, using either laser or etching techniques. Today, features can be formed in diamond with aspect ratios as high as 10:1 that are needed, for example, in micro-channel cooling.

Yes, we GaN
For development of GaN-on-diamond wafer substrate for use in military radar systems, the first goal was to increase by a factor of three the power-carrying capacity of GaN-based RF transistors. To do this, a CVD diamond substrate is brought to within tens of nanometres of the heat-generating layers of the GaN transistor. Owing to the high thermal conductivity of the diamond substrate, this proximity of diamond to the transistor’s gate enables the gate’s heat to be whisked out via conduction.

When the temperature of the transistor gate finger is reduced, adjacent fingers can be brought closer together without triggering thermal crosstalk. Simulation and modelling show that these adjacent fingers can be brought three times closer together on a GaN-on-diamond wafer than on GaN-on-silicon carbide (SiC) incumbent technology. Therefore, it is expected that a GaN-on-diamond wafer can produce three times more RF power density than a GaN-onSiC wafer held at the same temperature. As such, a GaN-on-diamond device can exhibit lower gate temperatures than a GaN-on-SiC device producing the same RF output power density would, thereby significantly prolonging the lifetime of a device.

Making wafers
Forming a GaN-on-diamond wafer requires several steps. Wafers are prepared by first removing the host Si (111) and transition layers beneath the GaN highelectron mobility transistor (HEMT) epitaxy, depositing a 50nm-thick dielectric onto the exposed AlGaN/GaN, and finally growing 100μm of CVD diamond onto the dielectric. The wafer is then mounted on a diamond carrier to provide sufficient flatness for processing the wafer substrate in a semiconductor fabrication.

A number of key challenges arose in developing this process. In the first step, a process had to be developed that would:

  • withstand high-temperature steps further down the process
  • preserve the GaN epitaxy’s mechanical, materials, and electrical qualities after the host Si has been etched away
  • not introduce defects or dopants into the GaN epitaxy throughout the subsequent process steps
  • be readily reproducible with high yield and low cost in any semiconductor fabrication

The third step required development of a proprietary dielectric that would serve as a thin intermediary between the GaN and the diamond. This intermediary would have to be as thin as possible, electrically insulating for RF applications, and be absorbing of any thermal expansion coefficient mismatches between GaN and diamond, to ensure the GaN stays permanently adhered to diamond.

The whole process flow was broken down to short experimental loops that could be executed quickly in order to vet a number of potential solutions. Each of the five steps was executed in isolation, with dummy Si wafers and thin diamond substrates (for faster turnaround) to test potential solutions. GaN was simulated with cheaper substitute films of silicon dioxide, polysilicon, or silicon nitride (SiN). Strength of the GaN–diamond interface was a key progress marker at each process loop. Dozens of thermal cycles, ranging from 23–1,000°C, were used to test the resilience of the test samples during the iterative loops. Closer to the start and the end of the whole process, electrical tests (such as Lehighton, Hall and CV) were used to assess the effects of the various experiments on the GaN epitaxy’s charge layer.

While most of these process-related challenges have now been overcome, several key wafer specifications have yet to be fulfilled for commercial application. Currently, the GaN-on-diamond wafer uses a thicker diamond carrier to flatten down the bow measurement to industry norms (around 20–30μm) – ideally, a freestanding wafer would not need a carrier to be flat. Further elevation of the diamond’s thermal conductivity would improve the price/performance ratio for economic use of this semiconductor wafer. And the robust four- and six-inch wafers necessary to fit well within the semiconductor infrastructure have yet to be demonstrated.

As power densities of semiconductor devices for defence applications continue to increase, their thermal management will also become ever more challenging. However, the results to date have so far proven CVD synthetic diamond to be a viable solution. Future research efforts will focus on refining these properties cost effectively as well as optimising the proximity of free-standing polycrystalline, singlecrystal diamond, and GaN-on-diamond substrates to the heat-generating junction of semiconductor die.

For more information, contact Bruce Bolliger bruce.bolliger@e6.com