A Study of Semiconductor Defects within Automotive Manufacturing using Predictive Analytics

Latent defects within semiconductors have been a primary cause of final inspection failures within the electronic control unit (ECU) manufacturing process. Finding the true root cause within the ECU circuit is a challenge due to the complexity of these devices. As automotive OEMs increase their Tier I testing requirements to eliminate failures within their manufacturing and in the consumer market, Tier I suppliers are reacting by introducing high voltage burn-in testing. The burn-in testing can induce electrical overstress (EOS) to the semiconductors, resulting in requests to Tier II semiconductor manufacturers to analyze these failures. However, the end result is a failure caused within the Tier I process, unrelated to the semiconductor manufacturing process. The analysis of these suspect manufacturers costs both the Tier I and Tier II time and money that could have been used more effectively. Historical data exists within the Tier I manufacturing process that can be analyzed to identify potential relationships between production variables, such as specific equipment used for testing, the temperature of the specific test, the type of measurement taken during that test step, and the result of a non-supplier related defect (NTF or EOS result). A prediction model is created to flag these as potential non-supplier related defects based on previous investigations. The results shift the focus away from the Tier II semiconductor supplier for non-related defects and towards solving the root cause which may be occurring within the testing or manufacturing process.