At much higher powers and voltages, things start to change. Engineers have found it makes sense to crossbreed the two types of transistor, producing hybrids, such as the insulated gate bipolar transistor (IGBT, see fig 1). This blends the FET approach to transistor control with the bipolar junction in an attempt to overcome problems that MOSFETs have at high power.
The blended design of the IGBT has, so far at least, made it the transistor of choice in systems that need high density, such as electric and hybrid cars. A Toyota Prius has a 50kW inverter built around tens of IGBTs, each of which is made up of an array of transistors on a single chip that work in parallel.
It is entirely feasible to use high power MOSFETs in electric vehicles, solar generators and other power delivery systems. The big advantage of the IGBT over the MOSFET is that it generally exhibits a lower voltage drop when switched on. This wastes less power, especially if the transistor has to be rated for breakdown voltages that approach 1kV.
In power MOSFETs, the on state resistance tends to increase with breakdown voltage – the point at which the voltage difference across the device makes it unable to block current flow when supposedly turned off by the gate. This increase is due to need to increase the resistivity and thickness of the drift region between source and drain so that it can resist breakdown.
At high voltages, intense electric fields can develop between the individual parts of the device which, to improve power efficiency, have to be packed more closely together. The higher the field strength, the more likely it is that the device will break down.
The answer in this case is to move one element – the drain – away from the gate and the source components onto the back of the chip. This allows many transistors to be packed tightly together typically in a hexagonal array to maintain a low overall resistance, thanks to the parallel design (see fig 3). This extra distance, all the way through the wafer, keeps the electric field between the source and drain sufficiently low. However, the carriers now have a comparatively thick and potentially resistive drift region to cross to make it to the drain.
The IGBT does not suffer from the resistivity increase. As a minority carrier device when it is switched on, a high concentration of those carriers in the drift region lowers overall resistance. The forward drop depends primarily on the thickness of the drift region and is largely independent of the material resistivity.
The problem for the IGBT – and the reason why it has not entirely supplanted the power MOSFET – is that it cannot switch as quickly. There is one other drawback: it can only conduct in one direction but, typically, this is not a practical restriction as circuit designers can simply use a 'freewheeling' diode in parallel that points in the other direction.
Another way to look at the IGBT is as the successor and alternative to another device used in very high power systems: the silicon controlled rectifier, or thyristor (see fig 2). Based, like the IGBT, on a junction transistor it is a four layer device, with an extra p-doped region added to an npn sandwich. Unlike a transistor, it stays on until the current is removed. Until comparatively recently, this could also be an issue with IGBTs as devices tended to suffer from unwanted latch up if pushed too far.
The thyristor is usually more complicated internally than just a sandwich of semiconductor layers. As temperature increases, so too does the voltage drop across the device. Not only that, the current needed before the thyristor triggers reduces. This means that, without protection for the gate triggering circuitry, spurious noise can lead to the thyristor switching on unexpectedly. Conversely, in a cold system, the triggering circuitry needs to be sensitive enough to turn on when required.
To provide more control over switching, a commonly used variant of the thyristor is the gate turn off (GTO) device. It achieves the effect by firing a negative voltage pulse between the gate and cathode terminals which makes the forward current fall so the thyristor will switch off. Typically, GTO thyristor will switch off more slowly than the silicon controller rectifier form, restricting the switching frequency to around 1kHz.
In use, the primary practical difference between a thyristor and an IGBT or MOSFET is that the former is good for AC circuits; the latter for DC. However, as its current handling ability improves, the IGBT is gradually taking over from thyristor – using different circuit designs – to take better advantage of the higher efficiencies offered by faster switching times. In very high power systems, such as railway locomotives, where thyristors have dominated for years, this is leading to smoother acceleration. Older trains needed 'geared' GTO thyristor circuits which were switched in and out as the locomotive picked up speed – something you could hear clearly as it pulled out of a station. The improved switching control provided by IGBTs means the gear switching noise has largely gone from newer vehicles.
Like thyristors, IGBTs can suffer from temperature effects. In this case it's thermal runaway. The non punch through (NPT) structure prevents this from happening. The name comes from a change in the manufacturing process. Where IGBTs can have an N+ doped buffer layer, which blocks an electric field that has 'punched through' the lightly n doped drift layer above it, NPT devices lack this buffer layer. Instead, they are designed not to let the electric field permeate all the way through the drift region and therefore 'punch' through into a lower layer.
NPT devices switch more slowly and, typically, cannot be used at such high voltages as the PT variants because of the electric field requirements. However, manufacturers have gradually been pushing the switching speed of NPT IGBTs to the point where they can supplant MOSFETs.
One of the developments that has made the NPT IGBT competitive in terms of speed is the use of thinned wafers. By thinning wafers to 100µm or less, it is possible to use a very lightly doped collector region, which normally sits on the back of the thinned wafer. Lighter doping reduces the amount of charge that can be stored in a device which it is conducting, which translates into better switching performance as fewer carriers need to be killed when the device is switched. Conventional PT devices use minority carrier killing techniques to improve their switching performance, which tends to translate into a more negative temperature coefficient – increasing the potential for thermal runaway.
Companies such as International Rectifier (IR) and STMicroelectronics have been making devices as thin as 85µm for a number of years and are moving to 60µm substrates – at these thicknesses, the substrate is more like a foil than a rigid disc – for high frequency IGBTs.
To obtain better performance, manufacturers are now looking at the use of alternative materials to silicon. Strong candidates are wide bandgap materials such as gallium nitride (GaN) and silicon carbide (SiC) as they allow reliable operation at high temperatures and so can be packed more tightly in modules with less need for cooling within the module – attractive features for electric vehicles.
Manufacturing cost is an issue, so companies are investigating different combinations. IR is looking to deploy GaN on silicon, while ST is working on using glass as a substrate for the semiconductor – at this stage for power MOSFETs in basestations, rather than IGBTs.
Breakdown voltage presents another obstacle. Many IGBTs operate with a 600V breakdown voltage, but the trend is to push that to 1200V. A substrate change could help here as well.
Using sapphire as a substrate, engineers at Matsushita managed to push the breakdown voltage for a power transistor from 2kV to almost 10kV, something they thought could not easily be done, even with SiC or GaN. But they had to develop a new way to punch holes all the way through wafers made from sapphire to do it.
The Matsushita design called for a mixture of aluminium and GaN to be deposited on a wafer of sapphire, which is a highly effective insulator. The problem lay in defining a conductive path between the drain and source. With silicon, holes can be etched all the way through chemically and then filled with metal. Unfortunately, sapphire is much more resistant to this process. Matsushita decided to use a high energy laser to drill the holes. The problem with using lasers is that it takes a long time to drill individual holes in a wafer that needs thousands of them.
So the Matsushita team used diffractive optics to perform parallel drilling, reducing the overall time to just 10s. The optics split the beam from a pulsed laser and focus them onto the wafer, slashing the number of steps the drilling tool needs to pepper the surface with holes. After that, conventional techniques could be used to fill the drilled vias with material.
As electric power becomes more pervasive, supplanting chemically powered engines in vehicles and generators, you can expect transistor designers to find even more novel ways of pushing up power ratings and densities.
