Semiconductor Production Equipment_ Emergence of Flash Memory for AI Calculation and Implications for Japan’s SPE Industry

Summary of Key Points from the Conference Call Industry Overview - Industry: Semiconductor Production Equipment in Japan - Focus: Emergence of Flash Memory for AI Calculation and its implications for Japan's SPE industry [1] Core Company Insights - Company: Floadia (unlisted) - Development: Floadia is developing inferential AI devices utilizing flash memory, specifically focusing on computing in memory (CiM) technology [3][6] - Technology: The CiM technology operates at ultra-low power using SONOS flash memory, which traps electrons on a silicon nitride film, preventing charge leakage [3][6] - Power Efficiency: Floadia claims its devices use only one thousandth of the power of a GPU, targeting edge AI applications [6][7] Market Dynamics - Current Demand: Demand for flash memory is currently weak, and a significant boost to baseline demand seems unlikely in the near term [6] - Potential Beneficiaries: Companies that could benefit from any future expansion in flash memory applications include: - Kokusai Electric - Tokyo Electron - Daihen - Japan Material - Micronics Japan [6][9] Technological Advancements - MAC Operations: Floadia's CiM technology allows for multiply-accumulate (MAC) operations similar to analog circuits, enabling neural network computations [7] - Prototype Development: Floadia is working on commercializing a programmable AI chip called FP CiM, with a prototype already completed [7] Industry Outlook - Investment Trends: The current trend in capital expenditure (capex) for flash memory is low due to an oversupply, leading to falling plant utilization rates [9] - Future Opportunities: An increase in production volumes and capex driven by new applications could positively impact equipment and materials makers in the semiconductor industry [9] Additional Insights - Historical Context: The concept of using memory for AI inference is not new, with previous developments by companies like IBM using phase-change memory (PCM) for similar purposes [8] - Market Positioning: Floadia does not aim to replace GPUs but is focusing on the inferential edge device market, indicating a niche positioning within the broader AI hardware landscape [7]