Like many companies around the world, RF Micro Devices and Transphorm are moving from the development of GaN FETs to their production. How far have these firms got? Richard Stevenson investigates.

There are now more than a score of companies developing GaN transistors for power electronics. All these firms share similar goals – to be a major player in this sector, and deliver high-quality, profitable products – but their backgrounds are vastly different. Some are silicon power electronics manufacturers that understand the promise of GaN; there are also those that have developed GaN for RF applications, and are looking to diversifying into new, lucrative areas; and there are also start-ups, focused solely on high-voltage GaN products.

Two firms with very different backgrounds, but similar approaches to GaN product development and manufacturing roadmaps, are: Transphorm, a spin-off from the University of California, Santa Barbara; and one of the world’s biggest chipmakers, RFMD.  Insights into both these US firms, which plan to qualify products this year, were provided at the inaugural CS International conference held in Frankfurt on 4-5 March.

The state of play today

An honest, open assessment of where GaN transistor developers are on the road to production was provided by Yifeng Wu, Vice President of Product Development at the west-coast start-up Transphorm, which was founded in 2007.

He began his presentation by pointing out that while there is no doubt that GaN is a superior semiconductor for power electronics, and while silicon is widely believed to be the best platform for enabling this transistor to realise broad market penetration, the industry is still waiting for successful qualification of products operation at 600 V or more. Reports of success are rarely about products, and instead tend to focus on demonstrations of high blocking voltages, low static on-resistances, and the fabrication of large-diameter GaN-on-silicon epiwafers, or efforts to understand current collapse measurements.

Wu argued that although many GaN developers know that this wide bandgap material has the potential to outperform silicon, there is little evidence of this happening in prototype products that should be delivering real-world system application advantages. He put this down to a very impressive set of attributes for the latest CoolMOS technologies: “If you think you can easily out-perform state-of-the-art silicon, you are wrong.”

To take on the incumbent, Transphorm has adopted a vertically integrated approach, performing epitaxial growth, wafer processing, product design and the production of discretes and modules in-house. It is targeting multiple markets: motor drives, power supplies, solar invertors, and motor invertors in electric vehicles. The modules incorporate GaN diodes and HEMTs with an AlGaN barrier. They feature proprietary GaN buffer layers for low leakage and high breakdown voltages, and they draw on the company’s exclusive epitaxial technology and gate insulator designs, which are claimed to enable excellent dynamic characteristics.

“Our structure is not complicated,” explained Wu, “but a lot of effort went in to making it happen.” He revealed that the company has switched substrates, and is now focused entirely on making GaN-on-silicon devices. Initially, SiC was used, but this is far more expensive.

Transform has recently produced its first-generation GaN-on-silicon diode. Feng compared a version rated at 4A and packaged in a TO220 to SiC and silicon equivalents, which have higher turn-on voltages (1.47 V and 2.3 V, respectively, compared to 1.4 V). In addition, the GaN diode produces lower conduction loss that stems from this lower operating voltage and the device features: a lower cost than a SiC rival, thanks to the cheaper substrate; and zero minority charge, compared to 60 nC for a silicon diode. Reducing minority charge improves performance at high temperatures, and also allows the device to handle any spikes in supply voltage in a better manner. Testing shows that the performance of the diode is not impacted by 100,000 shots of 990 V spikes.

From normally off to normally on

One of the weaknesses of GaN HEMTs is that they are normally on devices, while customers prefer normally off variants, because they are considered to be safer. To address this wish, Transphorm pairs its normally on HEMT with a normally off, very fast, low-voltage silicon FET to create a hybrid that is normally off (see Figure 1). The result is a device that is compatible with silicon drivers, and claimed to combine fast switching with a low on-resistance.



Figure 1. Transphorm pairs its normally on, high-voltage GaN HEMT with a normally off, low-voltage silicon FET to create a fast, normally off hybrid device.

A novel form of wiring is used in this high-speed GaN switch.  If a conventional approach were adopted, it could lead to spikes in the gate-source voltage that result from drain current transients. To avoid this, separate source terminals are used for the gate and drain currents (see Figure 2). According to Wu, this form of wiring configuration, known as Kelvin source wiring, is not available with silicon power T0-220 MOSFETs.



Figure 2. A conventional wiring scheme (top) leads to drain current transients that can cause current-voltage spikes. Transphorm addresses this issue with an approach known as Kelvin source wiring, which employs separate source terminals for gate and drain currents (bottom).

Comparisons of the current and voltage waveforms associated with high-speed switching in a CoolMOS device and in Transform’s HEMT – both packaged in a T0-220 - reveal that spikes are significantly supressed with the wide bandgap variant. This result led Wu to claim that CoolMOS, housed in a traditional TO-220 package, is not suitable for high-speed, high-power operation. In his opinion, Transphorm’s approach, known as Quiet-Tab, extends the limits of TO-220 to a new operation space.

Additional strengths of this device include an on-resistance that is slightly lower than that for silicon CoolMOS, the ability to handle high voltage spikes (no change in device behaviour after 100,000 shots at 850 V spikes), and no compromise in performance after operation at elevated temperatures (no degradation observed after 1000 hours of operation at 175 °C).

In 2012, Transform submitted its GaN-on-SiC devices to qualification by Jedec, an independent, international-standards organisation. Wu describes these tests, which were passed, as very stringent: “If one device fails, qualification fails.”

The next step for the company is to extend its qualification to silicon-based Gan devices. “I am very positive that we will be able to qualify our devices this year,” revealed Wu.

He finished his talk at CS International by highlighting the benefits that the company’s devices could deliver in electrical systems. He began by considering the combination of GaN HEMTs and GaN diodes in a boost convertor, a circuit providing DC-to-DC power conversion with an output voltage exceeding its input.



