NUS Computing’s Just-in-Time.AI wins first place at Micron Industrial Data Challenge

NUS Computing's Just-in-Time.AI wins first place at Micron Industrial Data Challenge

Held in conjunction with ICQSR 2025 in Singapore. The competition tasked participants with predicting wafer-level quality from complex, multi-sensor time-series signals, while accounting for tool ageing and refurbishment effects; finalists presented their solutions at Micron's workshop on 30 June 2025.

The challenge, titled Signals to Quality: Unlocking Insights from Process Data, evaluated teams on prediction accuracy, methodological suitability/innovation, and clarity of reporting. NUS SoC's Just-in-Time.AI comprised PhD candidates Yihao Ang, Yifan Bao, Shuyu Lu, and Professor Anthony K. H. Tung. Their winning solution, TSEMi, blends targeted feature engineering with a compact, high-generalisation tree-boosting model. The team converted variable-length sensor time series into comparable descriptors, improving signal separability across tools and steps. After feature extraction, a gradient-boosted decision-tree ensemble model delivered strong generalisation with minimal tuning. Analyses showed engineered features dominated importance rankings, and simple features often captured the key trends effectively.

Judges from academia and industry highlighted the team's clear framing of the semiconductor prediction problem and their pragmatic balance of accuracy, efficiency, and interpretability, hallmarks of deployable AI in manufacturing. The result underscores NUS Computing's strengths in data science and its translation to high-impact industrial applications.