GaN devices have the potential to capture a billion dollar power market for electrified vehicles. But if they are to be successful, they need to sell for little more than silicon products, have higher threshold voltages than they do today, show no signs of current collapse and deliver very high currents, argue Ming Su and Chingchi Chen from Ford Motor Company.

During the next decade a growing desire to increase automotive fuel efficiency and trim carbon dioxide emissions will spur sales of electrified vehicles. In turn, this will create a multi-billion dollar market for inverter power semiconductor devices.

This new market has already whetted the appetite of many wide bandgap device developers, as they aim to beat silicon hands-down with ‘thousand times’ better physical figures of merit. However, in the eyes of some automotive engineers, the case for discarding silicon is not as strong as some wide-bandgap device makers believe.
 
For example, some wide bandgap chip developers argue that their new, pricier devices can save money at the system level, because they can help to eliminate a dedicated power-electronics cooling loop that is needed to prevent silicon diodes and transistors from overheating. However, these engineers may not be aware that for a large automaker, the cost of the cooling circuit is a small fraction of what is paid for the silicon in a full hybrid electric vehicle (HEV) inverter. In reality, automakers will not be willing to pay much more for a new type of device, even if it does offer better performance.                                      

What is beyond question is that the prices of HEVs will have to get progressively closer to those of conventional vehicles to fulfill the anticipated boom of electrified vehicle sales. Power semiconductor manufacturers can play a critical role in making this happen. In particular, those developing GaN-on-silicon technology promise to combine the capabilities of wide bandgap semiconductors with the prices of silicon products. But are these GaN developers fully aware of the requirements of electric vehicle makers, and how they should tune their devices so that they become more attractive for the HEV inverter system?

Key pre-requisites for GaN devices, if they are going to compete side-by-side with silicon IGBTs, are a high-voltage rating for switches in the 600 V to 1200V range and growth on silicon substrates with sufficiently large diameters. As of today, 1200 V class devices have only been produced in labs, due to two constraints: The thickness required for the breakdown voltage and the mismatches between GaN and silicon. There are yet to be any demonstrations of scaling production of such high voltage transistors to 6 inch or larger wafers, an essential step for cost-effective device manufacturing.

Insufficient threshold voltages?

Other important requirements for switches in motor drive applications are that they are normally off and that they have a gate threshold voltage (Vth) large enough for noise and temperature margins. To meet these demands, modern silicon IGBTs are designed with a Vththat is at least 5 V.

If GaN devices are going to appeal to makers of electric vehicles, they need to mirror the gate control characteristics of the silicon IGBTs that they can supersede. Aside from the more complex cascode structure to hybridize silicon and GaN, the popular approaches for making normally off GaN switches all face challenges. MOSFETs can be made normally off, but the low channel mobility leads to high resistance. Meanwhile, gate injection transistors (GIT), which have been developed by Panasonic and feature a p-AlGaN cap layer for enhancement-mode gate control, have a limited gate junction potential that makes it very challenging to reach a high threshold voltage.

Another popular technique for constructing normally off transistors is that of gate recess etching on a HEMT structure. Lacking a good etch stopper in the AlGaN layer, this approach demands high-precision depth control, which could lead to a very low manufacturing yield and a wide spread in Vth distribution. In order to preserve the two-dimensional electron gas channel in the gate region, AlGaN must be removed with great precision.

Normally-off GaN transistors can also be formed with a hybrid MIS-HFET architecture. In this type of device, low gate channel mobility can substantially impact the device on-state resistance. Furthermore, it is possible to bring Vthinto the positive domain, at around 1V, by treating the gate region with fluorine ions. However, concerns over stability hinder wide acceptance of this method and its capability to reach higher threshold voltages.




Fig. 2 Many challenges must be overcome in order to produce a robust, high-performance GaN transistor

The performance of devices produced by leading GaN research teams with architectures outlined above are summarized in Figure 1. These results show that the Vth level is generally too low for use in electric vehicles, and their on-state resistance is yet to be further reduced. Nevertheless, encouraging progress is frequently reported, such as a 1200V 7.1 mΩcm2 hybrid MIS HFET in 2011 by Nariaki Ikeda from Furukawa Electric Company, Japan.



Fig. 1 There are very few reports of normally off GaN transistors with a threshold voltage of 2V or more. A threshold of 5 V or more is highly desired by makers of electrified vehicles

Reliability issues

One of the biggest issues afflicting GaN HEMTs is current collapse. There are many ways to suppress this, such as SiN passivation, the addition of a GaN cap layer and the introduction of field plates. However, many researchers were only able to combat current collapse for devices operating at half or less of their breakdown voltage. Since the peak surface electric field plays a dominant role in the severity of current collapse, and becomes stronger at high voltage bias, further development is required to enable devices switch at up to the rated voltage.

Device robustness also needs to be proven. It is possible that long-term electron trapping and degradation of on-resistance will not be detected by the conventional dynamic Ron test. Instead, failure criteria will be defined in the component qualification tests.

Those within the GaN community rarely talk about the short-circuit withstand capability of the transistor. In traction inverters, a short-circuit event occurs when the motor windings are accidentally shorted. This subjects the on-state power switch to high drain bias and pushes it into a fault current mode. Extreme power dissipation follows, which must be halted by a prompt turn off of the gate drive circuit.

The typical required short-circuit tolerance time is 10 µs, depending on the gate drive reaction speed. Since power dissipation density in this event is proportional to electric field strength and current density, GaN HEMT structures are expected to withstand much shorter duration than a vertical silicon IGBT. That’s because these wide bandgap devices have a field strength in the drift region that can be ten times higher, in addition to a highly concentrated current path in the two-dimensional electron gas. To address these weaknesses, researchers will probably develop advanced gate protection circuits that reduce the tolerance time needed.

Last but not least, makers of electric vehicles have concerns relating to optimal gate dielectrics and achievable current ratings. GaN RF devices do not require an insulated gate, and GaN is not as fortunate as silicon or SiC to have a thermally grown native oxide. As researchers work with deposited insulators such as SiNX, Al2O3 and SiO2, leakage currents, interface states and long-term stability all come into play and make the search more challenging.

Furthermore, to be used in motor drive circuits requiring 300 A to 600 A rated devices, a single-die current capability of at least 100 A is needed to enter this market. GaN HEMTs are not naturally suited to such high currents, due to the lateral nature of the device, and on top of this, there are challenges associated with hitting high manufacturing yields for GaN-on-silicon technology, which carries orders of magnitude higher defects than silicon.




An integrated power control unit for the HEV inverter, containing silicon IGBT modules shown in the inset

However, despite all these concerns, GaN power devices are undoubtedly a promising technology for the electrified vehicle market. A highly competitive silicon IGBT industry sets high standards for challengers, but GaN material and device scientists have a chance of success if they acquire a strong understanding of what electric vehicle manufacturers need.

In short, wide bandgap device performance is a beautiful promise, but in-depth knowledge of customer needs must be taken on board if this ever-improving technology is to spawn a billion dollar market.