Astrus raises $12M to automate chip design's physical bottleneck

What's the deal? Astrus, a Kitchener-Waterloo- and Toronto-based startup, has closed a $12-million (US$11.7-million) Series A round, co-founder and chief executive officer Brad Moon told BetaKit. San Francisco-based Caffeinated Capital led the round, with support from Khosla Ventures, Garage CapitalDealroom has a profile for this one. Try Dealroom →, MVP VenturesDealroom has a profile for this one. Try Dealroom →, RiSCDealroom has a profile for this one. Try Dealroom →, and RiSC Capital.

What's the endgame? Astrus is building an AI system that uses reinforcement learning to handle the physical design of integrated circuits — turning logic requirements into manufacturable silicon. The company says most AI chip-design tools focus on logic design, not the physical process it targets.

Why reinforcement learning? It came down to data. Public logic-design data is plentiful, but physical-design data is scarce and closely guarded. "The physical design side is incredibly secretive. There is no public data," Moon said, citing firms like Apple and TSMC.

To solve that, Astrus partnered with what Moon called "the largest bleeding-edge foundries in the world" to learn their manufacturing rules, then used reinforcement learning to generate synthetic training data. "In the next year, we'll have more physical design data than the rest of the world combined," Moon said.

Why now? The global integrated circuit market is estimated at over $800 billion this year, with hyperscalers, AI platforms, and consumer goods driving demand for new components. Astrus plans to launch its product in October, and the funding will support that launch and scale its team.

Staying in Canada: While many Canadian semiconductor companies have moved to the US or been acquired, Moon said Astrus has no plans to follow, pointing to local expertise in reinforcement learning and deep learning. "We'll probably have a go-to-market team in the US, but Canada is really well-positioned," he said.

The signal: At $12 million, the round sits in the top third of comparable raises — a bet that automating physical chip design, an area rivals have largely ignored, can carve out a defensible niche in a market otherwise crowded by logic-focused tools.

Read more: betakit.com

Image credit: le_hollandais_volant