Wide-bandgap electronic devices combine high efficiencies with good temperature- and current-handling capabilities, which makes them very desirable for a variety of applications. These include deployment in hybrid cars and trains to convert DC output into an AC signal to power the motor, and installation in base stations for the amplification of wireless signals.
Ideally these devices would be manufactured on a native material platform because this minimizes epilayer strain. With SiC this is possible, but substrate prices are high and different polytypes can form in the epilayers, impacting device reliability. However, these drawbacks are far less severe than those for GaN. It is very difficult to make this material in large sizes and it also suffers from dislocations that cause premature device breakdown (figure 1).
To address the problems with GaN, we have been developing a solution-based method for crystal growth at Osaka University, Japan. This process is based on an established technique – sodium-flux liquid phase epitaxy (LPE) – which can produce high-quality GaN crystals in a vessel that contains a gallium–sodium mixed metal melt and nitrogen gas pressurized to 50 atm (figures 2 & 3).
Sodium's role in this process is to drive the dissociation of nitrogen molecules at the gas–liquid surface. The nitrogen radicals that are formed dissolve easily in the metal melt at 800 °C. Without the presence of sodium, forcing nitrogen into a solution requires pressures of at least 10,000 atm.
GaN crystals were first produced from a gallium–sodium mixed metal melt over a decade ago by Hisanori Yamane and co-workers from Tohoku University. However, the pieces of GaN were 1 cm or less across. Since then, one of the primary goals for researchers in this community has been to increase the crystal dimensions to a more commercially useful scale.
We have done just this by inserting seeds of 2 inch GaN crystals into the metal melt. These seeds have high dislocation densities because they are grown by HVPE or MOCVD. However, the dislocations are not transferred to the LPE-grown material as they tend to annihilate close to the interface with the seed.
In 2005 we produced the first 2 inch GaN using this technique. This had a dislocation density of 2.3 × 105 cm–2 – more than an order of magnitude less than typical commercial GaN substrates. Scaling to this size demanded control of the thermal convection currents, alongside optimization of the crystal's growth conditions. If a 2 inch piece of GaN was dropped into the metal melt, nitrogen would not reach the crystal surface and would be lost to the growth of polycrystals around the gas–liquid interface.
We have recently developed proprietary apparatus and techniques that can increase the crystal sizes and growth rates. If many GaN wafers can be sliced from large single crystals, this could dramatically cut the manufacturing costs of these substrates. Last year we focused on improving the growth rate and tripled it to 0.6 mm/day. This is suitable for the mass-production of substrates for lasers, but faster growth is required to make a platform for LED manufacture.
With this new apparatus we have produced the thickest 2 inch GaN crystal ever made by sodium-flux LPE (figure 4). We are now working to extend the technique to 4 inch crystals by developing another, faster tool. If it is successful, this will be a major step towards the development of affordable GaN devices that can be grown on a native material platform.