Three companies tend to get top billing when pundits talk about the physical underpinning of the artificial intelligence boom.
Nvidia, the US computer chip designer whose graphics processing units fill the world’s data centers, is one of them. Another is Taiwan Semiconductor Manufacturing Company (TSMC), a contract manufacturer that builds most of Nvidia’s chips. Also in the triumvirate is ASML, a Dutch company that manufactures complex photolithography machines for projecting intricate patterns on silicon wafers to create chip circuit lines.
A recent New York Times opinion piece highlighted a less-acknowledged cadre of companies, many European, that also contribute to the AI boom by providing the “transformers, switching gear and energy management systems without which no data center operates.”
All true. But what the Times and many other observers don’t recognize are the chemical companies that are also critical to the AI revolution. Love data centers or hate them, they would be impossible without these firms. And as AI proliferates, technology from chemistry-centric companies and start-ups will be needed to make AI faster, cheaper, and easier on the environment.
A computer chip is essentially a highly engineered amalgam of ultrapure chemicals. It starts with the silicon wafer, which is made of polysilicon mostly from Chinese chemical companies like GCL Tech and Xinte Energy but also from the German firm Wacker Chemie and US-based Hemlock Semiconductor, majority owned by Corning.
ASML’s lithography machines create circuit lines by projecting light through a patterned mask and onto a thin layer of photoresist that has been applied to the silicon wafer. As lithography advances through shorter wavelengths of light, photoresist makers like JSR and Shin-Etsu Chemical, both of Japan, must develop new lithographic polymers to keep up.
High-end chips from Nvidia hold multiple metal layers so thin they can’t be laid down like a sheet of foil. Rather, chemical companies like France’s Air Liquide create organometallic precursors that are applied one molecule at a time by atomic layer deposition. As new metals—molybdenum, for example—are introduced, chemical makers must scramble to invent new precursors.
Applying layer after layer without regular smoothing would create chips with unacceptable hills and valleys. Thus, TSMC depends on chemical mechanical planarization, a process in which chip-covered silicon wafers are polished with polymer pads made by companies such as Qnity Electronics and Entegris, both of the US. The pads apply slurries of colloidal silica and other chemicals that gently smooth out any imperfections on the surface of the wafer.
And before a next layer can be added, the smoothed chip must be cleaned with ultra-high-purity versions of solvents such as hydrogen peroxide, isopropyl alcohol, and sulfuric acid. These chemicals are the province of multiple firms—including Entegris, Japan’s Mitsubishi Chemical, and Taiwan’s Sunlit Fluo & Chemical—that must continually raise purity levels as circuit lines shrink.
The chemical industry’s importance to AI doesn’t stop at the computer chip. For example, the data centers that house Nvidia’s chips are air cooled with the help of fluorochemical refrigerants. Increasingly, the chips themselves are being cooled by running chemical fluids through plates pressed on their surfaces or by immersing them straight into coolants.
AI is hardly an unalloyed good, and the chemicals that fuel it are often not benign. The US Department of Commerce recently awarded the AI software firm SandboxAQ $500 million to help develop new materials for semiconductor manufacturing, including alternatives to per- and poly-fluoroalkyl substances.
And smaller, more powerful chips are more energy-efficient chips. Inpria, one of C&EN’s 10 Start-Ups to Watch a decade ago, develops metal oxide photoresists specifically for the extreme ultraviolet light lithography used in ASML’s top-end machines. Acquired by JSR in 2021 for more than $500 million, Inpria is now seeing its technology edge toward use in the most-advanced chips.
Further chip advances will happen only with more chemical innovation. As in the case of Inpria, success can take a decade or more, but the returns can be good. Examples like this should be inspiration for investors and for chemists.
This editorial is the result of collective deliberation in C&EN. For this editorial, the lead contributor is Michael McCoy.
Views expressed are not necessarily those of ACS.