Figure 3. Compared to CoolMOS in a generic TO-220 package (right), Transphorm’s HEMT that is housed in its proprietary Quiet-Tab package (left) produces far weaker spikes during high-speed switching.

Efficiencies of 99 percent were possible with this circuit that operated at 100 kHz. Compared with a boost convertor incorporating CoolMOS transistors and QSpeed diodes, device losses were cut by a third at full load (1.5 kW). When operating frequency increased to 500 kHz, the performance gap widened, with GaN cutting device loss by 70 percent, and the load having to be restricted to 1.3 kW for the silicon converter in order for this system to operate in a safe manner.

Benefits of GaN extend beyond efficiency. Chip cooling is needed to deliver the very highest efficiency, with the 1.5 kW, 500 kHz GaN convertor rising from 97.95 percent to 98.05 percent when air cooling speed increased from 0.5 m/s to 5 m/s. Delivering this load was not possible with silicon components, due to thermal runaway at the highest rates of air cooling. “[And] at 0.5 m/s, within 10 seconds the device exploded,” explained Wu.

He also compared the performance of bridge circuits made from different classes of device. These can be formed without diodes when built from GaN. The speed of these wide bandgap circuits is very high – compared with silicon IGBTs, rise time is three-to-five times less, while fall time is five times less.

This form of circuit has been used to build a three-phase GaN module for a motor drive invertor. This was the industry’s first high-frequency module capable of 300 kHz operation, and it enabled the use of compact filters, leading to a pure sine-wave output that reduces motor stress (see Figure 4).



Figure 4. The waveform produced by the inverter that features silicon IGBTs (left) has a significantly more noisy output than that produced by an inverter incorporating Transphorm’s GaN devices (right).

The benefits of Transphorm’s GaN are also making an impact in the solar industry. Last December, a photovoltaic conditioner built by Yaswaka and featuring Transphorm’s GaN power modules generated considerable attention at PV Japan. This 4.5 kW inverter was the smallest and most efficient at the show, with a peak rating of 98.2 percent.

RFMD’s GaN development

Applications for GaN products were also discussed by RFMD’s A.J. Nadler, General Manager of Power Conversion Devices, who began his presentation by reviewing what the market is worth and how fast it can grow.

Nadler quoted figures from the French market analyst Yole Développement: It calculated that the power semiconductor device market was worth $17.7 billion in 2011 and predicts that it will more than double to $35.7 billion in 2020. In 2013, the high-voltage segment of this market - which is dominated by silicon power MOSFETs and IGBTs, but will face increasing competition from GaN and SiC devices - is worth $14.6 billion, according to Nadler, who based this figure on internal research, plus that from Yole and IMS Research.

He argued that the primary reason why silicon will not be able to see off the threat from wide bandgap products is that, in terms of device performance, it has run out of steam. Improvements since the 1970s have come from the introduction of new device structures, with the silicon bipolar transistors being surpassed by the VMOS structures in 1978, followed by the trench-MOS, the super-junction MOS and finally super-junction deep trench device.

Now, according to Nadler, the time has come when market penetration of GaN begins: “I expect things to explode in the next decade.” He claims that his company is agnostic when it comes to technology, and argued that GaN will succeed because, when it comes to on-resistance, it is possible to build devices with this material that are not only one-hundredth the size of silicon equivalents, but a tenth of the size of those made from SiC.

RFMD has been involved with GaN for more than a decade, with efforts beginning with the development of RF devices. In 2001, the Greensboro-based outfit acquired GaN device developer RF Nitro Communications, and six years later it started to progress its GaN technology towards high-volume production. In 2009 and 2010 it qualified its first and second GaN process, and in 2010 it also diversified, commencing its efforts in GaN power electronics.

According to Nadler, several criteria must be met to enable the adoption of GaN in high-power electronics: the creation of a trusted, high-volume supply chain; high reliability for every device; improvements in efficiency, coupled with system cost savings that make a compelling case for buying the relatively expensive GaN components; ease of use; and a roadmap to lower-cost products. He emphasised the need for reliability by showing a picture of a 1958 transformer that was still in service, indicating that products such as this may have a replacement rate of just two per century.

RFMD is aiming to meet all these requirements with products based on its rGaN technology, which is claimed to offer efficiency-sensitive power conversion up to 900 V. Its first offering, a 650 V FET that also features an ultra-fast free-wheeling diode, operates in a normally off fashion, has an on-resistance of just 45 mΩ and can handle up to 30A. “We are now sampling [this] to customers, and will qualify this product this year,” said Nadler.

One market that RFMD is targeting with its FET is the electrical system used to convert mains into a DC output for powering datacentres and high-end telecom infrastructure. This might be used, for example, to convert a three-phase AC input at 480 V into a 380 V DC output. Replacing silicon devices with those based on GaN delivers a one percent increase in efficiency. How valuable is that? Very, according to Nadler, who calculated a pay back period of just 15 months, based on the following assumptions: silicon switches cost $0.15/A, while GaN equivalents are double that; power costs $0.10 per kWhr; rated power per phase results from the supply of 20.8 A at 160V; and the units are run for 24 hours per day, and operate, on average, at half power.

RFMD is now considering the substrate that it using to make these devices. “Everything we are doing is on SiC, but we are substrate agnostic,” revealed Nadler, who said that the company is now looking at GaN-on-silicon. Transform has just made that transition, and many others – including the likes of International Rectifier – are already there, enjoying the benefits of the cheaper substrate. This leads to lower production costs, diminishing the additional cost of GaN devices over their equivalent silicon incumbents. The increased cost-competitiveness that results will help to drive the revolution in power electronics that RFMD, Transform, and many of their peers are tipping to take place throughout this decade.