Astrus, a Canadian AI startup building the world’s first physics-aware foundation model for chip design, has closed an $8 million USD raise led by Khosla Ventures with participation from Juniper Networks founder Pradeep Sindhu, 1517 Fund, Drive Capital, Alumni Ventures, and others.
While much of chip design is automated, the most critical parts of advanced chips are still hand-crafted by engineers. Astrus is changing that.
“Analog layout is one of the biggest bottlenecks in the design process for the most advanced microchips,” said Brad Moon, co-founder and CEO of Astrus, in a statement. “Our AI learns the underlying physics of chip design, allowing us to automate what has always been a manual process. This funding enables us to accelerate chip development for the world’s leading teams and drive faster progress in compute.”
Why this matters
Developing the analog portion of chip design — the circuits that rely on an understanding of real-world physics — is a painstaking, manual process. Engineers place transistors one at a time, guided by experience and intuition. For advanced analog blocks, designs can take months,, with costs reaching into the hundreds of millions.
Large analog design projects that once took months can now be completed in weeks. We achieve this by automating the core building blocks, individual analog layouts that used to take hours are now generated in seconds.
Rooted in world-class AI research
Astrus was founded in Toronto and Waterloo by Brad Moon, a former satellite-sensor chip designer, and Zeyi Wang, a reinforcement learning researcher trained by AlphaGo advisor Martin Müller. Their team also includes Kenny Young, who completed his PhD under reinforcement learning pioneer Rich Sutton.
Astrus’s architecture adapts the core ideas behind AlphaGo and applies them to chip design. Its AI learns from scratch, generating trillions of designs in simulation while developing the intuition behind what makes circuits exceptional.
Just as AlphaGo surpassed the world’s best players, Astrus is on track to outperform the world’s best layout engineers.
“We’re scaling one of the most under-utilized approaches in modern AI: training agents from scratch, in simulation and with a clear objective: to surpass human intuition,” said Kenny Young, Astrus’s Founding Research Scientist.
Looking ahead
Astrus is partnering with the world’s leading semiconductor foundries and top chip design companies to bring this breakthrough into production. Together, we are building the next generation of AI-designed chips, accelerating progress in compute and working to radically improve global computation.