Kioxia AI Smart Factory: 3 Billion Data Points in Yokkaichi

Title

Kioxia AI Smart Factory: 3 Billion Data Points in Yokkaichi

Content

Kioxia's Yokkaichi Plant in Mie Prefecture, Japan, is one of the world's largest and most efficient flash memory production facilities — approximately 1.4 km from one end to the other. Kioxia started operation of its new fabrication facility (Fab7) in the fall of 2022. The plant has been a smart factory since well before the current AI wave, with automated control of flash memory production as early as 20 years ago and full machine learning adoption in the mid-2010s. As Yukako Tanaka, an engineer from Manufacturing Engineering Department II at Kioxia's Yokkaichi Plant, explains, "The entire lifecycle of wafers, from the moment they enter the cleanrooms through the manufacturing process to the moment they leave as finished products, is converted into data." Every tiny piece of information (such as what day and time a certain wafer was processed, which manufacturing equipment was used, and what the settings of the equipment at that time were) is acquired as data. A large volume of data is also output in the course of evaluating manufactured flash memory, with detailed testing performed down to a single bit. As a result, the Yokkaichi Plant generates 3 billion data points every day from several thousand process and inspection/measurement tools. The scale of data exceeds what engineers can manually process, so the plant has been leveraging AI since an early stage to enhance the efficiency and speed of data analysis. Yokkaichi employs AI to analyze the 3 billion pieces of "big data" generated daily from all of its production facilities for use in producing high quality, high value added flash memory including BiCS FLASH. The data are mainly related to the results of sensing, tests, and measurements, and are collected into a massive integrated database for productivity improvement. A semiconductor memory fab is a large-scale plant consisting of several thousand tools that process nanoscale devices. Utilization of data is crucial to quickly grasp what is happening at the micro and macro levels. Kioxia has been working to enhance the sophistication of big data analysis using AI and machine learning technology, continually improving its capability to infer the root causes of device failure and detect signs of defects automatically. The Advanced Equipment Control/Advanced Process Control (AEC/APC) Symposium Asia 2019 awarded Kioxia the best research paper award for "Robust Estimation of Mixed-Type Wafer Map Similarity Utilizing Non-negative Matrix Factorization." With the rapid expansion of AI making data more valuable than ever, the role of flash memory is becoming increasingly important, and Yokkaichi Plant plays a central role in Kioxia's AI-era memory supply.