A misconception is holding back the development and deployment of GaN devices that are built on silicon substrates. This platform is widely blamed for compromising blocking voltages, but it doesn’t: It is possible to make diodes and HEMTs on silicon that have breakdown voltages of well over 2 kV, according to Timothy Boles and Douglas Carlson from M/A-COM Technology Solutions, Tomas Palacios from MIT and Mike Soboroff from the US Department of Energy.

GaN HEMTs are on the verge of revolutionizing the power electronics industry, thanks to their capability to take device performance to a new level. Their tremendous promise has already spurred widespread academic and industrial development of transistors for power switching applications that have an impressive set of attributes: Power densities of more than 2 W/mm; continuous current handling capabilities of 10 A or more; and very high reverse breakdown blocking voltages, which typically exceed 1 kV. Prototypes of these transistors have been built on sapphire, SiC and silicon −three platforms with differing pros and cons.

Efforts at device development have delivered much heralded, well-deserved technical successes, but this is yet to lead to significant commercial adoption of GaN. There are many explanations for this, and they tend to revolve around the view that while GaN diodes and transistors produce impressive results, they are far more expensive than their silicon rivals. Or, to put it in simper terms, they don’t get close to the bang-per-buck of the incumbent technology.

This explains what is happening in the marketplace today. In this arena, the limited success of GaN products can be accounted for by citing a widely held mantra: Once the minimum level of performance needed for the application is achieved, the cheapest solution will win. So, in order for GaN-based products to realize their full potential in a broader marketplace, two changes must take place: The cost of material must plummet; and device manufacturers must target applications that cannot be addressed by lower-cost rival technologies, such as those based on silicon.

One route to driving down the cost of GaN devices involves building them on silicon and processing them in silicon lines. Staff at the US Department of Energy (DoE) subscribe to that view, and they are funding a project to investigate and ultimately commercialize GaN-on-silicon power device technology. We are all involved, and we bring together expertise from MIT, MIT Lincoln Laboratories, M/A-COM Technology Solutions, and, most recently, Kopin Corporation.

Our goals for this project are to extend the blocking voltages of GaN-on-silicon transistors, diodes, and monolithic circuits to more than 5 kV, and to increase current carrying capability beyond 10 A. We are already well on the way to meeting this, as we have already exceeded blocking voltages of 2 kV and current handling greater than 10 A. These successes stem from focusing efforts on materials, device design and fundamental processing issues.

Lessons from the past

To understand where the GaN market stands today, it is helpful to have some knowledge of the evolution of GaAs devices in the microwave sector. Here, in the early days, the lack of commercial availability of high-quality GaAs epitaxy limited widespread adoption of GaAs-based devices. At that time −in the late 1970s and the early 1980s −many companies were essentially totally dependent on their own internally produced GaAs epitaxial wafers, and the overall marketplace was generally limited to low-volume, government-funded applications. But this situation changed completely with the introduction of multiple merchant suppliers of high quality GaAs epitaxy in the mid-1980s, and the commercialization of cellular technology, which provided a market that could not be addressed by silicon. These market conditions spawned a roadmap for exponential growth of relatively low cost, high volume GaAs devices operating at radio frequencies.

These trends in the GaAs market could be mirrored in the GaN power device sector. It is our view that these wide bandgap devices will only be cost-competitive with the incumbent technology when they are manufactured on silicon substrates, which are relatively cheap, available in wafer sizes up to 300 mm, and can be run through the lines of existing high-volume silicon foundries. As the GaN power industry moves towards this, the number of merchant suppliers for GaN-on-silicon wafers will grow, and their rivalry will push prices down as volumes take off.

We don’t expect the market for GaN and SiC devices made on SiC to evolve in the same way. That’s partly because there are a limited number of merchant suppliers of finished epitaxial wafers, and there are little more than a handful of SiC substrate manufacturers. What’s more, the cost of SiC substrates will always be far higher than that for silicon, due to fundamental differences in the crystal growth of both materials. Growth of SiC takes place at 2100 °C, 600 °C higher than that for silicon, and laws of physics dictate that its growth rate can be up to three orders of magnitude slower than that for silicon.

