Every advanced AI program ― from the widely available ChatGPT to those run by the Defense Department ― depends on access to semiconductors. These computer chips are in almost every digital device today, and newer semiconductor technology is essential for the enormous computing power that AI requires.

While the U.S. first developed semiconductors and led globally in semiconductor development and manufacturing in the last century, today it produces just 12% of all semiconductors. Following passage of the CHIPS and Science Act of 2022, the federal government has embarked on a sweeping effort to promote and protect U.S. leadership in the semiconductor industry to ensure economic and national security. A new National Academies report recommends actions for the U.S. Department of Defense ― coordinating with the commercial sector, universities, and other parts of government ― to secure its access to this critical technology.

We spoke with the chair of the committee that wrote the report, Liesl Folks, director of the Center for Semiconductor Manufacturing and professor of electrical and computer engineering at the University of Arizona, about how access to semiconductors is shaping AI and what the report’s recommendations could mean for its future.

Why is access to semiconductors so critical for advancing AI?

Folks: All AI models are trained on highly complex semiconductor chips that are custom designed to perform particular mathematical operations quickly on very large amounts of data. As amazing as AI seems today, it is expected to get much better, quickly, as better semiconductor chips are produced to meet this need. And, AI technology holds enormous promise for defense applications, both to assist in the processing of data in data centers and in so-called “edge-computing” situations such as for accelerated vision processing in augmented reality headsets used by pilots.

So, in order to remain competitive with other nations, the DOD needs access to the latest AI chips.  

The challenge that our new report highlights is that today these chips are likely to be designed in the U.S., but manufactured in Taiwan, and DOD’s current procurement policies and practices create barriers to use of chips manufactured in non-U.S.-based sites. Effectively, the DOD will be unable to keep up with innovations happening around the world unless it can gain access to the most advanced semiconductors being manufactured today.

The committee called for “reshoring” of semiconductor production capabilities. Why is this important? How do you think this shift might impact private companies in the U.S. working at the cutting edge of AI development?

Folks: As the CHIPS and Science Act of 2022 highlights, the growing dependence of much of our economy on semiconductors creates potential vulnerabilities, since a disruption to the supply creates negative ripple effects — such as we saw during the recent pandemic.

One obvious way to mitigate these risks is to reshore some production facilities, so that the U.S. will have a more reliable supply of chips to support its economy as well as national security. The chips needed for AI are already an intensely valuable resource, and of growing importance. Manufacturing AI chips in North America would increase the resilience of the supply. Further, it is plausible that we could see a ramp up of innovation in AI chip design and production, as a result of having shorter supply chains and closer working relationships between chip design companies and chip manufacturers.

It seems the U.S. is increasingly using AI in a variety of important industries, and that our reliance on it will only go up. How do you think increased demand will impact access to semiconductors, and how do you suggest DOD prepare?

Folks: The highest-performing AI chips are already hard for companies to buy in the needed volumes, and the demand seems likely to keep growing for quite a while. As our report makes clear, it is vital that the DOD find new ways to work closely with the U.S. companies that are designing the best AI chips, so that it can anticipate its own uses for next generation chips and be ahead of the pack in placing orders for the latest technology. The goal should be that the U.S. DOD should be an early adopter of, and the first military user of, new technologies as they emerge. But this will require new and sustained partnerships to be developed between the commercial chip companies and the DOD.

How do you think expertise in the U.S. tech workforce needs to change or grow to support a more robust domestic semiconductor industry?

Folks: Much has been written about the challenges the U.S. is facing to produce sufficient STEM workers to meet its needs and to compete on the global stage. The effort to reshore semiconductor manufacturing will only intensify the demand for STEM workers. So, we need a whole-of-government effort to encourage more students to consider careers in the semiconductor sector early in their academic pathways, so that they can be intentional about acquiring the needed skills. Importantly, new semiconductor-sector jobs are expected to span the academic spectrum, including high school graduates as well as those with post-secondary degrees.

Your report calls for the DOD to prioritize investment in “leap-ahead” semiconductor technologies. Does this include AI, and can you tell me more about what leap-ahead technology could enable?

Folks: Leap-ahead technologies are those that allow companies to fundamentally change the way they make products, rather than just evolving forward from the current technology. Today, we see that it is becoming harder and harder to evolve the current lithography processes to make ever smaller transistors and cram them into closer proximity to each other. Meanwhile, there are a number of promising candidate technologies emerging that may allow us to “leap ahead” to get past these issues and continue on our path to improved microelectronics.

Our report urges the DOD to be diligent in supporting R&D for leap-ahead technologies, so that these inventions will have an IP home in the U.S. and so that the DOD will be able to be among the first adopters of these new technologies. Again, this recommendation is aimed at ensuring that the DOD retains technology superiority over its potential adversaries.

How do you think your report’s findings and recommendations could be used by other government agencies looking to expand their own access to semiconductors?

Folks: All the recommendations that aim to increase the resilience of the semiconductor supply in the U.S. will be broadly beneficial to U.S. government agencies.