Pushing Power Limits: EPC and Cyntec on GaN for AI and Robotics

So Alex, the industry is moving towards 800-volt architectures, talking about data center, so megawatt scale AI racks.

And so increasingly distributed power conversion.

So which architectural shift do you see, do you believe will have the greatest impact over the next year? And what role we can play in enabling this one?

So with the 800-volt bus, which everybody is talking about, we've really split the function of power delivery into two different physical locations.

The sidecar will be taking AC offline, off the line voltage, and then we'll be converting it through a solid-state transformer to an 800-volt DC bus, which will then go on rails to the various server racks.

So the back of the rack now is receiving 800-volt DC.

So the next architecture is, how do you get from 800-volt DC to 0.5-volt GPU? Various opinions and various things are evolving.

One thing that I think is clear is that the 800-volt primary side is going to be done in a multistage ISOP topology.

That's inputs and series outputs in parallel.

It's a very good topology for this because basically you can take low voltage DC to DC converters, stack their inputs in series, stack their outputs in parallel, self-balancing.

Now, all of a sudden, you can get from, say, 800 to 50 volts or 800 to 12 or even 800 to 6 volts with very, very high conversion efficiency, 97%, 98%.

Whereas in the past, that might take a few stages.