Wide-bandgap device makers need to find high-volume uses

There s just one problem. Together, all of these applications still do not represent a large market. According to Yole Developpement, the French analyst company that specializes in wide-bandgap device markets (and which is currently putting together detailed reports on SiC and GaN technology for its clients), the prospective device-makers need to focus on other applications of the technology that will require much higher volumes.

Important recent developments that have driven the two technologies apart include Infineon Technologies, Germany, extending its range of SiC Schottky diodes and Japan-based Eudyna Devices progress with GaN HEMT RF devices. The latter is a clear indication that GaN has left SiC behind technologically as far as RF applications are concerned.

"Regarding RF and microwave applications, the latest work shows that GaN reliability, lifetime and linearity are more advanced than [that of] SiC components," said Roussel, who believes that this is principally due to the greater funding enjoyed by developers of nitride technology in recent times. So, where does that leave SiC? Schottky diodes are the only SiC devices on the open market, and Infineon s chips are used in a single application: power supplies for high-end PCs and servers, such as those used by telecom companies. The diodes are being supplied to at least two Taiwanese power-supply manufacturers. According to Roussel, the size of the SiC chip required presents the main obstacle to its deployment in low-end supplies: "The main issue is that you have to completely redesign the architecture," he explained. "It s not a chip-to-chip swap. That s why only the high-end market has been targeted."

Roussel is optimistic about the SiC devices, whose excellent heat properties mean that the need for expensive, heavy cooling fans and heatsinks in power supplies is much reduced. He believes that the mid- and low-end markets also present an opportunity. "We don t think that it will stay this way - it s just a question of price," the analyst said. Last year, Roussel estimated, the SiC-Schottky-device price was around $0.70 per amp. This year, the figure has dropped to $0.50 per amp. At $0.20 per amp, the low-end market could open up to SiC.

A number of developments will be required to drive prices down to this level. One key aspect that is starting to be addressed is the manufacturing switch from 2 to 3 inch wafers. Cree has announced its intention to do this and, as the US company is also Infineon s substrate supplier, the move will directly impact Schottky-diode selling prices. Roussel says that improvements in the speed and quality of epitaxial processes are also needed - he estimates the current device yield to be 70%. Roussel predicts that SiC Schottky diodes will breach the crucial $0.20 per amp figure in 2007, but admits that this could happen sooner if rival competitors to Infineon and Cree emerge before then. Competition looks likely, however, with the Switzerland-headquartered semiconductor giant ST Microelectronics and International Rectifier of El Segundo, CA, expected to enter the fray (see STMicroelectronics is set to drive SiC-device competition).

A pretty clear roadmap exists for SiC Schottky diodes, but there are some other high-volume applications that SiC can penetrate. "Everyone is waiting for [SiC] MOSFET and pin structures," said Roussel. Applications for such devices are likely in the transport market, for example, to replace some pin diodes currently used in high-speed trains that require a full silicon wafer to provide a single device. The problem with SiC in this area is the thickness of the epitaxial layer that is required, which is around 100 μm. Difficulties with defects, such as micropipes, has also meant that the devices take up a large area, with Cree managing to fabricate only 16 pin diodes on a 2 inch wafer.

However, micropipe problems could become a thing of the past thanks to impressive recent research by Japanese company Toyota (see Substrates and epitaxial growth dominate ECSCRM discussions). Although not previously noted for its SiC work, the Toyota research team has developed SiC substrates with a virtual absence of micropipes, which promise to move forward the development of SiC MOSFETs.

So, why is a Japanese car manufacturer like Toyota getting involved in the development of SiC substrates? The answer could lie in the future design of car engines, thinks Roussel: "Companies like Toyota are looking at this alternative power technology for hybrid electric vehicles." Something of a hot topic in Japan, innovative engines of the future could use a combination of SiC MOSFETs and Schottky diodes for drive electronics.

