Astrus Raises $8M to Expedite AI-Driven Microchip Design Process - InforCapital

Toronto and Waterloo-based AI startup, Astrus, which is creating the world's pioneer physics-aware foundation model for chip design, has secured $8 million USD in funding. Khosla Ventures led the round with contributions from other strategic investors, including Pradeep Sindhu, founder of Juniper Networks, 1517 Fund, Drive Capital, and Alumni Ventures.

The company aims to expedite chip development for the globe’s premier semiconductor design firms, pushing the boundaries of computing progress. The design of digital chips has largely been automated, but the analog component of advanced semiconductors remains a manual process, involving engineers placing transistors one by one. This method can take months and cost hundreds of millions for cutting-edge designs.

Astrus is revolutionizing this process with its AI. The company's reinforcement learning architecture understands the fundamental physics of chip design, producing thousands of high-quality layouts within seconds. This technology transforms a process that used to take months of manual engineering into an operation that can be completed in hours. It hastens the development of advanced chips and promotes quicker advancements in compute.

"Analog layout is one of the largest bottlenecks in the design process for the most sophisticated microchips," commented Brad Moon, Co-founder and CEO of Astrus. "Our AI comprehends the physics behind chip design, which lets us automate what used to be a manual process."

Brad Moon, a former satellite-sensor chip designer, and Zeyi Wang, a reinforcement learning researcher mentored by AlphaGo advisor Martin Müller, founded Astrus in Toronto and Waterloo. The team also includes Kenny Young, a PhD graduate who studied under reinforcement learning pioneer Rich Sutton. By adapting the fundamental ideas behind AlphaGo and applying them to analog layout, Astrus is creating a foundation model that can surpass human designers and discover completely new circuit architectures.

"We're working towards one of the biggest reinforcement learning training runs ever attempted, engineered to propel chip design beyond human limits," said Kenny Young, Astrus’s Founding Research Scientist.

The newly acquired funds will allow Astrus to grow its research and engineering team, scale its computing infrastructure for large-scale reinforcement learning training, and provide tools to the leading semiconductor companies working on the most advanced designs.