The Manufacturing Science and Technology Group (MSTG) two sessions this year address machine learning (ML) for both process control and materials discovery; and the related advanced characterization and metrology needed for such advanced manufacturing. As semiconductor manufacturing reaches its miniaturization limits in two-dimensions (2D), it has moved to 2.5D and 3D which greatly increases the number and importance of metrology steps as well as importance of decreasing process variability. ML although initially considered to be a subset of Artificial Intelligence (AI) and used to train AI, it is now a separate discipline focused more on probability theory and statistics that is becoming widely used worldwide especially in microelectronics’ to reduce process variability. In our morning session, speakers from a wide range of microelectronics manufacturing and research organizations will discuss improved or even autonomous process control. To date, commercial applications of ML for process control mainly have been for ex-situ parameter development. But for autonomous or other in-situ (real-time) uses of ML, energy use could become both an economic and environmental concern. Recently, however, experts have identified best practices that can lower in-situ ML energy use by orders of magnitude: For example, selecting efficient ML model architectures while advancing ML quality, such as sparse models versus dense modes, can reduce computation by factors of ~5–10 and using processors optimized for ML training such as TPUs or recent GPUs (e.g., V100or A100), versus general-purpose processors, can improve performance/Watt by factors of 2–5. For the afternoon session, we note semiconductor manufacturing trends that demand significantly advanced characterization and modelling: As device size shrinks toward the size of the probe being used, structures become more difficult to image accurately; Measuring structures or films that are not accessible from the surface or are hidden under pre-existing layers is a major challenge; In 3D, the complexity of structures increases geometrically with 3D device architectures and accessing some 3D features with non-destructive techniques can be difficult. The afternoon session also continues MSTG’s exploration of material discovery problems that are well suited to ML.
MS+HI-MoM: Machine Learning for Microelectronics Manufacturing Process Control
- Jeff David, PDF Solutions, “Progressing Process Control with Data-Centric AI”
- Di Du, ExxonMobil Technology and Engineering Company, “Machine Learning Accelerated Scale-up for Microporous Materials – An Industrial Perspective”
- Jun Shinagawa, Tokyo Electron America Inc., “Paths Toward Autonomous Plasma Process Tool Operation by Pairing of Plasma and Machine Learning Technologies”
MS+AP+AS+TF-MoA: Advanced Characterization and Metrology for 3D
- Bryan Barnes, NIST, “Semiconductor Metrology for Dimensional and Materials Scaling”
- Cornel Bozdog, Onto Innovation, “New in-Line Metrology for Advanced Semiconductor Nodes”
- Sergei Kalinin, Oak Ridge National Laboratory, “Autonomous Scanning Probe Microscopy: from Streaming Image Analysis to Learning Physics “
- Subramanian Sankaranarayanan, “Towards a Digital Twin for Spatiotemporal Experiments”
- Brian Valentine, DOE , “Applications of Artificial Intelligence AI and Machine Learning ML to Semiconductor Materials Discovery and Optimization”
MS3+HI: Machine Learning for Materials’ Discovery
MS-TuP: Manufacturing Science and Technology Poster Session