Startups from MATSUO Lab the University of Tokyo

Matsuo Lab's Ecosystem and the Cutting-Edge of Semiconductors × AI by Matsuo Lab Startups
Matsuo Lab at the University of Tokyo has launched 26 startups. We will introduce this ecosystem and share the latest advancements in semiconductors, materials, and manufacturing driven by AI, from Matsuo lab's startup ecosystem.

In recent years, the performance improvements and accompanying large-scale advancements in foundational model development have been remarkable, with semiconductor technology (computational resources) playing a significant role in these achievements. This lecture will explore the future prospects of foundational model development, addressing both hardware and machine learning aspects.

Emuni, a startup originating from Kyoto University and the Matsuo Lab, develops custom-made AI specifically for the manufacturing industry. In this presentation, we will introduce the latest trends in the utilization of generative AI in manufacturing. Additionally, we will showcase examples of joint development between our company and manufacturing firms, including semiconductor companies. We will explain how generative AI can be applied in various scenarios, such as in manufacturing sites, design, and intellectual property.

By integrating AI into semiconductor manufacturing process optimization, complex production flows can be analyzed and adjusted in real time, maximizing yield. This leads to significant improvements in cost reduction and efficiency, thereby enhancing competitiveness. In the future, AI-driven autonomous manufacturing processes will accelerate the adoption of new technologies, ensuring sustainable growth.

This talk will focus on how AI technology is advancing dramatically with the evolution of semiconductor technology. This has led to increased efficiency and cost reductions in the industry, with revolutionary changes awaiting in a variety of fields. How to maximize the synergy between AI and semiconductors to pave the way for the future will also be discussed.

Collecting the latest methods and findings from the vast number of academic papers published daily is essential for materials research but requires significant time and effort.In this talk, we will introduce the development of a domain-specific large language model (LLM) that uses academic papers as its primary training data.We will also provide details on the differences in output compared to general-purpose models, practical applications, and the time and cost required for development.

This lecture will explore the potential for digital transformation (DX) in the semiconductor and manufacturing industries through the use of generative AI and large language models (LLMs). It will focus particularly on a roadmap for digitizing internal documents and utilizing them effectively, as well as the effectiveness of local LLMs in industries that handle sensitive information.