回帰機械学習を用いた純水の高圧スプレーで生じる静電気量の予知技術

DOI Web Site

Bibliographic Information

Other Title
  • カイキ キカイ ガクシュウ オ モチイタ ジュンスイ ノ コウアツ スプレー デ ショウジル セイデンキリョウ ノ ヨチ ギジュツ

Search this article

Abstract

High-pressure spray cleaning is commonly used in the semiconductor device manufacturing process. In this study, a regression machine-learning algorithm was developed to predict generated static electricity, based on data from multiple sensors, to prevent electrostatic discharge （ESD） during high-pressure spray cleaning. As a countermeasure to ESD, generated current was suppressed by applying a high negative voltage to pure water, and by heating the water. A synergistic effect was achieved by combining these methods. We devised a system for predicting static electricity and avoiding ESD and made a regression machine-learning algorithm as a soft sensor for the system based on these data. Twenty-four regression machine-learning algorithms were prepared using the pressure of the spray and temperature of the pure water as control factors, and the flow rate and resistivity of the pure water, as observation factors, were evaluated in terms of root mean square error （RMSE）, coefficient of determination, predicted speed and learning time. The results showed that the ensemble learner could be predicted with an RMSE value of 0.408 µA. A coefficient of determination of 0.92, a prediction speed 13.0 µs/unit or lower, and a learning time of 5.56 s.

Journal

Keywords

Details 詳細情報について

Report a problem

Back to top