edfas.org ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 25 NO. 2 16 ARTIFICIAL INTELLIGENCE APPLICATIONS IN SEMICONDUCTOR FAILURE ANALYSIS Anna Safont-Andreu1, Konstantin Schekotihin2, Christian Burmer3, Christian Hollerith3, and Xue Ming4 1Infineon Technologies AT, Villach, Austria 2Alpen-Adria-Universität, Klagenfurt, Austria 3Infineon Technologies AG, Munich, Germany 4Infineon Technologies Asia Pacific Pte Ltd, Singapore konstantin.schekotihin@aau.com EDFAAO (2023) 2:16-28 1537-0755/$19.00 ©ASM International® INTRODUCTION Over the decades, semiconductors have become very complex devices appearing in all spheres of life. Growing complexity poses a significant challenge for further research, development, fabrication, and after-sales support of these devices. It is widely recognized that artificial intelligence (AI) has a huge potential to solve many issues in the industry and generate vast business value. A recently conducted survey of semiconductor executives indicates that about 77% of companies are adopting AI in their businesses, and among them, about 63% expect AI to have a crucial impact on their businesses being a key approach to mitigate fast-rising expenses for each new technology step, such as the transition from 7 nm to 5 nm process.[1] In particular, semiconductor manufacturers plan to increase productivity in all activities using AI techniques. According to McKinsey,[2] at the moment, AI is used to only 10% of its potential within the industry. Utilization of its full capabilities might increase the added value to annual company earnings from the current $5–$8 billion to $85–$90 billion per year. Among the critical contributions of AI, the report highlights aspects of failure analysis (FA), such as defect identification/prediction and automation of deviation handling, as well as automated testing and root cause analysis. FA supports design, prototyping, manufacturing, and after-sales operations of a semiconductor company by providing identification and localization of root causes for the electrical malfunction of devices, performing quality and robustness analysis, or executing production control. In all these stages, fast and precise feedback from an FA lab is essential for their successful completion. As a result, FA activities bring much value to the whole semiconductor company (Fig. 1). However, all FA activities are very knowledge-intensive and quite tedious as they require an engineer to have a deep understanding of physical processes, possess all relevant information about the device, and be aware of best practices and common failure patterns appearing in images and measurements gained from previous analyses. Although modern AI methods cannot replace an FA engineer, they can significantly reduce the load by automating many of the time-consuming tasks like recognition of failure patterns, search and retrieval of information about devices, or scheduling of FA jobs and tools to engineers. Fig. 1 Value added by FA to the life cycle of semiconductor devices.
RkJQdWJsaXNoZXIy MTMyMzg5NA==