AVS 65th International Symposium & Exhibition
    Vacuum Technology Division Tuesday Sessions
       Session VT-TuP

Paper VT-TuP5
Smart Diagnostics for Dry Vacuum Pumps Running in Semiconductor Processes

Tuesday, October 23, 2018, 6:30 pm, Room Hall B

Session: Vacuum Technology Division - Poster Session
Presenter: Wan-Sup Cheung, KRISS, Republic of Korea
Authors: W.S. Cheung, KRISS, Republic of Korea
J. Lim, KRISS, Republic of Korea
N.K. LEE, SK Hynix, Republic of Korea
J.B. LEE, SK Hynix, Republic of Korea
T.J. Park, SK Hynix, Republic of Korea
T.H. Kim, SK Hynix, Republic of Korea
Correspondent: Click to Email
This paper addresses logical and smart ways of diagnosing the operation conditions of dry vacuum pumps in the semiconductor processes of the SK Hynix company. In the last four years, SK Hynix has been attempting to complete the ‘production data logging’ systems, including even the collection of the state variables of dry vacuum pumps running in the semiconductor processes. Recent attempts have been added to exploit the logged state variables to diagnose individual dry vacuum pumps operating in the highly tough and risky production processes since their failures always accompany much time and cost. Adaptive parametric modelling (APM) approaches developed by the KRISS research team were considered to analyze the trend and statistical properties of the logged state variables of vacuum pumps. APM approaches are illustrated to provide merits of dramatic (more than 100 times) data reduction for the construction of ‘normal operation condition (NOC)’ reference batches from the beginning part of the logged state variables. The singular value decomposition (SVD) is exploited to decompose the selected NOC reference batch matrix into the sore and projection matrices, which are shown to enable the evaluation of two diagnostic indicators for each vacuum pump, the Hotelling’s T-squared (T2) value and the sum of squared residuals (Q). Both scalar values T2 and Q are demonstrated to play critical roles in diagnosing the current operation conditions of individual vacuum pumps, more specifically the similarity and difference of the current process state in reference to the NOC reference batches. In this work, several attempts have been carried out to improve the reliability and robustness of the diagnosis of vacuum pumps. The use of vibration measurements was the first selected candidate. Field tests are now in progress. It is expected that field test results are to be illustrated in the presentation of this paper.