The US Department of Commerce has awarded $500 million to the artificial intelligence software firm SandboxAQ to screen new chemicals and materials for use in semiconductor manufacturing. Among the targets are alternatives to per- and poly-fluoroalkyl substances (PFAS), which are used in several chipmaking steps but are under intense environmental scrutiny.
SandboxAQ, a 2022 spin-off from Alphabet, simulates, analyzes, and selects chemicals and formulations from virtual libraries using large quantitative models, which the firm describes as “AI systems trained on the laws of physics, chemistry, and biology, not human language.” Substances that get through the screening will be passed to US laboratories for real-life testing and development.
The money comes from the Creating Helpful Incentives to Produce Semiconductors (CHIPS) and Science Act, which was signed into law by President Joe Biden in 2022 to develop a US semiconductor supply chain through subsidies and policy support. SandboxAQ will focus on four topics that the Commerce Department describes as urgent for onshoring electronics manufacturing: catalysts, rare earth–free magnets, battery systems, and PFAS-free process chemicals.
“President [Donald J.] Trump is committed to strengthening America’s semiconductor supply chain and ensuring national security,” US secretary of commerce Howard Lutnick says in a press release. “This award will accelerate the discovery and innovation of critical materials and reduce our reliance on foreign-controlled materials.” As part of the award structure, the US government will take a minority stake in SandboxAQ.
The effort is worth a shot but far from guaranteed, says Mark Thirsk, an electronic-chemical expert with Linx Consulting. “I’ve seen great pieces of software come out that people say are going to solve the material-choice problem. Usually they struggle,” he says. “There are so many bits of information that we don’t know. We have tens of thousands of compounds, and we know maybe 20% of each compound’s physical data. Filling in the rest, even with AI, becomes pretty difficult.”
Still, Thirsk is encouraged to see at least some of the CHIPS funds being deployed this way. “I think doing this and attempting these difficult problems is way better than throwing up your hands and saying we’re never going to succeed.”
The PFAS problem is especially vexing. Fluorine’s exceptional electronegativity makes fluorinated compounds either highly stable or highly reactive, a useful duality for etching nanometer-scale features onto silicon wafers. “You’re taking a very powerful tool out of the toolbox when you remove fluorine,” Thirsk says. But regulators are cracking down on PFAS emissions, creating a need for PFAS-free materials for chip fabrication facilities.
Work is already progressing on PFAS-free surfactants used to create uniform photoresist layers, Thirsk says, but non-PFAS photo-acid generators and etching gases have proven harder to identify and qualify. Finding materials that can be manufactured in the US and dropped in at existing plants will be additional challenges.
“Is AI the best way to do it? I don’t know,” Thirsk says. “I know that lots of clever chemical simulation programs get thrown at these problems. Whether you call them AI or just machine learning or chemical databases, it’s still an immature science. And I’m not sure AI is going to be the magic key that unlocks the kingdom.”
