Workshop 3
Semiconductor Manufacturing with AI
Organizer : Munehiro Tada (Keio University)
Join us as we explore the transformative potential of AI in semiconductor manufacturing. We'll delve into its applications across various domains, including materials engineering, chip design, digital twins, smart factories, and big data analysis. Discover how AI is revolutionizing each of these areas, driving innovation, efficiency, and precision in the semiconductor industry.
About Organizer
Munehiro Tada received the M.S. and Ph.D. degrees from Keio University, Japan, in 1999 and 2007, respectively. He joined NEC Corporation, Japan, in 1999. From 2007 to 2008, he was a visiting scholar at Stanford University. He is a cofounder and executive board director at NanoBridge Semiconductor, Inc. From 2024, he is a full professor at Keio University. He published papers with 16-IEDM, 16-VLSI and 10-IITC. He holds more than 100 issued patents. He is an IEEE Fellow and a JSAP Fellow. He is an adviser at JST-CREST(2020-) and an executive board director at JSAP(2024-).
- 1. Revolutionizing Chip Manufacturing through AI, Gaurav Thareja, Applied Materials
-
Abstract
Artificial Intelligence (AI) plays a pivotal role in navigating the complexities, driving innovation, and enhancing the efficiency of chip manufacturing. This complexity involves managing vast amounts of data related to materials, processes, equipment, devices, circuits, wafers, and fabrication facilities. Innovation in this field is guided by the PPACt metrics: low Power, high Performance, reduced Area, low Cost, and rapid time to market. Achieving efficiency hinges on yield and sustainability, with innovations often requiring years or even decades of meticulous refinement, progressing from the Concept & Feasibility stage to High Volume Manufacturing. This exploration will delve into how AI can accelerate the PPACt-guided semiconductor manufacturing process from various perspectives, including materials discovery, process and equipment optimization, device engineering, and chip design.
- 2. Recursive AI in Material Engineering, Yong-Ju Kang, Synopsys
-
Abstract Material engineering is becoming critical for AI systems, addressing thermal issues and wire resistance in advanced chip technologies. New dielectrics improve thermal management, while first-principles material engineering and innovative patterning techniques help manage wire resistance in GAA and CFET technologies. Challenges in ForkSheet and CFET technologies, which surpass current litho-etch capabilities, can be tackled by re-inventing ALD and ALE for material-guided patterning. For HPC demands, bandstructure engineering of HKMG stacks and channels is essential to minimize NBTI and maximize transistor driving strength. We will present a breakthrough in capacity and turn-around time of the first principles atomistic material engineering that can be instrumental in resolving the challenges in the tightly coupled thermal, interconnect, patterning, and transistor engineering. To achieve the required throughput of atomistic analysis, we heavily rely on AI, which makes it recursive
- 3. Leveraging AI for Enhanced Efficiency and Quality in Semiconductor Manufacturing, Susumu Shuto, Toshiba
-
Abstract Pioneering the integration of AI technology into semiconductor manufacturing, our initiative aims to create smart factories that significantly boost efficiency and product quality. The presentation highlights key efforts, including the establishment of a "smart factory" at a semiconductor manufacturing plant, where AI is utilized to streamline production processes. AI-powered image recognition technology automates defect classification, offering more accurate and faster detection compared to traditional methods. Additionally, the smart yield report system leverages big data and AI to swiftly identify yield issues, drastically reducing analysis time and enhancing product quality. These advancements underscore the commitment to leveraging AI for a competitive edge in semiconductor manufacturing.
- 4. Deep Topological Data Analysis and Self-Supervised Learning for Yield and Quality Optimization, Janhavi Giri, Intel
-
Abstract In semiconductor manufacturing, analyzing high-dimensional data is essential for quality and yield optimization. Traditional supervised machine learning struggles with the data's volume and complexity. We will present a novel AI framework combining Deep Topological Data Analysis (DTDA) and Self-Supervised Learning (SSL) to overcome these challenges. DTDA reduces high-dimensional data into two-dimensional graphs, identifying key features for quality control. SSL uses unlabeled data to detect patterns and anomalies without external supervision. Transfer learning enhances model adaptability to new datasets, reducing computational costs. Validated on public wafer map datasets, this framework excels in image segmentation, defect detection, and classification, improving process monitoring and decision-making while reducing costs. Its applicability extends beyond semiconductor manufacturing to other high-dimensional image data domains.
- 5. Enabling a new paradigm in semiconductor Design-Manufacturing Co-optimization through Simulation Technologies, Lorenzo Servadei, Sony
-
Abstract The semiconductor industry is driven the pursuit of smaller critical dimensions, optimal PPA (Power Performance Area) characteristics and multi-dimensional integration techniques. The constraints imposed by fab unit economics and tight yield learning cycles makes this an extremely challenging pursuit. In such an ecosystem, the interplay between design and manufacturing has become increasingly complex, which demands a paradigm shift in how we approach semiconductor production. On one-hand, the engineering design problem is both multi-scale and multi-physics, while on the manufacturing side, multi-dimensional silicon integration (2.5D, 3D, etc.) has introduced new challenges in aspects such as mechanical stability, thermal robustness, signal integrity. In this talk, we will delve into the transformative role of simulation technologies enhanced by modern data driven artificial intelligence techniques which are enabling such pursuits. Such a paradigm is promising for next generation technology nodes across various types of semiconductors, i.e. logic, memory, and sensors. It is becoming increasingly evident that systematic learning through data and algorithmic exchange between pre-silicon design and post-silicon manufacturing will drive the next generation of design-manufacturing co-optimization.
- 6. Integrating AI and GPU accelerated Simulation to build Real-time Digital Twins for Semiconductor Manufacturing Processes, John Linford, NVIDIA
-
Abstract We will explore the transformative potential of integrating AI and GPU-accelerated simulation technologies to drive significant breakthroughs in semiconductor manufacturing. Our discussion will cover techniques from augmenting classical Monte Carlo method by adding Ray Tracing features, to applying CUDA based 3D Partial Differential Equation (PDE) Solvers for large-scale 3D spatial problems. We will also delve into Physics-Informed Neural Networks (PINN), which create AI surrogates for solving complex multi-physics problems with changing boundary conditions. Additionally, we will showcase real-time digital twin simulations created in a blueprint flow that integrates various 3D PED solvers, tools for meshing and geometry morphing, and AI surrogate modules with training and fine-tuning capabilities to enable real-time AI simulations at a speed of orders of magnitude faster.