Abstract: Artificial intelligence (AI)-driven electronic design automation (EDA) techniques have been extensively explored for VLSI circuit design applications. Most recently, foundation AI models for circuits (e.g., Large Language Models, Large Circuit Models) has emerged as a new technology trend. These models typically leverage a two-stage paradigm of pre-training on large-scale datasets followed by fine-tuning for specific applications, significantly enhancing adaptability across various EDA tasks. Their great potential has attracted wide attention from the EDA community, with representative works being relatively highly cited. In this panel, we invited 6 distinguished researchers with active publications in this field. We will discuss the latest research outputs, existing challenges, and future directions about the development foundation AI models for EDA methodologies.

Key Questions

• Question 1: There have been many general LLMs as commercial products, such as GPT or DeepSeek. Do you think foundation AI models (e.g., LLMs) will be adopted in the semiconductor industry as mature products in the near future (5 years)? What are the obstacles in the deployment?
• Question 2: Do you think circuit data will limit the development of foundation AI models? How can we solve the data availability problem? Do you think the computation resources will be the limit?
• Question 3: Do you think academia or industry will lead this research trend? Will the SOTA foundation circuit models in the future be open-source or in-house?
• Question 4: Do you think large foundation models for circuits will replace existing IC design jobs, considering both digital and analog design? Which positions?

Abstract: Integrating Large Language Models (LLMs) into IC manufacturing represents a transformative shift in how semiconductor processes are optimized, analyzed, and automated. LLMs have the potential to bridge knowledge gaps, accelerate design-to-manufacturing workflows, and enhance decision-making across DFM, TCAD, and Lithography. However, challenges remain in their trustworthiness, adaptability, and integration with existing workflows. This panel will bring together leading experts from academia and industry to explore where and how LLMs can bring real value to IC manufacturing, the technical and industrial barriers to adoption, and the long-term implications for semiconductor process optimization, cost reduction, and innovation acceleration. Through this discussion, we aim to uncover whether LLMs will merely serve as a complementary tool or drive a paradigm shift in how IC manufacturing is approached in the AI era.

Key Questions

• Question 1: Which areas of IC manufacturing can LLMs fundamentally transform, and where will their impact be most significant?
• Question 2: What are the key technical and practical challenges in applying LLMs to IC manufacturing, and how can they be addressed?
• Question 3: What unique opportunities do LLMs create for enhancing efficiency, collaboration, and knowledge transfer in semiconductor manufacturing?
• Question 4: How can LLMs be integrated with traditional AI/ML and physics-based simulation methods to achieve the best results?
• Question 5: What are the risks of relying on LLMs in a manufacturing process, and how can the industry mitigate them?
• Question 6: Will LLMs lead to a fundamental shift in IC manufacturing, or will they remain a supporting tool for existing methodologies?

Abstract: As semiconductor technology advances, multiphysics challenges—spanning thermal, electromagnetic, semiconductor physics and beyond—have become increasingly complex and computationally demanding. Traditional simulation and modeling techniques often struggle to balance accuracy, efficiency, and scalability, creating a pressing need for innovative approaches. AI has emerged as a powerful tool to revolutionize multiphysics analysis, enabling faster, more efficient, and intelligent solutions in EDA. This panel will bring together leading scholars and industry experts to discuss how AI is transforming multiphysics analysis in EDA. Topics will include AI-accelerated simulation, machine learning-based surrogate modeling, and physics-informed neural networks. The panelists will also explore the challenges of incorporating AI into existing design flows and its impact on design accuracy, efficiency, and manufacturability.

Key Questions

• Question 1: Can AI accelerate multiphysics simulations without compromising accuracy, reliability, or verification standards?
• Question 2: Which is the more promising direction: integrating AI into physics-based modeling, or embedding physics principles into AI architectures?
• Question 3: Can data-driven methods eventually earn enough trust to replace physics-based solvers in critical stages like design sign-off?