Astrus is bringing chip design into the future
Written By Alex Kinsella
We’re surrounded by integrated circuits, from the CPUs in our laptops and smartphones to the GPUs powering AI data centres to the simple chips that start our coffee makers before we wake up (bless). The pace of improvement in chip design and manufacturing means that your smartwatch has more processing power than the computer that powered the lunar missions back in the 1960s and 70s.
Despite the improvement in computing, much of the work that brings these chips to life is still a manual process that can take years to complete. There’s two sides to designing the components on a chip—the intent and the physical. The intent is pretty straight forward. Designers create a component to do a specific task, like manage communication between other components or manage power.
Despite the improvement in computing, much of the work that brings these chips to life is still a manual process that can take years to complete. There’s two sides to designing the components on a chip—the intent and the physical. The intent is pretty straight forward. Designers create a component to do a specific task, like manage communication between other components or manage power.
Those designs are then given to the layout engineers—and that’s where the bottleneck happens today. Intent is one thing, making it work with hundreds and thousands of other components on a chip is another. Each iteration of layout has to be done by hand, and there can be hundreds of iterations needed before the chip is ready for manufacturing.
Reducing the time to iterate through layouts is a challenge many chip manufacturers have wanted to address, but it has remained unchanged due to the complexity of chip design and manufacturing.
This is where the Astrus comes into the story.
Founded by Brad Moon and Zeyi Wang, Astrus is using AI to automate the layout process for chip components in virtually every type of integrated circuit. Founded in 2023, the startup has already raised over $25 million, including pre-seed and seed rounds led by Khosla Ventures and a recent $12 million Series A led by Caffeinated Capital.
The best part? They’re doing it in Canada, with offices in Toronto and here at their headquarters at Catalyst Commons in Kitchener.
Reducing the time to iterate through layouts is a challenge many chip manufacturers have wanted to address, but it has remained unchanged due to the complexity of chip design and manufacturing.
This is where the Astrus comes into the story.
Founded by Brad Moon and Zeyi Wang, Astrus is using AI to automate the layout process for chip components in virtually every type of integrated circuit. Founded in 2023, the startup has already raised over $25 million, including pre-seed and seed rounds led by Khosla Ventures and a recent $12 million Series A led by Caffeinated Capital.
The best part? They’re doing it in Canada, with offices in Toronto and here at their headquarters at Catalyst Commons in Kitchener.
A desire to solve tough problems with deep tech.
Moon and Wang are both University of Alberta alums, but they didn’t meet until they were both participants in the Entrepreneurs First program in Toronto. Before meeting, Moon’s post-university career involved stints at several accelerators, including the University of Waterloo’s Velocity program, and a role in the early days of edtech startup ApplyBoard. At the University of Alberta, Wang studied reinforcement learning under professor Martin Müller whose graduates worked on the research team for Google DeepMind’s AlphaGo program—the first computer system to beat a world-champion human Go player.
“They sat down and talked about their two areas of specialty—electrical engineering for Brad and machine learning engineering for Zeyi,” said Founding Head of Talent and Investor Steph Hector. “They started to talk about whether reinforcement learning could be used to solve chip design, which formed the startup idea, and then I started working with them to help build out the team.”
It was a running start that hasn’t slowed down. The startup has assembled a team of experts to bring their vision to life, including leaders in software and game engineering, reinforcement learning, and AI and machine learning.
“They sat down and talked about their two areas of specialty—electrical engineering for Brad and machine learning engineering for Zeyi,” said Founding Head of Talent and Investor Steph Hector. “They started to talk about whether reinforcement learning could be used to solve chip design, which formed the startup idea, and then I started working with them to help build out the team.”
It was a running start that hasn’t slowed down. The startup has assembled a team of experts to bring their vision to life, including leaders in software and game engineering, reinforcement learning, and AI and machine learning.
Analog design is part art, part science.
The potential for cost savings is huge, but Hector said the benefits for manufacturers go beyond time savings.
“If you can optimize layout by even 1%, you could save millions of dollars. Today, the entire process of layout design is 100% manual by humans,” she said. “Depending on the complexity of the circuit, the layout process takes three to five days between each iteration, with each pass finding better ways to optimize and design a new layout,” Hector said
Astrus’ approach aims to reduce those three to five day iterations down to minutes. The team is building a platform that allows an analog designer to enter the requirements and have potentially hundreds of layout options generated by its custom designed and trained AI.
