Toshiba taps AI to boost productivity at memory plant

Toshiba expects to root out trouble spots in production much more quickly thanks to a new analytics system based on artificial intelligence at its mainstay flash memory plant in Japan.

At the Yokkaichi plant in Mie Prefecture, AI is being used in processes that chiefly monitor the semiconductor yield rate. For circuit-containing wafers that fail to meet quality standards, AI will automatically classify the defects, detect causes and analyze incident trends, among other tasks.

Up until now, the wafers’ surface temperatures and voltages would be collected and scrutinized by technicians during major steps of the fabrication process, such as lithographic exposure and polishing. During the testing phase of the AI system, the time needed to discover faulty manufacturing processes or equipment was cut to less than two hours from the five or six hours it used to take, according to the Tokyo-based electronics maker.

The company will also incorporate AI technology into wafer defect tests sometime this fiscal year. The AI program will engage in deep learning by reading and analyzing image data of defective wafers and other components beforehand. This will allow it to automatically detect specks of dust or irregular surface colors on wafers.

Toshiba is looking to adopt the system in other production facilities, as well as sell it to other manufacturers.

Technologies advance quickly in the semiconductor field, resulting in smaller chips with greater circuit densities. Combined with other innovations, the manufacturing process has become increasingly complex. Fabrication can take around three months and incorporate 20,000 steps both minor and major. Because of this, many companies see wide latitude for process streamlining using AI.

Since 2005, Sony has been using AI analysis to identify causes of defective products in semiconductor fabrication. Its Nagasaki Prefecture production center has reportedly seen the yield rate increase by 3%, leading to 3.6 billion yen ($35 million) in annual cost savings. The company’s image sensor plant in Kumamoto Prefecture also employs an AI-based system.

Renesas Electronics employs AI to predict when defective products will be created by looking at voltages to ascertain surface conditions of wafers when special gases are applied to them.

Give it a share: