The Chip Insider®– AI in the Semiconductor Industry

AI in the Semiconductor Industry: AI is much more than a semiconductor demand driver…As for the semiconductor industry, it was using AI for decades before it became a substantive market for semiconductors. The earliest AI applications were in lithography, process diagnostics, yield optimization, early detection of tool failures, and scheduling. In the late 70s and early 80s there were known efforts using…the success of Intel and TSMC was due to… They also pointed to the use of digital twins and…used for lithography today…

By the late 1980s, the first applications of today, which would be called 'Agentic AI', began to show up in the fab. They were 'agentic' in the current definition of making decisions and acting autonomously with minimal human intervention. In this timeframe, insiders learned Intel was using…AMD was the first to apply Agentic AI across a full fab. This was in Austin…led by Gary Heerssen…The head of manufacturing did not allow computers into the other fab. It was a perfect A:B test…Agentic AI controls for the fab using feedback loops between tools to optimize yields down-flow in the 1990s and early 2000s…Tom Sonderman (who is now CEO of SkyWater)…all of AMD's Silicon Valley production fabs were soon shut down because Austin alone was outproducing their combined total…more successful AI efforts continued to come in the 2010s and 20s. Examples include: digital twins that identify lithographic hot spots in a design, automatic defect cataloging, getting more information out of noise from metrology tool images, and listening for audio signatures to identify and predict tool failures…Symbolic AI solutions…Transformer AI models are relatively new…The power of being able to assign 'attention' in an AI model quickly made its way into the design and ATE communities…Synopsys and Advantest were the first known to aggressively push Transformers…Sassine Ghazi…for finding power-performance optimums…The key to success…was an ability to meld separate groups with problem and solution knowledge to develop a complete product.

Transformers are also quietly making their way into the fab…It's hard to find anyone replacing engineers with AI…Looking back, the most successful AI developments have come from the bottom up, with engineers…Another common AI failure mode is when…Or as Denzel Washington put it, "Don't confuse movement with progress." The bottom line is, understand the problem to be solved with AI, its value to the company, and the inclusion of problem holders with solution providers.

“No computer is ever going to ask a new, reasonable question. It takes trained people to do that. – Grace Hopper