Silicon alternatives offer a new way to boost power device performance

4 min read

Over the past 40 years, the biggest advances in power conversion efficiency have come from improving conversion technique, although important enhancements have also been made to the characteristics of the silicon used.

But silicon power devices are reaching their theoretical limits in terms of on resistance (Rds(on)) and gate charge (Qg) for mosfets and of forward voltage and reverse recovery time in diodes. Improvements in these characteristics will only be made more slowly. Worthwhile performance improvements are more likely to come from alternatives to silicon – and the most promising are silicon carbide (SiC) and gallium nitride (GaN). These wide bandgap materials have properties – such as critical field, electron mobility and thermal conductivity – that not only allow power converters to operate at higher voltages, higher temperatures and higher frequencies, but also to reduce energy losses substantially. This, in turn, promises to enable smaller and lighter power supplies. In spite of the favourable properties of SiC and GaN, however, chip manufacturers are yet to develop broad product portfolios featuring these materials or to gain wide adoption for them. So what is stopping these materials from replacing silicon? And at which applications should SiC or GaN devices be targeted? SiC technology The theoretical advantages of SiC technology are obvious, but there are several barriers hindering its exploitation in power electronic circuits. While an SiC device will operate effectively at high temperatures, this means higher temperatures at the heat sink. When other silicon components must be mounted on the same heat sink, the use of SiC becomes impossible. But, by combining SiC diodes with a SiC mosfet, this problem is overcome. Just because a power semiconductor has a high junction temperature rating, it does not mean it has to be driven at or close to this temperature. In fact, the weak link in power circuits is more likely to be soldering or wire bonding lifetime. Solder or wire bond cannot tolerate high temperatures, so the chip's ability to tolerate them becomes worthless. Meanwhile, SiC manufacturing capacity is limited, which tends to constrict the supply of finished SiC devices. There is, however, a device for which SiC is advantageous: a SiC Schottky diode has ideal dynamic behaviour. In hard switching applications, switch-on losses are influenced mainly by the diode's reverse recovery time (trr), which is extremely short in a SiC Schottky diode. This keeps the reverse recovery charge (Qrr) very low, reducing switching losses significantly (see fig 1).

At the same time, a short trr enables faster switching, meaning smaller passives. Lower power losses mean the cooling system (heat sink and/or fan) can be smaller, so replacing a silicon fast recovery diode with a SiC diode in a switched mode power supply contributes to end product miniaturisation. In addition, SiC diodes are more stable with temperature than silicon based devices (see fig 2). A constant forward voltage over the temperature range simplifies the design of parallel connections and prevents thermal runaway. A constant trr enables high temperature operation without increasing switching losses. Switching speed and temperature stability are important mosfet characteristics and SiC devices are inherently superior on both counts than silicon and manufacturers such as Cree and Rohm are offering 600V and 1200V SiC mosfets. For example, Cree's first 1200V power SiC mosfet, the CMF20120D, has a maximum Rds(on) of 130mO over the whole temperature range and a typical gate charge of less than 100nC. It should not be surprising that 1200V SiC mosfets have appeared: they compete with igbts, which have dramatically higher switching and on-state losses. Rohm is in production with the first 1200V 'Full SiC' power module, which combines a half bridge SiC mosfet with parallel SiC Schottky diodes for a rated current of 120A. This module, half the volume of a conventional silicon module, can operate safely in temperatures up to 200°C. Implementing GaN GaN, meanwhile, has some favourable attributes when compared to silicon, but the main problem is the economics of production: pure GaN is only available on 2in wafers. In addition, the production processes for high quality epi-ready GaN substrates are less mature than those for growing SiC substrates. While an SiC epitaxy is grown on a SiC substrate, the trend is to grow a GaN epitaxy on silicon. Because of differences between the two crystal lattices, a buffer layer has to be added between them (see fig 3). This complex process results in a higher cost per die for GaN than for a silicon equivalent. Nevertheless, GaN-on-Si devices appear to have found a sweet spot: dc/dc power conversion at less than 100V. Because a GaN dc/dc converter can operate at several MegaHertz, it enables higher power density since it requires smaller associated components. Another advantage is that GaN fets do not exhibit body diode characteristics, which means there is no reverse recovery loss. This increases conversion efficiency, which in turn reduces the amount of waste heat generated. With a smaller requirement for heat sinking, system size can be reduced. International Rectifier (IR) was the first company to offer GaN-on-Si power switching devices, introducing a complete output stage (high and low side drivers and GaN fets) for use in point of load dc/dc converters. The iP2010/11 features an input voltage range of 7V to 13.2V and an output voltage range of 0.6V to 5.5V. Output current is up to 30A with frequency up to 5MHz. A discrete GaN transistor is also available from Efficient Power Conversion. The EPC2010FET is a 200V GaN FET with on resistance of 25mO at 5V and a pulsed current rating of 60A. While most GaN products available today are suited to low voltage dc/dc conversion, several manufacturers have announced products that support drain to source voltages as high as 600V. One of these, IR, will offer a discrete switching device that pairs a low voltage silicon fet in series with a 600V normally on GaN HEMT (high electron mobility transistor, see fig 4) in a cascade configuration. This means the device can be driven as if it were a classic normally off mosfet, since it allows the use of standard gate drivers that require no special consideration of voltage limitation, over voltage limits or reliability.

Overcoming the last barriers For a wide bandgap material to gain widespread adoption, a sustainable cost of $3/cm² for the substrate and epitaxy layers must be achieved, with wafers of at least 150mm in diameter. To hit this target, a manufacturer must be in high volume production using a high yield fabrication process which is scalable and flexible. This explains why IR has bet on a heterogeneous GaN on Si epitaxy technology: processing can be performed on a standard silicon manufacturing line with little modification to either equipment or process. So, while SiC devices have been on the market longer than GaN parts and currently enjoy a higher level of sales, GaN seems to be better equipped to meet the economic conditions necessary for commercialisation of a new wide bandgap material. Antoine Descazot is a power and analogue field application engineer with Future Electronics (France).