Let's face it. Even though silicon (Si) is a great material for logic, it is not the material of choice for many of the other parts of an embedded system. This is particularly true for systems involving high power, high frequency, or optical functionality.
For those applications, gallium nitride (GaN) is a much better material. It has a wider bandgap, higher breakdown voltage, a larger critical electric field, and higher thermal conductivity. Bandgap is the energy required for an electron to jump from the top of the valence band to the bottom of the conduction band within the semiconductor. Stated another way, it is the amount of energy needed to free an electron so that it can become mobile. It is typically on the order of a few electron volts (eV). It is 3.4eV for GaN and 1.1eV for Si.
This means that GaN devices can operate at higher voltages and higher switching frequencies, handle higher power density, and (perhaps most importantly) provide higher power efficiency than Si devices. GaN is also being used more often for optical components such as LEDs, laser diodes, and optocouplers. This is because GaN has brighter emission characteristics than silicon and other materials. A TechNavio report (purchase required) predicts that the GaN semiconductor device market will grow at a compound annual rate of 18 percent from 2012 to 2016, versus 3 percent for silicon. One area of disagreement about GaN is which substrate to use. The most common ones are sapphire and silicon carbide, but they come with problems, including high cost, poor thermal conductivity, and small wafer size. For example, a six-inch sapphire wafer costs about $400, compared with $15 for a silicon wafer of that size.
High levels of integration are required these days. There are companies that would like to get all the benefits of GaN on Si, but this is not easy. Thermal mismatch and lattice mismatch have limited the effectiveness of growing GaN directly on silicon, because they cause issues such as bowing and cracking, which can lead to device failure. However, whenever a problem presents itself in this industry, an army of people will look for a solution.
Most of the research in this area has centered on applying a seed layer on to the silicon before the GaN is deposited. This seed layer could be aluminum nitride (AlN), which is created by depositing aluminum and then subjecting it to ammonia (NH3 ). During epilayer growth, compressive stresses are generated. It has been shown that the residual stress is dependent on the impurity levels, and that doping can enhance the epilayer's tensile strength. If the tensile strength is comparable to the compressive strength, it is possible to make the GaN epitaxy crack free. (Epitaxy is where one crystalline structure is deposited on to another. In some cases, there is a specific, defined orientation of the two structures.) Patterning substrates by masking or etching the substrate or buffer layer has also been shown to be effective in reducing cracking.
Yield is more of an issue in the power domain than for optoelectronics. The reason is simple. GaN is a defect-laden material. With an LED covering 0.1mm2 and a 10A transistor spanning several square millimeters, it is clear that yields will be higher where GaN use can be minimized. Also, for an LED of this size, redundancy could be built in easily.
Another area of investment is in wafer size. Six-inch wafers are common today, but larger ones are coming and are necessary for further cost reductions. Still, production is beginning to ramp up for LEDs, and the knowledge learned from this could easily make other optoelectronics devices available within a few years. Electronic design automation companies are already gearing up for this.
Are LEDs such a big market opportunity that most of the investment dollars will go into this narrow field, or is this a necessary step to optimize the process?