Let's do more compute yes? Well, it's not so simple anymore. Every chip generation now buys performance by cramming more transistors into less space and running them harder, which means more heat in a small space.

The solution is, as we all know, cooling. And yes, we've spent 20 years cooling chips from the outside: chilled air, water, cold plates, metal spreaders bonded to the back. All of it works on heat that has already escaped the die.

But think about it, that doesn't actually address the root of the problem. The heat is made at the transistor, buried under hundreds of microns of silicon and a stack of metal interconnect, and by the time it reaches anything a cooler can touch it has already throttled your chip or killed it.

What's the best solution? Well, yes, put a better heat conductor inside the die, right next to the transistors. And what is the best heat conductor? Diamond. 10 times better than silicon. And yet it is nowhere near a production chip, yet. The catch is temperature. Growing diamond normally needs temperatures of 800°C, and while the transistors can take that, the metal interconnect layered on top can't survive much past 400.

So this has limited diamond's usefulness. You use it either as the starting substrate before you build anything, which means you need to change the entire process and tooling. Or you bond it onto the thinned-down back of the chip, miles from the hotspot, with a thermal resistance sitting at exactly the interface where the heat is trying to cross.

Or you deposit diamond at lower temperatures. But that's impossible right?

The 400°C trick

PROUD deposits diamond at 400°C. That's not quite the whole thing, but it's certainly a huge thing. At 400 you can grow it after the transistors are built, during back-end-of-line, sitting right on top of the heat. And it grows straight onto silicon, SiC, GaN and glass with no buffer layer, which matters more than it sounds.

Doing it with no buffer layer is the hard part, the bit you don't expect to work if you know CVD. Instead of a buffer they engineer the interface itself. That doesn't take the resistance at the contact to zero, physics won't allow that, but it gets it extremely low, nothing like the barrier a buffer throws up, so the diamond keeps almost all of its advantage right where the heat is. The films are thin, 500nm to 10 microns against 300-plus for backside bonding, a 500nm film still shifts heat at 600 W/m·K, and a deposition takes about an hour.

Does it actually work?

There are three places diamond can go in: as an interlayer once the transistors are formed, on top of the finished interconnect to pull heat sideways between stacked dies, or on the thinned backside, swapping copper or aluminium spreaders for something 10 times better. Even a 25°C drop in junction temperature is a big deal for things like automotive radar and power electronics, real lifetime gains on parts that run hot. PROUD has already shown 50°C off GaN HEMTs and 70°C off PIN diodes, and grown diamond on live GaN-on-Si dice with the devices still working after polishing. That last one is the hardest of the three, and it's the one they've actually done.

On the team, because for a process this specialised the founders are the moat. Ligia Colina, the CEO, is a plasma physicist who spent a decade designing CVD reactors, which is the exact discipline this whole thing lives or dies on. Mehdi Naamoun, the CTO, invented the core process at EPFL, after 15 years doing nothing but CVD diamond for power electronics. So you have the person who can build and tune the reactor and the person who invented what it makes, both of them deep in this one niche rather than generalists who wandered in. And they bootstrapped from 2022, funding their own lab by selling scratch-resistant diamond coatings to Swiss watchmakers. That reads like a quirky origin story, but it's the part I'd underline. Swiss watchmakers are about as exacting a customer as you'll find, so shipping them diamond coatings at quality, again and again, means PROUD can already make this stuff to a standard and repeatably. That is a different game from a lab demo, and at pre-seed it is rare. Most deep-tech teams this early have a nice result in a paper and nothing you could actually put in a product.

So who else is doing this?

Most of the money and the noise is still outside the die: direct-to-chip liquid, immersion tanks, fancier cold plates. Useful, all of it, but it's working on heat that has already left the transistor, so it runs into the same ceiling we started with. I don't really file that under competition. If anything, it's the thing PROUD makes less necessary over time.

The competition that actually interests me is the small group also trying to cool at the source. The main one is microfluidics, etching tiny channels into or under the die and pushing coolant through them, which Corintis out of EPFL is doing well. It's a real approach and it works. But it's plumbing inside your chip: channels, fluid, pumps, and all the reliability and leak questions that come with moving liquid through silicon. PROUD is the other kind of answer, a solid passive layer with no moving parts that just spreads the heat better. If you're a foundry, "deposit one more layer in your existing flow" is a far easier yes than "let us plumb your wafer".

Then there is the diamond crowd, which sounds like head-on competition and mostly isn't. The GaN-on-diamond and diamond-substrate names, Akash and the diamond power-device companies, grow or bond devices onto diamond as a starting material, at high temperature, before anything else is built. That's a substrate play: you start from diamond and build on it. It does nothing for the billions of chips already designed around silicon. PROUD's whole edge is going in late, at 400°C, after the transistors are there. Same material, completely different point in the process.

The risk worth naming is that diamond isn't a secret, and a big materials house (Element Six is the obvious one) or a foundry could decide to build the low-temperature, no-buffer process in-house. Could happen. But that process is the hard, non-obvious part Mehdi spent 15 years on and patented, and "we'll just build it ourselves" has been the famous last words of plenty of incumbents who underestimated how nasty the deposition chemistry really is.

How big this gets

The thermal wall is now the thing gating the entire AI-compute roadmap. Nvidia's GB200 ships liquid-cooled out of the box, data centres are running into power limits, and direct-to-chip liquid cooling is on track to become the default in AI training builds by 2027. Once you've maxed cooling from the outside, and we basically have, the only direction left is inside, at the transistor. There isn't another bit of physics hiding behind that.

So whoever owns how you cool from the inside owns a step in every advanced wafer, and owning a step in the wafer is a completely different business from selling a cooling part. It's the materials-and-equipment layer: high margin, and once you are qualified into a node you are stuck in there for years. The external cooling stack this sits above is already around an $8bn market for AI accelerators (Yole's numbers), on its way past $25bn by 2030. The in-die layer doesn't even have its own line on the chart yet, because until you could deposit at 400°C nobody could build there.

They don't have to win all of that on day one, and the plan doesn't ask them to. Land in compound semis first, GaN and SiC power and RF at 6 and 8-inch, where the heat pain is worst and qualification is quicker, then climb to 12-inch silicon logic and AI where the numbers are 100 times bigger.

What has to go right

None of this is free. The one thing that has to work is qualification. Getting a brand-new material into a foundry flow, at the reliability bar these layers are held to, is the gate every materials company eventually runs at, and plenty of them die there. It's also why this is the sort of deal we like. We are a dedicated semiconductor fund with foundry LPs. We know the foundry world and how qualification actually works, and helping a pre-seed company get onto that path is exactly why we exist and will win.

The bet is classic first-principles stuff:

If you want to know more, ping us.