Another Japanese car-maker, Nissan, is involved in a US project in conjunction with SemiSouth Laboratories, a spin-out from the Mississippi State University, that is aiming to develop a prototype multichip power module made entirely with SiC-based control electronics by around 2010. The $2.5 million project recently received backing from the US Commerce Department through its Advanced Technology Program awards.

With GaN technology streaking ahead of SiC in terms of RF and microwave applications, the roadmap for this material ought to look pretty rosy. The problem is that wireless base stations, defense applications and satellite communication do not amount to much in terms of the volume of demand for wafers. Although the base-station market was estimated at $6 billion in 2003, this translated into roughly $650 million for PA modules. And even though 70% of PA costs arise from the transistors used, the market is dominated by silicon LDMOS, accounting for 85% of PAs in base stations today. Silicon-friendly Freescale, the US-based spin-off from Motorola, owns 60% of this chip market and appears unlikely to switch to any III-V solution.

Despite this, there is a clear opportunity for GaN transistors as they promise to reduce the cooling expense in base stations, which is one of the major costs encountered. Although current 3G base stations use LDMOS and GaAs technology to achieve the necessary 300 W output at 2-2.5 GHz, the required electronic architecture is highly complex, something that a GaN solution could capitalize on.

Defense is another area touted as ripe for GaN infiltration, with applications for advanced radar and X-band communications. For example, the US company Nitronex, as well as developing base-station solutions, has always been closely aligned with US military bodies such as DARPA. But again, the defense market is not one requiring high volumes, even though it would still be lucrative given the relatively small supplies required.

Roussel reckons that 1000 processed wafers will be enough for the approximately 150,000 RF transistors likely to be needed in 2006, and it s a similar story for satellite communications. With 50-100 satellites launched each year, and each satcom unit (including almost 50 PAs) needing about four RF transistors, the total market would only amount to something in the region of 15,000 GaN RF transistors per year. All of this begs the question: why build extensive GaN capacity for the RF transistor market? Another communications application may be one reason to do so. So-called very small aperture terminal (VSAT) technology looks to be an attractive tool for building communications infrastructure in developing countries such as China and eastern Europe, where the immense cost of digging trenches makes copper wireline and fiber-optic deployment unfeasible.

Today s typical VSAT transceivers use two MMIC modules operating at 2-4 W power, and are currently based on GaAs HEMT technology. With 0.8 million units sold in 2003, and the sector s compound annual growth hitting around 13%, VSAT manufacturers believe that the annual market potential could run into several million units over the next five years. Roussel says that GaN technology could help to reduce the cost and size of VSAT, because of the reduction in the number of passive heat-controlling elements that it allows.

Roussel understands that French media and technology giant Thomson MultiMedia is targeting a price of $300 for a low-cost VSAT module, meaning that the challenge for GaN device-makers would be to produce a 5 W MMIC costing only $50. This could be possible, he believes, through the use of GaN on-silicon technology that takes advantage of the low-cost substrate to enable 4 or even 6 inch wafer production.

In a more radical piece of thinking, Roussel says that the ideal application for GaN RF transistors, certainly in terms of high-volume production, could be sitting in the kitchen. Microwave ovens are very heavy, mainly because of the microwave-producing magnetrons required to cook the food. Replacing these magnetrons with GaN transistors is a serious possibility, insists Roussel - provided that there is a fall in the cost of transistors.

"It sounds funny, but we have heard Matsushita [Panasonic] suggesting that to bring down the cost of microwave ovens, you also need to reduce their weight," said Roussel. Replacing the magnetron with transistors would reduce weight and therefore transportation costs, and render the oven s rotating plate - which is often the cause of a breakdown - obsolete. The required operating frequency would be 2 GHz - the same as that of a base station - with a power requirement of 1 kW. Technologically, this is no problem; the issue will be to bring cost down. After all, magnetrons cost only $10, so the price per GaN transistor would need to be a challenging $3-5.

The rewards could be huge, however. With an annual market currently exceeding 50 million units ready to be tapped into, microwave ovens may just turn out to be the star application for GaN-based RF transistors.