Method and apparatus for determining cause of abnormality in a semiconductor manufacturing chamber

A non-transitory computer-readable recording medium having stored therein a program that causes an information processing apparatus to execute a process comprising:

• acquiring sensor waveform data represented with respect to a sensor value axis and a time axis, measured by a semiconductor manufacturing apparatus that is executing a process according to a same recipe;
• calculating a monitoring band for the sensor waveform data represented with respect to the sensor value axis and the time axis used in a waveform monitoring method from a predetermined number or more of the sensor waveform data;
• monitoring a waveform of the sensor waveform data using the monitoring band for the sensor waveform data represented with respect to the sensor value axis and the time axis and detect an abnormality sign of the semiconductor manufacturing apparatus;
• calculating a result of monitoring a waveform of the sensor waveform data using the monitoring band for the sensor waveform data as a score for a sensor and a processing step;
• identifying a target sensor showing abnormality and its corresponding processing step as a target pair based on the abnormality sign detection;
• calculating correlation coefficients between scores of each combined pair of the sensors and the processing steps for both normal and abnormal time;
• calculating a contribution rate for each combined pair based on:
• (i) a difference between a latest degree of divergence and an average degree of divergence during learning, and (ii) a difference between correlation coefficients at normal time and at abnormal time;
• creating a contribution rate table on combinations of sensors and processing steps that are highly related to behavior of the target sensor based on the calculated contribution rates;
• estimating a cause of the abnormality sign from a factor estimation rule of the cause of the abnormality sign by combining (i) the contribution rate table with (ii) a factor estimation rule stored in a factor estimation rule storage;
• and notifying a user of the cause.

A monitoring method executed by an information processing apparatus, the method comprising:

• acquiring sensor waveform data represented with respect to a sensor value axis and a time axis, measured by a semiconductor manufacturing apparatus that is executing a process according to a same recipe;
• calculating a monitoring band for the sensor waveform data represented with respect to the sensor value axis and the time axis used in a waveform monitoring method from a predetermined number or more of the sensor waveform data;
• monitoring a waveform of the sensor waveform data using the monitoring band for the sensor waveform data represented with respect to the sensor value axis and the time axis and detecting an abnormality sign of the semiconductor manufacturing apparatus;
• calculating a result of monitoring a waveform of the sensor waveform data using the monitoring band for the sensor waveform data as a score for a sensor and a processing step;
• identifying a target sensor showing abnormality and its corresponding processing step as a target pair based on the abnormality sign detection;
• calculating correlation coefficients between scores of each combined pair of the sensors and the processing steps from the score for both normal and abnormal time;
• calculating a contribution rate for each combined pair based on:
• (i) a difference between a latest degree of divergence and an average degree of divergence during learning, and (ii) a difference between correlation coefficients at normal time and at abnormal time;
• creating a contribution rate table on combinations of sensors and processing steps that are highly related to behavior of the target sensor based on the calculated contribution rates;
• estimating a cause of the abnormality sign from a factor estimation rule of the cause of the abnormality sign by combining (i) the contribution rate table with (ii) a factor estimation rule stored in a factor estimation rule storage;
• and notifying a user of the cause.