Maintenance scheduling for semiconductor manufacturing equipment

1. A method, comprising:

collecting a first plurality of sensor measurements from a first plurality of sensors associated with a first component of semiconductor manufacturing equipment during regular production runs of the semiconductor manufacturing equipment;

determining a relationship between each of the first plurality of sensor measurements and each of a plurality of failure modes of the first component;

selecting a first subset of the first plurality of sensor measurements determined to be most indicative of a first failure mode of the plurality of failure modes based on the determined relationship;

selecting a second subset of the first plurality of sensor measurements determined to be most indicative of a second failure mode of the plurality of failure modes based on the determined relationship; and

evaluating a difference between the first and second subsets of the first plurality of sensor measurements in order to predict one of the plurality of failure modes of the first component.

2. The method of claim 1 , further comprising:

predicting a remaining useful life of the first component based on the evaluation of the first and second subsets; and

scheduling maintenance of the first component when the predicted remaining useful life of the first component is less than a threshold value.

3. The method of claim 1 , further comprising:

configuring a first model to perform the step of selecting flan corresponding subsets of the first plurality of sensor measurements; and

configuring a second model to perform the step of evaluating the corresponding subsets.

4. The method of claim 3 , further comprising:

the first model is a physics-based model and the second model is a data-driven model.

5. The method of claim 4 , further comprising:

the first model and the second model are implemented as machine learning models trained on historical sensor measurements.

6. The method of claim 4 , the first model further comprising:

simulating a first failure mode of the first component and modeling a second plurality of sensor measurements from the first plurality of sensors as a result of the simulated first failure mode, each of the second plurality of sensor measurements corresponding to a respective one of a first plurality of sensor measurements;

measuring a plurality of signal shifts between respective ones of the first plurality of sensor measurements and corresponding ones of the second plurality of sensor measurements;

evaluating the plurality of signal shifts in order to determine the relationship between each of the first plurality of sensor measurements and the simulated first failure mode of the first component; and

selecting the first subset for of the first plurality of sensor measurements based on the relationship between each of the first plurality of sensor measurements and the simulated first failure mode exceeding a first threshold criteria indicative of the first failure mode.

7. The method of claim 3 , further comprising:

adapting and tuning the first model and the second model as more sensor measurements are collected.

8. The method of claim 7 , further comprising:

collecting additional data sets of sensor measurements that are beneficial for tuning the first model and the second model to recognize a false positive, a false negative, or a false classification; wherein the false positive results when the evaluation step identifies the first failure mode of the first component in error, the false negative results when the evaluation step fails to identify the second failure mode of the first component when present, and the false classification results when the evaluation step identifies a third failure mode of the first component but misidentifies a root cause of the third failure mode.

9. The method of claim 1 , further comprising:

evaluating additional pluralities of sensor measurements, each of the additional pluralities of sensor measurements associated with one or more components of the semiconductor manufacturing equipment, in order to determine a relationship between each one of the pluralities of sensor measurements and each one of a plurality of failure modes associated with respective components;

for each of the plurality of failure modes associated with respective components, selecting a corresponding subset of the pluralities of sensor measurements determined to be most indicative of the failure mode for a respective component; and

evaluating the corresponding subsets of sensor measurements to determine potential failure modes of each respective component.

10. The method of claim 4 , the second model further comprising:

a classification model for classifying a plurality of known failure modes.

11. The method of claim 10 , further comprising:

detecting an unknown failure mode;

classifying the unknown failure mode; and

adding the unknown failure mode to the classification model.

12. A method, comprising:

determining, using a physics-based model, a relationship between each of a first plurality of sensor measurements obtained from a first plurality of sensors associated with a first component of semiconductor manufacturing equipment and each of a plurality of failure modes of the first component;

selecting, using the physics-based model, a plurality of subsets of the first plurality of sensor measurements, each of the plurality of subsets selected to represent a corresponding one of the plurality of failure modes on the basis of the determined relationship of each of the first plurality of sensor measurements to the corresponding failure mode; and

evaluating, using a data-driven model, differences among the corresponding subsets of the first plurality of sensor measurements to diagnose one of the plurality of failure modes of the first component.

13. The method of claim 12 , further comprising:

predicting a remaining useful life of the first component based on the evaluation of the corresponding subsets; and

scheduling maintenance of the first component when the remaining useful life of the first component is less than a threshold value.

14. The method of claim 12 , further comprising:

the relationship between each of the first plurality of sensor measurements and each of the plurality of failure modes is determined by inducing a first failure mode in the first component and comparing shifts in the sensor measurements of the corresponding subset.

15. The method of claim 14 , further comprising:

for each of the plurality of failure modes, the corresponding subset is selected based on the impact the failure mode has on the sensor measurements.

16. The method of claim 14 , further comprising:

for each of the plurality of failure modes, the corresponding subset is selected based on the relationship exceeding a first threshold criteria indicative of the first failure mode.

17. The method of claim 12 , further comprising:

evaluating additional pluralities of sensor measurements, each of the additional pluralities of sensor measurements associated with one or more components of the semiconductor manufacturing equipment, in order to determine a relationship between each one of the pluralities of sensor measurements and each one of a plurality of failure modes associated with respective components;

for each of the plurality of failure modes associated with respective components, selecting a corresponding subset of the pluralities of sensor measurements determined to be most indicative of the failure mode for a respective component; and

evaluating the corresponding subsets of sensor measurements to diagnose one of the plurality of failure modes of each respective component.

18. A method, comprising:

using a physics-based model for feature engineering of sensor measurement data to select a subset of the sensor measurement data to represent a failure mode of a semiconductor manufacturing component; and

using a data-driven model to evaluate the subset of sensor measurement data to diagnose the failure mode.

19. A diagnostic engine for components of semiconductor manufacturing equipment, comprising:

a physics-based model configured to (i) evaluate a first plurality of sensor measurements associated with a first component of semiconductor manufacturing equipment; (ii) determine, based on the evaluation, a relationship between each of the plurality of sensor measurements and each of a plurality of failure modes of the first component; and (iii) for each failure mode of the first component, select a corresponding subset of the first plurality of sensor measurements determined to be most indicative of the failure mode; and

a data-driven model configured to evaluate the corresponding subsets of the plurality of sensor measurements in order to determine at least one of the plurality of failure modes.

20. The diagnostic engine of claim 19 , further comprising:

a prognostic engine for predicting a remaining useful life of the first component based on the evaluation of the corresponding subsets; and

scheduling maintenance of the first component when the remaining useful life of the first component is less than a threshold value.