Time series anomaly detection in integrated circuit (IC) fabrication environment
An IC manufacturing system including an anomaly detection and classification engine including a trained convolutional autoencoder (CAE) module coupled to a pointwise outlier detector (POD) module that is fitted to a latent feature space associated with the trained CAE module. The trained CAE module is operable to encode a candidate time series signal into a compressed representation in the latent feature space. The POD module is operable to determine the number of compressed datapoints of the compressed representation predicted to be outliers, which may be used in classifying whether the candidate time series signal is an anomalous signal.
1 . A method of fabricating an integrated circuit (IC), the method comprising:
configuring a convolutional autoencoder (CAE) engine based on trace data relative to a tool parameter of a fabrication tool deployed at a targeted process step for processing semiconductor wafers, the trace data comprising a plurality of time series signals associated with the tool parameter;
obtaining a candidate time series signal from the fabrication tool comprising datapoints measured relative to the tool parameter during processing of at least one semiconductor wafer by the fabrication tool;
determining, using the CAE engine, one or more channels in a latent feature space corresponding to the candidate time series signal, the one or more channels each having a plurality of compressed datapoints fewer than the datapoints in the candidate time series signal;
on the condition that a quantity associated with the compressed datapoints in the latent feature space determined to be outliers is greater than a first threshold, identifying the candidate time series signal as an anomalous signal;
actuating, responsive to the identifying, an out-of-control action plan (OCAP) module configured to determine a disposition action;
adjusting at least one of a tool setting relative to the tool parameter of the fabrication tool and a process parameter of the targeted process step in response to the disposition action; and
processing a subsequent semiconductor wafer at the targeted process step, the subsequent semiconductor wafer containing the IC at an intermediate stage of formation.
2 . The method as recited in claim 1 , wherein a compressed datapoint is determined to be an outlier based on a pointwise outlier detection (POD) engine fitted to the plurality of time series signals compressed in the latent feature space.
3 . The method as recited in claim 1 , further comprising preprocessing the plurality of time series signals and the candidate time series signal to obtain alignment of the time series signals based on an index before processing by the CAE engine.
4 . The method as recited in claim 1 , further comprising:
configuring a second threshold with respect to the compressed datapoints, the second threshold greater than the first threshold; and
on the condition that the quantity of the compressed datapoints in the latent feature space determined to be outliers is greater than the second threshold, servicing the fabrication tool prior to processing the subsequent semiconductor wafer.
5 . The method as recited in claim 1 , wherein the at least one semiconductor wafer and the subsequent semiconductor wafer are obtained from a single wafer lot.
6 . The method as recited in claim 1 , wherein the at least one semiconductor wafer and the subsequent semiconductor wafer are obtained from different wafer lots.
7 . The method as recited in claim 1 , wherein the disposition action is dependent upon the quantity of the compressed datapoints determined to be outliers and a relative importance of the tool parameter with respect to the targeted process step.
8 . A semiconductor fabrication tool, comprising:
a manufacturing stage configured to receive an integrated circuit at an intermediate stage of formation; and
an anomaly detection and classification (ADC) module coupled to the manufacturing stage, the ADC module including a processor and a persistent memory having executable program instructions configured to perform following acts when executed by the processor:
training a convolutional autoencoder (CAE) engine based on trace data relative to a tool parameter of the semiconductor fabrication tool, the trace data comprising a plurality of time series signals associated with the tool parameter, the CAE engine including a latent feature space corresponding to compressed representation of the trace data;
training a pointwise outlier detection (POD) module fitted to the latent feature space; and
classifying a candidate time series signal operable to be generated by the fabrication tool as an anomalous signal responsive to an outlier determination from the POD module with respect to the candidate time series signal.
9 . The semiconductor fabrication tool as recited in claim 8 , further comprising a data preprocessing module configured to remove translations in the trace data and the candidate time series signal.
10 . The semiconductor fabrication tool as recited in claim 8 , further comprising an input data quality module configured to ensure quality of the trace data and the candidate time series signal.
11 . The semiconductor fabrication tool as recited in claim 8 , wherein the POD module comprises a Local Outlier Factor (LOF) module.
12 . The semiconductor fabrication tool as recited in claim 8 , wherein the POD module comprises an Isolation Forest (IF) module.
13 . The semiconductor fabrication tool as recited in claim 8 , further comprising an alarm generator for triggering an out-of-control action plan (OCAP) module in response to identifying the anomalous signal.
14 . The semiconductor fabrication tool as recited in claim 8 , wherein the program instructions further comprise instructions configured to classify the candidate time series signal at different levels of anomalousness based on a plurality of thresholds.
15 . The semiconductor fabrication tool as recited in claim 8 , wherein the manufacturing stage is configured to facilitate one of etch operations, deposition operations, implant operations, wafer clean operations, thermal treatment operations, and wafer polishing operations.
16 . The semiconductor fabrication tool as recited in claim 8 , further comprising at least one sensor selected from one or more position sensors, pressure sensors, thermal sensors, optical sensors, chemical sensors, motion sensors, level sensors, proximity sensors, and humidity sensors.
17 . The semiconductor fabrication tool as recited in claim 8 , wherein the manufacturing stage includes a process chamber configured to add to or subtract from a material layer of the integrated circuit.
18 . An integrated circuit (IC), comprising:
a material layer formed over a semiconductor wafer at a targeted process step of a fabrication flow using a fabrication tool, the semiconductor wafer forming a substrate for the IC; and
the material layer reworked responsive to determining that a candidate time series signal associated with the fabrication tool is identified as an anomalous signal by a trained convolutional autoencoder (CAE) module coupled to a pointwise outlier detector that is fitted to a latent feature space associated with the trained CAE module, the candidate time series signal comprising trace data from the fabrication tool obtained during processing of the semiconductor wafer.