Speakers:
Haralampos-G. Stratigopoulos, Sorbonne Université, CNRS, LIP6, FR
Yiorgos Makris, The University of Texas at Dallas, US

Part I will motivate the need, the challenges, and the benefits of using machine learning and will discuss its utility on actual test- and yield-related industrial problems. We will give an abstract representation of problems that can be tackled using machine learning. We will also illustrate the link between machine learning and semiconductor manufacturing and test.

Speaker:
Yiorgos Makris, The University of Texas at Dallas, US

Part II will provide a concise and comprehensive overview of applications of machine learning in semiconductor manufacturing and test. For each application, we will define the problem, we will explain how machine learning can come to the rescue, and we will show a case study on industrial datasets. Applications include: alternate test for analog/mixed-signal/RF ICs, test compaction, fault diagnosis, yield learning, post-manufacturing tuning, outlier detection, adaptive test, wafer-level spatial & lot-level spatiotemporal correlation modeling, analog test metrics estimation, neuromorphic on-chip testers, hotspot detection, board-level fault diagnosis, trimming, die inking, pre-silicon verification and post-silicon validation, yield estimation in fab-to-fab migration, yield estimation when transitioning from one design generation to the next.

Speaker:
Haralampos-G. Stratigopoulos, Sorbonne Université, CNRS, LIP6, FR

Part III will illustrate the main practical issues when applying machine learning techniques. It will provide several recommendations based on the presenters' own experience in developing several applications in the past. Practical issues that will be discussed include: types of learning machines, feature extraction, feature selection, training and validation processes, dataset preparation, limited and unbalanced datasets, non-stationary datasets, metrics for generalization error, mitigating the generalization error, explainable artificial intelligence.

Speakers:
Haralampos-G. Stratigopoulos, Sorbonne Université, CNRS, LIP6, FR
Yiorgos Makris, The University of Texas at Dallas, US

Part IV will describe in more detail selected applications of machine learning in semiconductor manufacturing and test. We will delve into the following four mainstream applications: alternate test for analog/mixed-signal/RF ICs, adaptive test, yield learning, and hotspot detection. For each application we will discuss the collection of training data, the choice of learning models, the training procedures, etc., and we will provide several cases studies on actual industrial data.

Speaker:
Haralampos-G. Stratigopoulos, Sorbonne Université, CNRS, LIP6, FR

Part V will discuss emerging applications. In particular, we will discuss whether deep learning methods open new opportunities for solving efficiently test and semiconductor manufacturing problems. We will also discuss the "inverse" problem of testing machine learning hardware. In particular, we will discuss to what extent testing machine learning hardware is any different from testing any regular integrated circuit. We will also discuss fault tolerance methods that gain interest thanks to the integration of machine learning hardware in autonomous vehicles and systems.