artificial intelligence for semiconductor industry

The topics of environmental sustainability and artificial intelligence have always been fascinating to me. Their wide-reaching impact on society and potential significance in shaping our future opportunities make them particularly meaningful. Undoubtedly, both topics offer great potential but also come with a multitude of risks - from the pressing issue of climate change to the numerous ways in which AI could potentially go awry.

Even though I didn't study AI in graduate school and am not an expert in the field, I consider myself fortunate to be working on research and product development projects related to it in my company. The semiconductor manufacturing process is already very complex and it's impossible for any one person to understand every integration step, even for older-generation chips. The process development for each step is becoming more complex due to the atomic-level challenges and the need for increased tool precision. That's why, the opportunities for AI in semiconductor industry feel almost boundless.

According to McKinsey, semiconductor companies could potentially earn nearly $100 billion in revenue within the next five years through the use of AI. Although the exact figure may be subject to discussion, the rationale behind the projection is evident. As the technological complexity and capital expenditure for each subsequent logic and memory node continue to rise at an almost exponential rate, the industry requires innovative approaches for analyzing data and gaining valuable insights.

Nevertheless, the implementation of AI and ML in industries will differ greatly from the more typical uses by tech giants in areas such as customer-facing products, marketing, and content generation. Although AI technology continues to evolve, the expertise and knowledge of human professionals will still be essential in fields that require advanced technology. During my time in the engineering rotation program, I had the privilege of working with a team of researchers at Lam Researchers who emphasized the significance of a human-AI hybrid approach.

I was in charge of creating a controlled virtual process game to systematically benchmark the performance of humans and computers for the design of a semiconductor fabrication process. We found that human engineers excel in the early stages of development, whereas the algorithms are far more cost-efficient near the tight tolerances of the target. We show that a strategy using both human designers with high expertise and algorithms in a human first–computer last strategy can reduce the cost-to-target by half compared with only human designers. More details can be found in our publication.

I continue engaging in this space through my role in the Equipment Intelligence® strategic initiative. The future will only bring more and more AI solutions that we need to ensure meet our customers' objectives. I also explore the environmental impact of ML dataset training and processing among other virtualization and digital twin techniques. While these calculations can be very relatively intensive, they still use significantly less resources than standard laboratory and fab work, as we demonstrate in a recent article.