Artificial intelligence (AI) and machine learning (ML) are becoming increasingly pervasive in everyone’s lives. The field of semiconductor test is no exception, and we see a boom in the number of companies wishing to deploy AI/ML into their chip development, validation, and production test environments. However, as we strive to do that, several challenges arise:
AI/ML methods typically require access to large and extensive data sets from across the semiconductor integrated workflow for training (and retraining)
These workloads tend to be computationally complex, and the existing compute infrastructure available in test is often not up to the task
When these intricate models are deployed into production to generate real-time inferences within a touchdown (i. e., while the parts are still being contacted), they require fast access to significant amounts of data which must be “crunched” quickly to minimize the impact on test time and cost
We are working with customers and partners to develop a comprehensive solution to address these problems called the ACS Real-Time Data Infrastructure (RTDI). ACS RTDI has multiple components which work in concert to provide secure deployment of a wide variety of analytics for real-time test data access and decision making in captive or disaggregated manufacturing situations. ACS RTDI begins with the concept of an open solution ecosystem, which has an architecture designed to allow development of in-house AI/ML applications or to procure them from a large array of 3rd party providers. These applications are housed on the ACS Container Hub cloud storage facility, and securely deployed into the test cell using the ACS Unified Server. Analytics run on the ACS Edge high performance and highly secure computation platform, and data communications from the test cell to the edge appliance, the floor server, and external databases is handled by ACS Nexus, which also facilitates communication back to the test cell for tester and test flow control.
In this presentation, we will discuss our experiences in designing, developing, and deploying ACS RTDI and describe current and expected future use cases.