Process control techniques for semiconductor manufacturing processes
Techniques for measuring and/or compensating for process variations in a semiconductor manufacturing processes. Machine learning algorithms are used on extensive sets of input data, including upstream data, to organize and pre-process the input data, and to correlate the input data to specific features of interest. The correlations can then be used to make process adjustments. The techniques may be applied to any feature or step of the semiconductor manufacturing process, such as overlay, critical dimension, and yield prediction.
1. A method for predicting whether a semiconductor product will pass or fail a multi-step semiconductor process, comprising:
receiving input data from a first plurality of metrology measurements for a current process step and from a second plurality of metrology measurements for at least one upstream process step;
analyzing the input data to determine that a first set of the input data has a multi-variate relationship to at least one targeted process parameter for the current process step;
evaluating the multi-variate relationship of the first set of the input data to form an initial prediction as to whether the current process step will cause a final semiconductor product formed by the semiconductor process to pass or fail;
determining a confidence metric corresponding to the initial prediction; and
generating a final prediction that the final semiconductor product will fail when the initial prediction is that the final semiconductor product will fail, or generating the final prediction that the final semiconductor product will fail when the initial prediction is that the final semiconductor product pass and the confidence metric is less than a threshold value.
2. The method of claim 1 , further comprising:
creating a model for analyzing the input data, evaluating the multi-variate relationship of the input data for the targeted process parameter of the individual process step, and determining the confidence metric;
training the model using the input data; and
deploying the model for subsequent production runs of the semiconductor process.
3. The method of claim 2 , further comprising:
receiving new input data from the subsequent production runs; and
retraining the model using the new input data.
4. The method of claim 2 , wherein the model utilizes machine learning, predictive analytics and data mining techniques to analyze the input data, evaluate the multi-variate relationship of the input data, and determine the confidence metric.
5. The method of claim 2 , further comprising:
identifying the input data that are relevant to the targeted process parameter;
collecting the identified input data;
cleansing the identified input data;
transforming the identified input data; and
training the model using the cleansed and transformed input data.
6. The method of claim 1 , wherein the individual process steps of the semiconductor process include wafer fabrication, wafer test and sort, chip assembly and packaging, chip final testing, system level testing, and chip burn-in.
7. The method of claim 1 , wherein the input data includes key performance indicators, parametric measurements, and product performance measurements.
8. The method of claim 1 , wherein the metrology measurements include optical and non-optical measurements of the wafer fabrication process step.
9. The method of claim 7 , wherein the input data includes data from final wafer electrical testing, wafer sort tests, and wafer acceptance tests.
10. The method of claim 1 , wherein the final prediction is a weighted function that includes results of chip final testing, the confidence metric, and a yield classification.
11. The method of claim 2 , further comprising:
creating a virtual metrology model based on data received from the at least one upstream process step; and
providing an output of the virtual metrology model as input data to the model.
12. The method of claim 1 , further comprising utilizing the final prediction to determine next steps in the semiconductor process.
13. A method useful in a multi-step semiconductor process, comprising:
receiving sensor data from a plurality of pieces of production equipment during prior production runs of the semiconductor process;
receiving metrology measurements from a plurality of sampled wafers of the prior production runs;
determining a multi-variate relationship between the sensor data, the metrology measurements, and a targeted process parameter for one step of the multi-step semiconductor process; and
predicting metrology measurements for subsequent production runs based on the multi-variate relationship.
14. The method of claim 13 , further comprising:
creating a virtual metrology model for predicting metrology measurements;
training the virtual metrology model using the sensor data and metrology measurements from prior production runs; and
deploying the virtual metrology model for subsequent production runs of the semiconductor process.