Based on these fundamental differences in crystal growth, it is difficult to see how GaN-on-SiC products could ever be cost-competitive with those incorporating GaN-on-silicon technology. Then factor in the differences in substrate diameters (150 mm or less for SiC, compared with 300 mm in high volume production for silicon), along with the overall market drive of silicon versus SiC, and it is beyond question that, from a cost perspective, a silicon based technology will always be the winner.

Market opportunities

The market opportunities for driving the sales of GaN-on-silicon epiwafers and devices fall into several categories.  One is a range of power-switching applications requiring devices capable of handling 1 kV. These devices could increase the efficiency of power transmission in the grids of tomorrow, which will be integrated with a growing number of solar and wind farms. Efficiency improvements translate into cost savings of $50 billion over the next 25 years, due to a reduction of new power plant construction. 

Another opportunity for GaN-on-silicon is in electric cars, including hybrid, plug-in, and fuel cell designs. These vehicles have roadmaps though 2014 to push to a 2 kV operating buss a move that will reduces on-state and switching losses and boost efficiency. The latter gain is highly beneficial, because it eliminates the need for a cooling loop for inverter circuitry, it increases battery life, and it delivers savings in terms of weight, volume and cost. Efficiency gains resulting from higher voltage-handling capabilities also make it attractive to deploy GaN-on-silicon devices in industrial drives, electric trains, and military and merchant fleet applications.  Our low cost GaN-on-silicon technology has the potential to target all of these applications. It is based on the use of commercial GaN-on-silicon epitaxy, and involves HEMTs with an AlGaN buffer and barrier layer (see Figure 1).



Figure 1. A typical GaN HEMT epitaxial structure

These structures can be grown by either MOCVD or MBE, using a process that begins with the deposition of a nucleation layer that is unique to the starting substrate material (either silicon, SiC, or sapphire high-resistivity substrates). A relatively thick AlGaN buffering layer or super-lattice structure is then deposited onto the nucleation layer. This serves two purposes: It mitigates lattice mismatch and induced strain associated with the substrate/epitaxy transition; and it provides electrical isolation of the active device region from the substrate and substrate/epitaxy interface.  Leakage from this buffer, which ultimately sets the limit for the overall device reverse breakdown, is determined by the composition and total thickness of the buffering layer.

After the buffer has been deposited, growth continues with the addition of the active GaN and AlGaN Schottky barrier layers. The aluminium fractional content in this barrier is generally between 22 percent and 32 percent, with the precise value chosen to deliver the best compromise between on-resistance and leakage characteristics.  A GaN cap often completes the structure. This reduces oxidation of the underlying AlGaN film and improves the device’s contact resistance.

We are by no means the only developers of GaN-on-silicon technology, but the efforts of most groups in this field have been limited to 600 V applications. This focus has led many within the power electronics industry to incorrectly believe that GaN-on-silicon devices are limited to this operating range, and GaN single-ended devices must be built on SiC if they are to deliver voltage standoff capabilities above 1 kV. We have no doubts that GaN-on-SiC devices can operate at 1 kV and more, but they are prohibitively expensive – and we can’t see that changing. Our mission is to address the misconceptions regarding the limits of GaN-on-silicon technology, and show that it is capable of creating high blocking voltage devices capable of carrying high currents. 

Thanks to recent advances in commercial GaN-on-silicon epitaxy – especially in the construction of advanced buffer layers coupled with field plate design technology – HEMTs and Schottky diodes can now deliver standoff voltages in excess of 1.5 kV. Devices made by us can even hit much higher values than this, with GaN-on-silicon HEMTs producing blocking voltages in excess of 2.5 kV, and Schottky diodes incorporating the same materials technology delivering stand-off voltages of more than 3 kV.  These levels of performance are rebuffing the accepted wisdom that GaN devices must be built on SiC, if they are to serve power electronics applications requiring enhanced voltage operation.

