By Ken Butler, Business Development Senior Director, Advantest Cloud Solutions (ACS)
Most semiconductor market sectors saw a softening of business in 2023. However, 2024 should herald a rebound in demand. One driver will be the U.S. CHIPS and Science Act, which has already awarded its first contract as of this writing, and the similar efforts in other countries, each intended to minimize reliance on foreign sources within their respective semiconductor supply chains. Another part of that increase will be attributed to recent significant growth in the use of artificial intelligence (AI). In day-to-day life, there are many examples, such as AI-powered text and image generation, large language model (LLM)-enabled search engines from all the major providers, and AI productivity enhancers for many endeavors in STEM fields and others.
AI in Semiconductor Test
Semiconductor test is no exception to the infusion of AI technology into workflows and will also benefit from these new advances. In late 2023, Advantest announced the availability of the ACS Real-Time Data Infrastructure (RTDI), fueled by customer demand for adding complex AI/ML analytics to their test solutions. For all modern semiconductor products, but particularly for highly integrated “More than Moore” chiplet-based designs, product teams will continue to look for ways to incorporate advanced inferencing into the test process to optimize this valuable resource. Test optimization will be a frequent goal – every device should see “the right” test content – no more and no less than it needs. But today’s neural network-based techniques will also be used to detect devices that are quality or reliability risks, and should either be scrapped or have additional screening applied to ensure they will meet the demands of the end application and do so throughout the expected lifetime. Yield improvement will be another target, and AI will be applied to maximize product learning during the initial product ramp and into high-volume manufacturing. These inferences will increasingly be written in the languages favored by data scientists, such as Python, and made to run in concert with a test program written in a more traditional programming language.
