New improvements in storage, measurement, and control methods in semiconductor engineering are rapidly producing more data. Today, there are valuable tools for monitoring and gathering time-based data for manufacturing devices such as integrated circuits. Given the huge volume of data available for detecting faults and anomalies in devices, how should quality control industrial algorithms be designed to take advantage of numerous data points for such things as temperature, gas tide, electrical load, and electric partiality? In an effort to improve the engineering of semiconductor devices, Zhu and He present a method for clustering the time series of manufacturing data.

Advocating for the effective engineering of semiconductor devices, the authors present algorithms for concurrently clustering 2D time series of device manufacturing data. They present optimization algorithms that use alternative metrics to identify relevant variables for manufacturing semiconductor devices. The performance of the algorithms was experimentally evaluated using the typical categories of power, gas, and pressures for device engineering control practices. Other than a few classification errors, the co-clustering algorithm was used to accurately identify the comparable chamber and adjustable procedure actions for manufacturing semiconductor devices. The accuracy of the proposed co-clustering algorithm, in pinpointing the variables for engineering semiconductor devices, outperforms reputable clustering algorithms such as k-means and complex-valued linear dynamical systems (CLDS).

Clearly, the authors present a reliable generic clustering algorithm, adaptable to alternative distance functions for recognizing the pertinent variables for manufacturing semiconductor devices. Experiments were performed with benchmark, synthetic, and real-life manufacturing data to investigate the effectiveness of the proposed method. The results are adequately reliable. Computational statisticians should read this paper and consider ways to investigate the co-clustering framework beyond the limited 2D manufacturing data.