For economic and national security reasons, the semiconductor industry is rebuilding itself on U.S. soil after more than three decades of concerted offshoring. The scale of this reshoring is immense—the $52.7 billion infusion from the federal CHIPS and Science Act of 2022 being only the beginning.
The challenges of reshoring are equally immense. Building a multibillion-dollar fab takes perhaps 6,000 construction workers three years, and more than 60 high-profile projects have been announced. SIA forecasts that these fabs and other facilities will employ about 115,000 people—with the caveat that, without a serious effort to attract engineers and technologists in a big way, 67,000 of them will go unfilled.
Then there’s the pressing question of the competitiveness and profitability of the semiconductors these U.S. fabs will produce. These facilities will cost about 30% more to build and operate than ones in Taiwan, South Korea or Singapore, and up to 50% more than one in China.
The post-pandemic years haven't been especially kind to the high-tech industry, but there are signs that things are looking up. Find out the top priorities for tech companies as they look to grow business.
The issue with reshoring won’t be a lack of data. The 400-1,200 steps of a semiconductor manufacturing process generate 10-100× the data volume of any other manufactured product. The challenge is how to gather and rationalize it for AI and other systems that monitor and optimize production.
A typical fab’s equipment and manufacturing execution systems (MES) crank out terabytes of data in a litany of formats each day. Typically, less than 10% of it feeds into analytics systems. But the other 90% surely holds clues to spotting nascent problems which, if addressed with a design change, a new mask set, or other adjustments, can improve yields.
There are tools that can help with this. They harmonize dozens of data types—from fault detection and classification (FDC) on through test, assembly and packaging—into cloud-based semantic models ready for interactive analytics or machine-learning applications.
Similarly, data related to the status of the tools of production themselves can feed into AI predictive maintenance systems to reduce costly downtime.
Also key to the competitiveness of a reshored U.S. semiconductor industry will be the far deeper integration of production and business systems—connecting the “shop floor” to the “top floor,” as it were.
There are other possibilities here, but you get the idea. The union of shop-floor and top-floor data comes with added benefits of improved product traceability and precise sustainability management and reporting. Better traceability helps quickly nail down where in the fab or the supply chain defects crop up and the nature, cause and extent of the issue. That improves communication with customers and faster problem resolution.
With respect to sustainability, semiconductor firms must prepare to provide verifiable, product-specific emissions numbers that will customers will need as carbon accounting rules, including for scope 3 value-chain emissions, take effect. Having that capability will help fabs manage reporting related to more stringent U.S. environmental regulations.
Manufacturers will kick digital transformation into high gear in 2025 as they adopt advanced technologies like AI to build resiliency and drive growth.
The U.S. semiconductor industry’s initial success in reshoring will depend on massive buildouts, enormous hiring sprees, innovative training, and workforce-development efforts. Its ultimate success, though, will hinge on its ability to maximize yields and ensure profitability.
As we start anew, we should cast the widest possible net for our data and maximize the gains we distill from it.
Editor’s Note: This article first appeared in EE Times and is republished here with permission.
