Semiconductor manufacturing apparatus, failure prediction method for semiconductor manufacturing apparatus, and failure prediction program for semiconductor manufacturing apparatus

A failure prediction method for a semiconductor manufacturing apparatus, the method comprising:

• determining one or more measured feature quantities showing a state of a first device, wherein the first device is a component among a plurality of components of the semiconductor manufacturing apparatus, based on one or more physical quantities showing the state of the first device detected at the first device included in the semiconductor manufacturing apparatus, each measured feature quantity being set to a value based on each corresponding physical quantity among the one or more physical quantities;
• determining whether the first device is normal or abnormal by evaluating the measured one or more feature quantities with an identifier generated by learning with a support vector machine, using as training samples a normal data set of feature quantities representative of normal operation and failure data set of feature quantities representative of abnormal operation, or by evaluating the measured one or more feature quantities based on a degree of deviation of the measured one or more feature quantities from the normal data set;
• when the first device is determined to be abnormal, calculating a deviation between a measured feature quantities vector comprising the measured one or more measured feature quantities and a plurality of model feature quantities vector, wherein each of the one or more model feature quantities vectors comprises one or more model feature quantities corresponding to the measured feature quantities, and the calculation is performed at each time in a plurality of pieces of model data, wherein each piece of model data includes a temporal change in corresponding one or more model feature quantities in the model feature quantities vector, until the first device fails from a normal state of the first device, determining one or more pieces of model data including the one or more model feature quantities vector with the minimum deviation between the measured feature quantities vector and the model feature quantities vector at each time in the plurality of piece of model data, among the plurality of pieces of model data;
• and
• calculating one or more predicted times until failure from one or more differences between a failure point in time in the determined one or more pieces of model data and one or more points in time corresponding to the one or more model feature quantities that is obtained based on the one or more measured feature quantities which causes the determination that first device is operating abnormally, selecting the minimum predicted time from among the plurality of predicted times,, and
• stopping reception of a new substrate to prevent an introduction of defects on the new substrate when the minimum predicted time until failure is smaller than a predetermined threshold.

A non-transitory computer readable storage medium that stores a program for causing a computer to perform failure prediction for a semiconductor manufacturing apparatus, the program including

• determining one or more measured feature quantities showing a state of a first device, wherein the first device is a component among a plurality of components of the semiconductor manufacturing apparatus, based on one or more physical quantities showing the state of the first device detected at the first device included in the semiconductor manufacturing apparatus, each measured feature quantity being set to based on each corresponding physical quantity among the one or more physical quantities,
• determining whether the first device is normal or abnormal by evaluating the measured one or more feature quantities with an identifier generated by learning with the support vector machine, using as training samples a normal data set of feature quantities representative of normal operation and failure data set of feature quantities representative of abnormal operation, or by evaluating the measured one or more feature quantities based on a degree of deviation of the measured one or more feature quantities from the normal data set,
• when the first device is determined to be abnormal, calculating a deviation between a measured feature quantities vector comprising the measured one or more measured feature quantities and a plurality of model feature quantities vectors, wherein each model feature quantities vector comprises one or more model feature quantities corresponding to the measured feature quantities and the calculation is performed at each time in a plurality of pieces of model data, wherein each piece of model data includes a temporal change in corresponding one or more model feature quantities in the model feature quantities vector, until the first device fails from a normal state of the first device,
• determining one or more pieces of model data including the plurality of model feature quantities vector with the minimum deviation between the measured feature quantities vector and the model feature quantities vector among the plurality of model feature quantities vectors at each time in the plurality of piece of model data, among the plurality of pieces of model data, and
• calculating one or more predicted times until failure from one or more differences between a failure point in time in the determined one or more pieces of model data and one or more points in time corresponding to the one or more model feature quantities that is obtained based on the one or more measured feature quantities which causes the determination that the first device is operating abnormally, the minimum predicted time is selected from among the plurality of predicted times, and
• stopping reception of a new substrate to prevent an introduction of defects on the new substrate when the minimum predicted time until failure is smaller than a predetermined threshold.