Our results include a portfolio of devices combining 1.5 kV blocking voltages with current handling capabilities in excess of 10 A. These devices were fabricated from commercially produced, 100 mm diameter, GaN-on-silicon epiwafers. 

Optimizing the location and dimensions of the field plate is one of the keys to realizing a high blocking voltage. The importance of this addition to the transistor is clear from plots of the drain source breakdown as a function of the dimension of the field plate (see Figure 2).



Figure 2. The breakdown voltage of a GaN-on-silicon HEMT, which has a fixed 20m gate-to-drain spacing, is influenced by the design of a source connected field plate (SCFP). The average breakdown value achieved at a 4.5 m SCFP overlap dimension was 1322 V.  This translates to an average field strength in the drain region of 66 volts/mm (6.6x105 V/cm) .  The highest breakdown voltage achieved was 1632 volts, corresponding to a drain field strength of 82 V/µm (8.2x105 V/cm)

We have also measured the on-resistance of these high-voltage GaN-on-silicon FETs, and compared this data against theoretical limits for a range of materials, and results from the technical press for various field plate design geometries and material systems (see Figure 3). This benchmarking exercise reveals that our GaN-on-silicon results compare extremely favourably with values for reported state-of-the-art devices formed on a range of substrates, including GaN-on-SiC, and various field plate approaches.


Figure 3. Material theoretical limits and a plot of Ron verses reverse breakdown voltage for various field plate geometries

To build a circuit for high power inverters, diodes with a high reverse breakdown need to be used in conjunction with high blocking voltage transistors. Until now, wide bandgap diodes with reverse breakdowns of 1 kV or more have only been realized in GaN-on-SiC and SiC, but our DoE-sponsored project has shown that GaN-on-silicon is more than capable of delivering this level of performance.

Schottky diodes architectures can be divided into lateral and vertical designs, and we have focused on the former, because this enables integration with transistor structures. Pairing these two devices together holds the key to the creation of monolithic power circuits. The reverse breakdown of our devices, known as lateral GaN-on-silicon anode connected field plate (ACFP) Schottky diodes, depends on the anode-to-cathode spacing (see Figure 4). Optimize this, and this diode on silicon can produce a reverse breakdown in excess of 1.5 kV.



Figure 4. Lateral ACFP GaN Schottky diode reverse breakdown as a function of anode to cathode spacing

Our GaN-on-silicon devices also have values for on-resistance that compare favourably with equivalents built on other substrates and incorporating different designs (see Figure 5 for the comparison and Figure 6 for a cross-section of a typical field-plate structure). Current carrying capability can exceed 10 A, and its value can be adjusted by scaling device dimensions.

Although our device development is still in its early stages, it shows that GaN-on-silicon technology can produce HEMTs and Schottky diodes with blocking voltages well in excess of 1 kV. Thse devices  should win significant sales in the power electronics marketplace, because they will be more competitive than GaN-on-SiC, which delivers impressive device results but is hampered by the limited availability of substrates and epiwafers (these are pricey and cannot scale to the same dimension as those based on silcion). We are convinced that while there is still some work to do to improve the capability of GaN-on-silicon power devices, they will be the only significant wide bandgap alternative to the silicon incumbents. They are destined for sigificant success, and we are sure that they will win widespead acceptance in the marketplace.


Figure 5. The GaN Schottky diode on-resistance of the device made on a silicon substrates compares favourably with those built on sapphire, SiC and single crystal GaN. 



Figure 6. The anode-coupled field-plate GaN Schottky diode produced at M/A-COM Technology Solutions

The work reported here has been sponsored by Department of Energy under Contract Number DOE IA No.: DE-AI26-OE0000121 Award No. :DE-AI26-07NI43294/006