With new funding, Astrus prepares to automate one of the biggest bottlenecks in chip design

Kitchener-Waterloo- and Toronto-based Astrus has closed a new, $12-million USD ($16-million CAD) Series A round, co-founder and CEO Brad Moon told BetaKit in an interview. The round was led by San Francisco-based Caffeinated Capital, with support from Vinod Khosla’s Khosla Ventures, Garage Capital, MVP Ventures, RiSC (Canada), and RiSC Capital.

Astrus is developing an AI system that uses reinforcement learning (RL) to perform the physical design of integrated circuits. Its system is built for analog designers, meaning the people who take logic and intent requirements and turn them into physical designs for manufacturing.

“They’re essentially applied physicists, and then they have to manually create this data because there’s no automation,” Moon said. “How do you make this manufacturing reality of the most performance-critical components?”

“In the next year, we’ll have more physical design data than the rest of the world combined.”
Brad Moon, Astrus

The value of the global integrated circuit market is estimated to be over $800 billion USD this year, with hyperscalers, AI platforms, and consumer goods continuing to drive demand for new components. AI tools play a critical role in the design of these next-generation chips. The challenge, Moon said, is that the majority of AI companies developing tools for chip design only focus on the logic design part of the process, not the physical design process.

“If you watch an intent engineer, it looks like they’re writing code. They’re explaining logic. Large language models (LLMs) are really well-positioned to help there,” Moon said. “What we’re doing is more about manufacturing realities: the actual, physical silicon. It’s three-dimensional, and you have to run physics and simulation. That’s really where we’re focused, and that’s really where no one else is focused.”

Astrus’ choice of reinforcement learning over LLMs came down to one thing: data. Moon said there is a large amount of public logic design data available to train LLMs. That’s not the case for physical design.

“The physical design side is incredibly secretive. There is no public data. This is because companies like Apple or TSMC, there’s no way they’re going to let their physical design data out,” he added.

Moon said Astrus partnered with “the largest bleeding-edge foundries in the world” to understand their manufacturing requirements. The startup’s developers then applied reinforcement learning to create what Moon referred to as “infinite synthetic data.”

“We essentially have this AI learn. It plays around in this physics environment that understands the manufacturing rules, and it generates synthetic data. And you can just keep training this thing, and it’ll get better and better,” he said. “In the next year, we’ll have more physical design data than the rest of the world combined.”

Staying, and scaling, in Canada

Moon said Astrus is preparing to launch its solution in October and the new funding will support the launch and help scale its team. While many other Canadian companies in the semiconductor space have moved to the US or been acquired by US companies, Moon said Astrus has no plans to follow.

“We’ll probably have a go-to-market team in the US, but Canada is really well-positioned. We have Rich Sutton from University of Alberta, the father of reinforcement learning, and deep learning with Geoff Hinton from U of T. We have pipelines to two of the most important AI foundational technologies in the world,” he said.

Moon said potential customers include major semiconductor manufacturers, and Astrus is looking at potential collaborations with companies working on logic design tools.

“We can still partner with the LLM providers—they handle the intent side—but once they have a design they want to turn into manufacturing reality, you have to look at the physical design side,” he said.

With funding in place, including continued support from Vinod Khosla, Moon is confident that Astrus has the right solution for the moment.

“I think it’s going to take off like wildfire because this is one of the biggest bottlenecks in chip design process,” Moon said.

Feature image courtesy Alex Kinsella for BetaKit.