Transphorm: GaN power devices prove their worth at higher voltages

At the first of the year I interviewed John Palmour, CTO at Cree, about the future of high-performance, high-voltage silicon carbide (SiC) in power switching devices. Recently I spoke with Carl Blake, VP of marketing at Transphorm, a gallium nitride power start-up that came out of stealth mode at last February’s APEC conference when it announced it was working with selected customers to deliver 600V gallium nitride (GaN) devices, and would have fully-qualified devices for sale at the end of 2011.

Here is some of the basic information that Transphorm has released about their GaN HEMT devices: The blocking voltage is 600V;  R_DS(on) is 310 mΩ (max) for the TPH2002PS and 180mΩ (max) for the TPH2006PS.

EDN: What process does Transphorm use?
Blake: We have GaN-on-silicon and we also have GaN-on-SiC.

EDN: National recently came out with its LM5113 eGaN FET driver IC designed especially to ease the challenges of threshold gate voltages on the eGaN devices from EPC. Does Transphorm’s process require a similarly tight tolerance on the gate voltage?

Blake: First, some background: GaN transistors are inherently normally on.  Since power circuit designers are comfortable working with MOSFETs that perform in a normally-off mode, what Transphorm has done is use a normally-on GaN HEMT combined with a low-voltage silicon device connected to the emitter (source) of the GaN HEMT – it’s a cascode connection. [Editor’s note: This is the same architecture that IR has announced for the next generation of 600V GaN devices it will introduce at the end of the year.] Nobody has figured out how to make a GaN gate that survives more than 6V.

By definition,  a normally-off device is one that you have to put a positive bias voltage on the gate in order to turn it on, and that positive bias can be as little as a half a volt – at least for the professors in academia who want to be able to claim that they’ve made a normally-off GaN device. However, in a practical application, a half a volt wouldn’t get you sufficient noise immunity. You really need a higher turn-on voltage. [Editor’s note: The following is Carl’s interpretation of how EPC’s product works.] What EPC did is add a gate section to their GaN device, but in GaN rather than in silicon. The problem with doing it in GaN is that you are still limited to 6V. In order to get the part fully turned on, you have to apply at least 4.5V to the gate. 1.5V is not much room to control the gate-drive signal over temperature, power line variations, and variations within the circuit.

EDN: The  LM5113 designed specifically address these drive challenges. Care to comment? 

Blake: There are consequences to trying to hit those tight tolerances. EPC chose one route, and Transphorm chose another.

EDN: Transphorm’s GaN power transistors and diodes have blocking voltages of 600V and above. Doesn’t that put you head-to-head with power SiC devices like Cree’s?

Blake: Absolutely!  But a SiC 6-in wafer costs $2500. If you use a SiC substrate, that’s your base cost. A silicon 6-in wafer costs $25, and GaN can be grown on silicon. That’s why competition from SiC is not a concern.

EDN: But Transphorm also has GaN-on-SiC parts?

Blake: Our initial parts were all made on SiC. Our initial parts were built on SiC because its crystal structure is a closer match to GaN and it’s a bit easier to build on it.  We could get high-voltage parts that were stable and show to customers and they could test in circuits.

EDN: The parts that you’ve announced you’ll introduce at the end of the year - 600V, fully qualified – which will they be?

Blake: We’re running both qualification programs in parallel. We’ll introduce the GaN-on-SiC parts, and then the GaN-on-silicon. Also, we’re planning on going to higher than 600V parts. going to higher voltage parts. When we release 600V Gan-on-silicon, we’ll then start making higher voltage parts with GaN-on-SiC and basically do the same thing.

EDN: So you use the SiC platform to bootstrap to the next voltage level?

Blake: Yep.

EDN: Why aren’t you going after the lower-voltage devices?

Blake: The higher the voltage, the more advantage there is to GaN. At 30V, a power converter has to be switching at higher than 1 MHz before there’s an advantage to GaN. And because GaN is a new technology, it’s not yet at the cost levels of silicon MOSFETs. So why would somebody by a new part which is more expensive if it only matches a silicon MOSFETs performance?

EDN: Because R_DS(on) is not significantly better at less than 200V?

Blake: Right. We looked at the market and asked, where did the parts add significant value for the customer that their customer will get a benefit that is worth paying more for. And that was easier to find at 600V than at the lower-voltage applications. But solving the problems of higher breakdown voltages was much more difficult.