What will be the impact of AI on semiconductor applications in the next 5-10 years?

One way to look at how AI can impact semiconductor development is to consider those activities within the development phase that are the highest effort, or are error prone (examples are verification, translating customer requirements to internal specifications). If AI can help improve the productivity of these activities, then it would be a great step forward. Another aspect to consider would be improving those activities where engineers spend the most time (E.g. documentation). As semiconductor development is a niche industry, and therefore faces significant talent shortages, should such activities be improved using AI, large benefits may be gained.

AI can be used to build smart user interfaces and simplify use of EDA and TCAD tools. This will will flatten the learning curve for new users. AI can assist in calibrating TCAD models and optimization of semiconductor devices. AI can also help to systematize/organize EDA and TCAD archives. For EDA and TCAD vendors, AI may provide assistance in customer (EDA and TCAD user) support. This will increase support scalability and reduce response and resolution time. All of the above will increase productivity.

With the strides AI is making, I would expect AI to help improve the Architectural exploration phase which is the key to optimal solution. The next stage Micro-Architecture is where AI can make lots of difference where it can aid engineers to explore various micro-architectures to result in balanced power-performance-area output. Providing the metrics for divergence of the real design from the Architecture/MRD would be something more that would help the overall product quality from features point of view.

With the advances of AI technology, the importance of human readability in computing is being reevaluated, which may have implications for the future of test cases in many fields including semiconductor. Computer readability has been a trade-off between machine readability and human readability. Machine readable code (or test cases) are already in practice; while AI can take it further to human readable using natural language processing. This would change the landscape in semiconductor test (and other functions) as well!

The points above discuss how AI algorithms might be used in the development and manufacturing of semiconductors, but another influence that the growth of AI will have on semiconductors is that many new semiconductor designs will incorporate enhancements in order for the resulting semiconductor chip to be able to more efficiently perform AI calculations. AI deep learning algorithms can be very compute intensive, needing to run on chips that can efficiently perform the necessary dot product multiplication and accumulation operations. There is a need to design semiconductor chips that can include engines to assist with running these AI algorithms.