Taiwan R&D Group Rolls AI for Fault Prediction

　　Taiwan’s Industrial Technology Research Institute (ITRI), one of the world's largest high-tech research institutions, has introduced AI software that can predict breakdowns in production-line equipment, helping to eliminate factory downtime and boost human productivity.

　　ITRI’s Prognostic and Health Management Software in Semiconductors (PHM) uses AI-based machine learning algorithms for fault prognosis to predict equipment failure and component wear in chip fabs and other manufacturing operations. While similar attempts have been made for over 20 years, prediction accuracy has up to now fallen short of manufacturer requirements, according to ITRI.

　　Malfunctions during production still require equipment engineers to diagnose the problem, which can take several days and may influence production capacity and quality. PHM has high scalability, without restriction to specific facilities, according to ITRI. In addition to applications in the semiconductor industry, PHM is effective in fault prognosis for equipment used in medical and power generation facilities, greatly reducing the uncertainty of machine malfunction.

　　The key advantage is cost savings, according to one of the ITRI people who developed the software.

　　“Unscheduled downtime can be avoided,” said ITRI Manager Chung-Wei Lin in an interview with EE Times. “Our solution realizes fault diagnostics, failure prediction and remaining useful life prediction so factory scheduling and dispatching can be performed well.”

　　ITRI has conducted trials of the PHM software with seven companies, one of which is Winbond, a memory chip manufacturer in Taiwan. Another company that has implemented the software is Marketech International Corp., an LED manufacturer.

　　While most of the companies working with ITRI have declined to disclose their names, three manufacturers have implemented the software in their production lines, and the others are still in the trial stage, Lin said.

　　PHM can predict when a hardware component or software indicator will fail, giving a technician or engineer enough time to eliminate problems and take corrective measures before damage is done, according to ITRI.

　　The PHM software can achieve more than 90 percent to 95 percent prediction accuracy in different production situations, ITRI says. The AI software effectively ensures product quality especially for high-mix, low-volume operations, according to the Taiwan government-funded R&D institute.

　　While Lin declined to disclose the cost of the software to a potential customer, he said that after implementation, the software can reduce manpower requirements by about 15 percent, allowing engineers and equipment operators to spend time on higher priority tasks.

　　ITRI aims to license the software to more manufacturers in the electronics industry.

　　“Generally, we need three months to run our trial program to evaluate whether our solution is suitable for their production line,” Lin said. “Then it takes one to two years to deploy the solution in production.”

　　ITRI is the R&D organization credited with the creation of a number of spinoff companies in the semiconductor industry such as Taiwan Semiconductor Manufacturing Co. and United Microelectronics Corp. Many of the key people in Taiwan’s semiconductor industry are former ITRI employees.