A compound of silicon and carbon, silicon carbide - close-up

Silicon carbide and gallium nitride set to oust silicon circuits

Silicon's long-held dominance as the IC material of choice is being challenged. Novel new compounds, such as silicon carbide and gallium nitride, are enabling electronics power switches that will not break down quite so easily.

If the history of technology has taught us anything in the field of microelectronics, it's that silicon is a winner. It has displaced most other materials, and only been kept out of applications such as photonics by physical properties that make it innately unsuitable. Where silicon can compete directly with materials that offer better nominal properties, it almost always wins. However, when it comes to power electronics, silicon may well have met its match.

As a material for power electronics, silicon has done quite well, but it has a weakness: when hit with a high voltage, it breaks down and starts to conduct uncontrollably. If power is not removed from the device it will simply burn itself to death, with a strong likelihood of roasting the other electronics that it was meant to protect. It is possible to boost the breakdown voltage – the point at which the transistor stops being able to control power – but this leads to higher resistance when switched on, which means more heat and power wasted.

To deal with the breakdown issue, while working at General Electric in the early 1980s Indian electrical engineer Dr B Jayant Baliga invented the insulated-gate bipolar transistor (IGBT) – a hybrid of the two most commonly-encountered types of transistor used now. For the most part, it behaves like the bipolar transistor that was first invented by William Shockley in the 1950s; but it has a gate, sitting behind a fairly thick layer of insulation, much like the metal-oxide semiconductor field-effect transistor (MOSFET) used in every computer.

This blended design has made it the power device of choice where breakdown is a problem. A hybrid electric car may have tens of IGBTs that are used to convert battery power into drive power – and back again.

In general, an IGBT with a given breakdown voltage will beat a power MOSFET in terms of on-state resistance; but MOSFETs are more flexible and much faster. This has become increasingly important in power designs, as circuits that switch faster tend to be smaller. For one, they can get away with smaller passive components, such as inductors and capacitors that form a necessary part of most power circuitry.

Herv' Branquart, director of automotive solutions at ON Semiconductor, says while there is certainly scope to improve IGBTs, "there is a limit to integration... One option is to use other materials than silicon". Power-device designers have looked closely at the elements that surround silicon in the periodic table for possible alternatives. There is no clear winner in any of them – but some alloys of these elements look very promising indeed.

Drew Nelson, president and CEO of specialist UK-based wafer maker IQE, claims that with compound semiconductors, "you can engineer practically any property that you want". He adds: "When you look at the material properties of compound semiconductors they are fundamentally better than silicon."

Two key materials that stand out for use in power devices are silicon carbide (SiC), and gallium nitride (GaN). One key advantage that GaN and SiC have over silicon is that, for a given on-resistance they do not break down as readily as silicon due to a much larger band gap – the energy it takes to move electrons out of bonds with atoms in the crystal lattice and turn them into unbound, conduction electrons.

The strength of electric field that a device can resist rises roughly with the square of the band-gap energy, so that both SiC and GaN with a band gap more than three times higher than silicon can withstand a field ten times stronger. This allows devices to be made smaller for a given breakdown voltage requirement.

The Figure of Merit pulls together a number of factors that govern the effectiveness of power semiconductor. There are several in common use now, although designers commonly employ the Figure of Merit proposed by Dr Baliga in 1989. This focuses on electron mobility and band gap, and so helps predict how much energy a device will lose when conducting. Because GaN has much higher electron mobility than either silicon or SiC, but its band gap is not very different to that of SiC, it tends to win on this figure of merit.

Conductive channels

Electrons move differently through GaN than through silicon. Rather than being made of a single GaN crystal, careful doping by impurities such as aluminium creates a sandwich structure that lies parallel with the gate electrode above it. When a voltage on the gate allows a conductive channel to form, the electrons that are able to move are forced into the interface between these layers, so that they can move from side-to-side but not up-to-down, as it were. The result is what physicists call a two-dimensional electron gas – electrons in this gas move far more easily than in the rest of the material, making them highly conductive – just as long as the surrounding electric field allows the gas to form.

As well as GaN, International Rectifier (IR) considered using silicon carbide, says director of systems and applications Marco Palma: "We decided to go straight in the direction of the best solution, to GaN. We started work on this technology in 2008." IR's first GaN devices are going on general sale this year, but "we are already in production with some major customers", Palma reveals. By using GaN, IR was able to not only almost halve the size of a 400W module by putting the guts of three-phase inverter into a single-chip package but increase its blocking voltage to 500V, and remove the need for a heatsink. "That is only the beginning of the revolution," believes Palma.

IR is not alone and some companies are already in production. The first company to launch GaN-based MOSFETs for power circuits was Efficient Power Corporation (EPC), co-founded by Alex Lidow, the former CEO of IR. Transphorm launched high-voltage GaN devices in 2012 and has raised enough funding to acquire part of Fujitsu Semiconductor together with the Japanese company's fab capacity for GaN. Canada-based GaN Systems is working on power transistors able to deal with more than 100A at high voltage, and it expects to roll-out products this year.

Not everyone is convinced by the value of GaN. UK-based Anvil Semiconductor, a spin-out of the University of Warwick, has instead turned to SiC. Although it will not switch as quickly as GaN, SiC has one important advantage in an environment where even the most efficient devices generate copious amounts of heat. Its thermal conductivity is more than three times higher than that of either silicon or GaN. Heat produced by the devices can quickly move to the heatsink and out of the lattice, making it possible to have transistors that work happily at 300°C, rather than at the 125°C at which silicon often tops out.

GaN also deals well with high temperatures but the market seems to be splitting into two distinct groups. Power devices based on SiC could take over from the large, specialised components such as thyristors needed in the electricity distribution grid or locomotives, with GaN focused on smaller components controlling the power that goes into individual motors and industrial equipment. Professor Phil Mawby, chair of power electronic, applications and technology in energy research (electrical and electronics division) at the University of Warwick's School of Engineering, draws the line between SiC and GaN at a capacity of around 10kVA to 100kVA.

EPC's Alex Lidow claims that the advantages of GaN are on its side against silicon, arguing that the GaN transistor costs will fall to parity with silicon by the end of 2015 as volumes rise as the main differential in cost today is the time and effort required to deposit thin layers of the material. Most of the other processes are, in fact, very similar to those of silicon.

Compound-on-silicon growth

The breakthrough for GaN came with the ability to put the material on top of standard silicon wafers. Originally, GaN power devices were constructed on a sapphire substrate, which is expensive to make and work with. Not only that, the transistors were primarily depletion-mode devices – they need a voltage on the gate to turn them off. For safety and reliability reasons, power switching devices should be designed such that they are normally turned to off. Thanks to the move to silicon wafers, GaN MOSFETs are now appearing that offer enhancement-mode, normally off operation; but it was not a simple transition.

GaN and silicon have very different crystal structures. When one is grown on top of the other, these differences manifest themselves as stress that usually results in the thinner layer splitting apart – and cracked semiconductors do not work very well.

The materials science that led to the speed improvements of strained silicon in logic transistors has been translated to the growth of GaN. Buffer layers and careful control over the deposition process have now made it possible to grow thin layers of GaN on silicon without it turning into a form of microscopic crazy paving. EPC's Lidow says that the epitaxy process used to grow GaN is more expensive than that for silicon right now, but volumes will tend to push the cost down and bring more advanced devices into direct competition with silicon.

The availability of silicon processing for GaN is one reason why it may encroach on SiC's core market. SiC has a structure very similar to that of diamond – and a hardness to go with it. Manufacturers have had to develop novel ways to cut through the material, using high-powered lasers for example, to build power semiconductors the way they have with silicon.

University of Warwick spin-out Anvil Semiconductor says that it has developed a way of depositing SiC layers on silicon wafers, that the company claims will lead to a step reduction in the cost of manufacturing power switches. According to Anvil, the process will make it possible to produce SiC devices at costs close to those of conventional silicon.

For once, silicon looks as though it is going to give way to other materials in the quest to improve power efficiency – although it seems destined to continue in the game by providing the literal foundation for these new classes of device.

Further information

Sign up to the E&T News e-mail to get great stories like this delivered to your inbox every day.

Recent articles