Roland St Pierre from Power Integrations talks to PSD about the industry's first 2,200V GaN technology.
The 2,200V GaN switch will expand the company’s PowiGaN platform for power conversion up to bus voltages that the data centre and automotive industries are planning to adopt in the mid-term future. The announcement was a technology reveal rather than a product launch, with the first physical products expected to be announced as early as next year.
PowiGaN has steadily climbed in voltage capabilities since its introduction. The company has a 1250V version that is intended for 800V data centre supplies, 800V EV battery architectures, renewable energy and other applications that use voltages around that level. It also has a 1,700V version available for axillary power applications. The migration towards ±400V/800V DC distribution is intended to increase efficiency, as well as reducing the cost and size of copper conductors. The EV industry and data centres are looking to 1,500V as the next node for the same reasons, and Nvidia has already published papers making the case for migrating to higher voltages. There are also applications available today, such as 1,200 V automotive systems, 1,500V photovoltaic installations, and auxiliary supplies for emerging high-voltage DC transmission that need devices with a higher blocking voltage than current PowiGaN devices can provide.
The new technology will be fabricated on the company's proprietary GaN-on-sapphire process and with its depletion-mode (d-mode) GaN HEMT architecture. Because the HEMT is normally on, it is paired in cascode with a low-voltage silicon MOSFET that performs the actual switching while the GaN handles the full blocking voltage. In practice, this means that the device can be driven easily, while the alternative enhancement-mode GaN devices are notoriously difficult to drive and vulnerable to overvoltages. The cascode design can be driven like any silicon MOSFET at 12V, 10V or even 5V. Third-quadrant operation is also more efficient than e-mode alternatives, which is important for bridge topologies.
The measured off-state breakdown voltage of the new technology is actually 4,000V, but it is intentionally overspecified to guarantee reliability. Semiconductor failure-in-time (FIT) calculations use temperature and voltage as the two acceleration factors, so the gap between the published 2,200V rating and the point where leakage begins to climb is deliberate headroom created to ensure very low FIT rates. Power Integrations already holds automotive qualification for its GaN devices and points to its HTRB FIT performance that it claims is unmatched. The extra margin also provides safety for one-time events and dynamic RDS(on) stability, as well leaving headroom for degradation if the device has to be operated above 2,200 V continuously.
The higher voltage technology brings another less obvious benefit in that a higher primary-side rating allows a larger reflected voltage in a flyback, which reduces stress on the secondary-side synchronous rectifier. Low-RDS(on) SR MOSFETs above 150V are scarce, so being able to stay with a 100V part while handling very high input voltages is a real benefit to designers.
To prove the 2,200V concept, Power Integrations conducted a practical test with a flyback supply of roughly 40W, and a peak current 1.5A. It used a GaN switch mated to a suitable low-voltage silicon MOSFET in an InnoSwitch package. Input voltage was raised to 1,500V, which is 80% of peak rating. The converter ran continuously for more than 20 hours, with normalised RDS(on) monitored across the cascode found to be flat, indicating no dynamic RDS(on) effects under real power-supply stresses.
The practical test was intended to demonstrate that GaN can offer an alternative to silicon carbide at higher voltages. The lower switching frequency of SiC provides lower efficiency. SiC exists at 1,200V and 1,700V, but practical switching tops out around 200 to 400kHz, whereas GaN can run beyond 1MHz. That higher frequency allows the shrinking of the size of the circuit’s magnetics dramatically.
Roland St Pierre, VP Product Development at Power Integrations summarizes by saying, “this technology is future proofing. Applications are currently emerging, but there will be a point where power conversion directly off 1500V is desirable. A 1500V LLC that can operate at a megahertz, is an amazing technology to be able to deploy. Right now, the only technology that can achieve this is a multi-level converter using 650V GaN devices, and that is complex. We can eliminate those extra bias supplies, floating gate drivers and additional parts that are required for such a topology, and still be able to enjoy high frequency conversion. It has the benefit of very small magnetics because it operates at higher frequencies than SiC designs. And, our customers tell us that one of their main bottlenecks is the magnetics, along with the high current SR MOSFETs, and the PCB.”
