Anomalous equipment trace detection and classification
Scheme for detection and classification of semiconductor equipment faults. Sensor traces are monitored and processed to separate known abnormal operating conditions from unknown abnormal operating conditions. Feature engineering permits focus on relevant traces for a targeted feature. A machine learning model is built to detect and classify based on an initial classification set of anomalies. The machine learning model is continuously updated as more traces are processed and learned.
1. A process for detecting and classifying anomalies in semiconductor equipment trace data, comprising:
receiving a multiplicity of original traces from respective semiconductor equipment sensors into a processor-based model;
identifying at least a first key feature for the multiplicity of original traces using the processor-based model;
identifying by the processor-based model a plurality of anomalies in at least a first set of traces of the multiplicity of original traces;
modifying each of the first set of traces on the basis of the first key feature using the processor-based model;
generating a basis set of features from each of the modified first set of traces using the processor-based model;
creating a plurality of reconstructed traces from the modified first set of traces and the basis set of features using the processor-based model;
training a machine learning model to detect and classify anomalies based on the first set of traces, the modified first set of traces, and the reconstructed traces; and
deploying the machine learning model in a semiconductor equipment operating environment for identifying and classifying anomalies in trace data.
2. The process of claim 1 , further comprising:
receiving additional traces from the respective semiconductor equipment sensors;
identifying additional anomalies in a first set of the additional traces;
modifying the additional traces to remove a basic set of statistical characteristics;
generating the basis set of features from the modified additional traces;
creating reconstructed additional traces from the modified additional traces; and
updating the machine learning model based on the first set of original traces, the modified additional traces, and the reconstructed additional traces.
3. The process of claim 2 , further comprising:
determining residual features as unknown trace variations not captured in the basis set of features;
determining supplemental features having a known impact on the first targeted feature; and
updating the machine learning model based on the first set of original traces, the modified additional traces, the reconstructed additional traces, the basis set of features, the residual features and the supplemental features.
4. The process of claim 1 , further comprising:
identifying an effective correlation between the first set of traces and the first predefined targeted feature.
5. The process of claim 1 , further comprising:
identifying an effective correlation between the first set of traces and a plurality of predefined targeted features.
6. The process of claim 1 , the step of generating a basis set of features further comprising:
calculating a set of statistical characteristics of the first set of traces;
modifying the first set of traces to remove the set of statistical characteristics;
calculate an algorithmic solution for the modified first set of traces;
wherein the algorithmic solution and the set of statistical characteristics form the basis set of features.
7. The process of claim 1 , wherein the multiplicity of original traces includes mostly normal traces and the plurality of anomalies represents only a few anomalies.
8. The process of claim 1 , further comprising;
providing a graphical user interface for reviewing a selected trace anomaly;
receiving as first input from the graphical user interface one of a plurality of predefined classes to be associated with the selected trace anomaly; and
receiving as second input from the graphical user interface one of a plurality of predefined actions to be assigned for the selected trace anomaly.
9. A non-transitory computer-readable medium having instructions, which, when executed by a processor cause the processor to:
receive a multiplicity of original traces from respective semiconductor equipment sensors;
identify at least a first key feature for the multiplicity of original traces using the processor-based model;
identify a plurality of anomalies in at least a first set of traces of the multiplicity of original traces;
modify each of the first set of traces on the basis of the first key feature;
generate a basis set of features from each of the modified first set of traces;
create a plurality of reconstructed traces from the modified first set of traces and the basis set of features; and
generate instructions for training a machine learning model to detect and classify anomalies based on the first set of traces, the modified first set of traces, and the reconstructed traces;
the computer-readable medium deployed into a processor that monitors a semiconductor equipment operating environment for identifying and classifying anomalies in trace data.