Tool health monitoring and classifications with virtual metrology and incoming wafer monitoring enhancements

1. A method of evaluating tool health of a plasma tool, the method comprising:

providing a classification model that identifies a plurality of failure modes of the plasma tool;

performing an initial test on an incoming wafer by performing a measurement on the incoming wafer;

determining that the incoming wafer meets a preset requirement based on the initial test, in response to determining that the incoming wafer meets a preset requirement based on the initial test, executing a plasma etching process, and predicting a wafer characteristic associated with the plasma etching process using a virtual metrology (VM) model that is configured to predict the wafer characteristic based on parameters measured by module sensors and in-situ sensors of the plasma tool;

determining that the wafer characteristic predicted by using the VM model is outside a pre-determined range and identifying a current failure mode using the classification model; and

based on the current failure mode, adjusting a recipe of the plasma etching process or taking a corrective action for the plasma tool.

2. The method of claim 1 , wherein providing the classification model comprises:

determining predictor parameters;

removing collinearity among the predictor parameters to obtain key predictor parameters;

selecting a subset of the key predictor parameters based on relevance to the plurality of failure modes; and

building the classification model using the subset of the key predictor parameters.

3. The method of claim 2 , wherein determining the predictor parameters comprises:

determining target wafer characteristics; and

determining failure modes for the target wafer characteristics based on occurrence and sensitivity of the failure modes so that the parameters from the module sensors and the in-situ sensors are classified into different categories for the failure modes.

4. The method of claim 3 , wherein a fault detection model is, constructed with one or more parameters from the module sensors without using parameters from the in-situ sensors, the method further comprising adding the one or more parameters from the module sensors to a first subgroup of predictor parameters.

5. The method of claim 4 , wherein building a fault detection model entails using one or more parameters from the in-situ sensors, the method further comprising adding the one or more parameters from the in-situ sensors to a second subgroup of predictor parameters.

6. The method of claim 5 , wherein determining the predictor parameters further comprises:

obtaining a third subgroup of predictor parameters by processing the parameters from the module sensors and the in-situ sensors using domain knowledge including knowledge of the plasma tool, a plasma process associated with the plasma tool, metrology and/or the wafer; and

processing the third subgroup of predictor parameters to remove error and variance.

7. The method of claim 5 , further comprising building a VM model associated with a failure mode using the second subgroup of predictor parameters.

8. The method of claim 5 , wherein the classification model comprises a plurality of fault detection models.

9. The method of claim 5 , further comprising integrating a plurality of fault detection models into a single multi-class classification model by applying a machine learning algorithm.

10. The method of claim 2 , wherein providing the classification model using the subset of the key predictor parameters comprises regression analysis that includes at least one of a logistic regression, a support vector machine regression, a decision tree regression or a linear regression.

11. The method of claim 1 , wherein performing the initial test comprises:

measuring a reflectivity of the incoming wafer;

providing a test model that predicts the wafer characteristic based on the reflectivity; and

predicting the wafer characteristic using the test model.

12. The method of claim 11 , wherein the incoming wafer meets the preset requirement when the wafer characteristic predicted by using the test model is within a preset range.

13. The method of claim 1 , wherein the current failure mode allows for a process control, and the recipe of the plasma etching process is adjusted.

14. The method of claim 13 , wherein the current failure mode comprises a worn part of the plasma tool.

15. The method of claim 1 , wherein the current failure mode does not allow for a process control, and the corrective action is taken.

16. The method of claim 15 , wherein:

the failure mode includes deposition on a chamber wall, and

the corrective action includes seasoning to reset the chamber.

17. The method of claim 15 , wherein:

the failure mode includes radio frequency (RF) generator power output, and

the corrective action includes RF generator service.

18. The method of claim 1 , wherein the module sensors include at least one of a pressure manometer, a gas flow meter or RF power meter.

19. The method of claim 1 , wherein the in-situ sensors include at least one of a reflectometer, a plasma sensor, an RF sensor or a voltage and current (VI) sensor.

20. The method of claim 1 , further comprising determining that the wafer characteristic predicted by using the VM model is within the pre-determined range, and continuing to process a new wafer.

21. The method of claim 1 , wherein the wafer characteristic is selected from the group consisting of a critical dimension (CD), an etch rate (ER), particles and defects.