Skip to Main Content
An adaptive neural network-based advanced process control software, the Dynamic Neural Controller™ (DNC), was employed at National Semiconductor's 200 mm fabrication facility, South Portland, Maine, to enhance the performance of metal etch tools. The installation was performed on 5 identical LAM 9600 TCP Metal etchers running production material. The DNC produced a single predictive model on critical outputs and metrology for each tool based on process variables, maintenance, input metrology and output metrology. Although process metrology is usually measured on only one wafer per lot, the process can be closely monitored on a wafer-by-wafer basis with the DNC models. The DNC was able to provide recommendations for maintenance (replacing components in advance of predicted failure) and process variable adjustments (e.g. gas flow) to maximize tool up time and to reduce scrap. This enabled the equipment engineers to both debug problems more quickly on the tool and to make adjustments to tool parameters before out-of-spec wafers were produced. After a comparison of the performance of all 5 tools for a 2-month period prior to DNC installation vs. a 2-month post-DNC period, we concluded that the software was able to predict when maintenance actions were required, when process changes were required, and when maintenance actions were being taken but were not required. We observed a significant improvement in process Cpks for the metal etchers in this study.