Instead of training on human data, Astrus trains its AI by having it play a game that mimics the manufacturing environment of a chip foundry. The training process teaches the AI the art of analog layout in a similar way Google’s DeepMind team taught its AI to play Go.
“Our AI can quickly produce a number of layouts and the designer can see the options, pick the one that closest meets the requirements and then iterate off of that one,” she said.
“If you can optimize layout by even 1%, you could save millions of dollars. Today, the entire process of layout design is 100% manual by humans,” she said. “Depending on the complexity of the circuit, the layout process takes three to five days between each iteration, with each pass finding better ways to optimize and design a new layout,” Hector said
Astrus’ approach aims to reduce those three to five day iterations down to minutes. The team is building a platform that allows an analog designer to enter the requirements and have potentially hundreds of layout options generated by its custom designed and trained AI.
Instead of training on human data, Astrus trains its AI by having it play a game that mimics the manufacturing environment of a chip foundry. The training process teaches the AI the art of analog layout in a similar way Google’s DeepMind team taught its AI to play Go.
“Our AI can quickly produce a number of layouts and the designer can see the options, pick the one that closest meets the requirements and then iterate off of that one,” she said.
Taking an uncommon approach to collaboration.
Astrus has quickly gained the attention of chip design companies, chip manufacturers and VC firms. Khosla Ventures is one of Astrus largest investors, including its founder Vinod Khosla.
“Khosla was the co-founder of Sun Microsystems, he knows the semiconductor space. He is personally working with our company. It’s not just partners within his VC — we’re working directly with him,” Hector said.
Today, the Astrus team is spread between its office in Toronto and its space in the Glasgow Hall area at Catalyst Commons.
Hector was the sixth hire for Astrus, but has been part of the team since Moon and Wang started thinking about the type of team they needed to assemble.
“I’ve helped bring on each member of our team. They are incredibly smart and exceptionally talented,” Steph said. “They’re a strong mix of principal-level engineers who value clear communication and are deeply focused on solving complex problems. We operate with a high degree of respect across disciplines, give people real ownership and autonomy with the support to execute, to solve challenging and unique technical problems.”
Astrus’ workplace is built around in-person collaboration, but they have built in systems to make it easy to connect when people work from home or are in different offices.
“We’re always connected over Zoom between the two offices, so you just walk over and talk to whoever you need to. If you’re working from home, you join the Zoom so you’re not missing out on the random conversations that happen,” she said. “You’re always connected to the team, to help keep the speed of a startup.”
Choosing Catalyst Commons was an easy choice for Hector and the Astrus team.
“We looked at it as part of our recruitment strategy. Catalyst removes the frustrations of going to an office. There’s parking, it’s on the Iron Horse trail, it’s near an ION station,” Hector said. “It’s the tech space to be in.”
“Khosla was the co-founder of Sun Microsystems, he knows the semiconductor space. He is personally working with our company. It’s not just partners within his VC — we’re working directly with him,” Hector said.
Today, the Astrus team is spread between its office in Toronto and its space in the Glasgow Hall area at Catalyst Commons.
Hector was the sixth hire for Astrus, but has been part of the team since Moon and Wang started thinking about the type of team they needed to assemble.
“I’ve helped bring on each member of our team. They are incredibly smart and exceptionally talented,” Steph said. “They’re a strong mix of principal-level engineers who value clear communication and are deeply focused on solving complex problems. We operate with a high degree of respect across disciplines, give people real ownership and autonomy with the support to execute, to solve challenging and unique technical problems.”
Astrus’ workplace is built around in-person collaboration, but they have built in systems to make it easy to connect when people work from home or are in different offices.
“We’re always connected over Zoom between the two offices, so you just walk over and talk to whoever you need to. If you’re working from home, you join the Zoom so you’re not missing out on the random conversations that happen,” she said. “You’re always connected to the team, to help keep the speed of a startup.”
Choosing Catalyst Commons was an easy choice for Hector and the Astrus team.
“We looked at it as part of our recruitment strategy. Catalyst removes the frustrations of going to an office. There’s parking, it’s on the Iron Horse trail, it’s near an ION station,” Hector said. “It’s the tech space to be in.”
