In TSMC, machine learning is used to implement auto defect classification (ADC) and keep the accuracy of Advanced Defect Classification (ADC) identification. The machine learning approach we use is the Human-in-the-loop (HITL) which leverages both human and machine intelligence to create machine learning models. With enough data and human tuning, those machine algorithms can identify and analyze images quickly and incredibly accurate without the need for engineers to teach the machine constantly what exactly defects look like. Therefore, the human intelligence will play a significant role in machine learning, which means the experiences and algorithms provided are critical and distinctive for successful machine learning models.

TSMC has widely deployed machining learning in Inline edge computing ADC and offline cloud computing ADC, especially focus on key manufacturing stages for tool defect defense. The function and benefit of ADC is to deal with above millions of images per day, improve image defect inspection sampling rate. For inline edge computing ADC, so called “eagle’s view”, which is embedded in the tool and detect the defects while processing and isolating the material. For offline cloud computing ADC, which is inserted into process afterward while detecting the defect and then holding lots to prevent furthermore loss. The flexibility of machining learning is developed and extended to sustain variant engineering requirements, such as size Measurement, defect classification and particle detection to extend the scale of quality management.

For committed of customer trust, lot-level and wafer-level traceability are now insufficient. Even more, the comprehensive approach to material and die traceability is the reliability requirements for packaging manufacturing.

TSMC constantly has enhanced the data management of the full traceability even the per-die mapping to the fab wafer position. We have used 2D barcode to encode all the product data and to trace back all traceability information such as source wafer position, bin codes, and engineering experiment labels. The 2D barcode mark is individually generated for every package. The generated 2D barcode are combined with per-package unique lithography patterns in a patterning software system. When uploading the unique mark, all the process and product information including process history, tool log, material and production yield, will be listed and searched, so called “product resume”. Using this product resume, we can quickly define impact scope for problematic material or process issue and analyze the root cause of low yield by data correlation which can minimize defect impact scope.