I was discussing the compound semiconductor landscape with fellow attendees at APEC 2010. Our discussion was probably one of many prompted by IRís further introduction of devices and the presence of Cree, TranSiC, and others at APEC. I posed the question, “Where are the high voltage gallium nitride (GaN) power switching devices? After all, GaN is reputed to be a high voltage technology by leading technologists in that field.”
In the past few years, compound semiconductors have become the focus of development as semiconductor engineers have strived to get to the next better device. The bulk of the development has been for applications such as radio frequency (RF) power transistors and light emitting diodes (LEDs). Now, as of this writing, two suppliers have introduced low voltage GaN power switching devices.
However, we saw that GaN might well be the technology to provide 600 volt and 1200 volt semiconductor devices for every type of high voltage power conversion, including variable-speed motion control, solid-state lighting, electric vehicle drives, wind and solar converters, uninterruptible power supplies, and, yes, eventually the higher power distribution, transmission, and traction markets.
WHY GALLIUM NITRIDE?
GaN is the most practical, lowest loss, power semiconductor material available today. SiC can achieve reasonable performance as a low-loss device, but experts have assured me the performance of fully developed GaN will be two times (2X) better than SiC.
GaN provides the fastest path to market. As one expert told me, “GaN has always had a high natural speed of device development and market accessibility.” Technologists will develop GaN materials and products faster than either gallium arsenide or silicon carbide devices. The reason: material problems plague GaN far less than other compound semiconductors. For instance, if a non-GaN LED had as many imperfections as a GaN LED, it would not emit any light. It would be absolutely dead! Yet the GaN LED operates extremely well, even with imperfections.
Let's use previous compound semiconductor LED development history as a guide. The figure illustrates GaAs development taking 26 years to obtain 15 lumens per watt (lm/W) which is the equivalent of the incandescent lamp. SiC never got onto the roadmap before GaN development took off. In only six years, GaN LEDs reached 15 lumens per watt. GaN material development was considerably shorter than either GaAs or SiC so therefore product development could commence sooner. In the development of both GaAs and SiC devices, it was first necessary to create a near-perfect material. For instance, perfect SiC material has taken over a decade to develop! According to Yole's market analysis, it is still plagued with micro-pipe and basal plane dislocations. GaN material development was much faster! Thus, GaN engineers could start on product creation a lot sooner in the development process than their SiC or GaAs counterparts could. Experts tell me that we can expect the same fast GaN development time line for high voltage power switching devices. A leading GaN technologist recently shared with me, “If we needed to develop GaN power device material to the extent that they had to develop GaAs material, we would not have seen the devices that are already introduced. We would have to spend a billion dollars before the material became clean. That would make the whole industry not only late in device development but also make the devices expensive. Fortunately, that will not be the case!”
GaN has much better device size scalability. We can also conclude from the figure that technologists were able to scale GaN four times (4X) beyond either SiC or GaAs materials, reaching a benchmark of 168 lm/W. We can expect this to be the case for GaN power switching device scalability. GaN will happen sooner, and then will scale to a much higher device size. Presently, the low voltage GaN designers are proving that very point.
GaN is low cost. It is higher cost than silicon, but GaN will always cost less than SiC because GaN is compatible with silicon substrates affording a large-area foundation substrate - and SiC is not compatible. At APEC 2010, low voltage GaN speakers very accurately emphasized this point. However, if there ever is a fundamental breakthrough to make SiC compatible with silicon, then that will seriously reduce its cost. If we look ahead, diamond is the only material better than GaN, but that is most probably more than 20 years away from becoming a reality. Even if developed, experts cannot even get a glimpse of a roadmap to cost-competitive devices yet.
Continue to next page
GaN offers high efficiency concurrent with high power density. GaN offers almost zero switching losses and lower conduction losses per mm2. We see time and time again the present highly efficient converter designs achieved at the expense of power density. Designers, utilizing either silicon superjunction or standard silicon technologies, are intentionally reducing the switching frequency to achieve ultra-high efficiency. Of course, lower-frequency converters require larger power-passive components - especially magnetic elements. So, designers choosing that route are doing so at the expense of power density. I cited in a previous editorial that high-voltage standard silicon transistors were preferred over superjunction devices in ultra-high efficiency design. GaN offers the opportunity to increase efficiency while raising the switching frequency and thus increasing power density.
So getting back to the low-voltage GaN power switching devices we saw at APEC 2010, I asked, “What have the low-voltage developers achieved?” The response I got from those knowledgeable in that field was that GaN products with ratings of 28 V and 48 V have been available for several years. High volume RF applications use them. However, those working in the low-voltage GaN arena have recently strived very successfully to scale the technology to higher current reaching 40 A or greater. They have achieved large-scale devices as one might expect from the GaN LED scalability exhibited so dramatically in the figure. That is a significant accomplishment! They have proven that GaN is no longer a university technology.
Yet, their path to product success may not be that easy. Unfortunately, low voltage GaN is uncomfortably close to silicon in performance and cost expectations. This will make their road to success difficult.
Further, their attempt to scale voltage appears to be somewhat elusive. I have received comments from those who have tested presently available GaN transistors. They advised me that they have experienced a considerable decrease in efficiency when the devices operated at voltages approaching 80% of product voltage rating.
Upon learning of this, I asked how one might minimize this increased power loss. An expert in the field of GaN device development informed me that operating the devices at voltages considerably lower than rated voltage would minimize this increase in power loss.
Considering this to be a waste of device capability, I inquired whether this was an insurmountable problem. He assured me that GaN devices could be made that did not exhibit this phenomenon. I was shown device performance data that proved a GaN transistor, rated for several hundred volts, did not have this significant power loss increase when operated close to its rated voltage. Therefore, we can rest assured that future GaN devices will have excellent voltage scalability.
Continue to next page
If we look at the high-voltage compound semiconductor power switching devices, we see a struggling landscape. SiC devices have tried for a decade to get traction with a high-voltage Schottky diode. It performed well, but cost has remained high and it lacked the partner transistor. The high-voltage SiC transistor is finally making its debut - but at what price? It may get some traction in boutique markets, but may still be plagued with yield problems - which will further affect cost.
There is one high-voltage GaN supplier with a diode product purported to be available this year. History has shown that the various previous attempts to introduce a solo diode product are far from stellar ventures - in any technology! For that reason, even if they produce an excellent diode, they may not get enough successful traction in the market unless they also develop the transistor.
As one expert advised me, even though GaN is a high-voltage material, developing GaN product with ratings of 600, 1200, and 1700 V is not a trivial matter. Another shared that GaN is not an easy material to master for high-voltage devices. Although GaN does not require a “perfect material”, there are other challenges for designers. Technology papers presented over the past few years cite some of those challenges. Further, if it were easy, power conversion designers would already have devices and would be designing them into their circuits. He concluded by remarking that potentially, designing high voltage GaN into power conversion systems could be a simple process making rapid market adoption a likely scenario.
So from whence will they come? There certainly are a number of players working diligently on GaN technology. Besides the low voltage players, International Rectifier and Efficient Power Conversion, they include Fujitsu, Furukawa Electric, HRL Laboratories, IMEC, Panasonic, Sanken Electric, Toshiba, and Velox Semiconductor. A recent Google Scholar search indicates both Cornell and University of California at Santa Barbara are very active. Most others cite them both as leading the way to significant breakthroughs.
GaN conversation certainly dominated the Power Conversion and Intelligent Motion (PCIM) Conference in Nuremburg, Germany this year. In the race to be first, potential suppliers are announcing that they will have product “soon”. Some promise product by Q3 this year while others are sampling a few prototypes with product promised for 2011. My concern is that while promising product qualification, they are still discussing fundamental changes, such as maybe the product should be vertical rather than lateral. If we are going to get to qualified high voltage GaN product soon, then suppliers should put these issues to bed and provide product! I believe the successful winner in this race will be an entirely new player arriving on the scene with qualified products.
The bottom line is that GaN high voltage devices offer promise of a quantum increase in efficiency and power density. Compared to other materials, the GaN time-to-market will be much faster. Development with be less because a large fraction of funds will not have to be spent on fabricating “perfect” GaN material. This will enable companies will ramp up to product development and produce working devices sooner.
Unlike its low-voltage counterpart, high-voltage GaN performance is far removed from the market pressures that competitive silicon or silicon superjuction devices face. Thus, the likelihood of success is significantly greater - as long as the supplier has dedicated focus on the power conversion market. High-voltage GaN offers the remarkable opportunity to positively impact the complete ecosystem (economic, performance, and design structures) of possibly every high voltage power electronic product. Relative to other compound semiconductors, GaN-on-silicon high-voltage devices are the best and cheapest next step in power conversion, and therefore offers the most hope.