In-situ detection of anomalies in integrated circuits using machine learning models

An integrated circuit (IC) for in-situ detection of an anomaly in the IC, the IC comprising:

• one or more sensors configured to generate a sensor dataset including information indicating one or more conditions in the IC;
• a memory configured to store the sensor dataset;
• a preprocessing module configured to receive the sensor dataset and to extract features from the sensor dataset;
• a network of in-memory computing bit cells configured to:
• receive the extracted features, and
• determine one or more classifications of the anomaly in the IC using a first subset of the features,
• wherein the network has been trained to detect and classify anomalies in integrated circuits;
• and
• a training module configured to:
• identify a second subset of the features, the second subset of features including one or more features that are not in the first subset;
• and
• further train the network by using the one or more features in the second subset and one or more labels of the one or more features.

One or more non-transitory computer-readable media storing instructions executable to perform operations for in-situ detection of an anomaly in an integrated circuit (IC), the operations comprising:

• receiving, by a processing unit in the IC, a sensor dataset generated by one or more sensors in the IC, the sensor dataset including information reflecting one or more conditions of the IC;
• extracting, by the processing unit in the IC, features from the sensor dataset;
• inputting, by the processing unit in the IC, the features into a model in the IC, wherein the model has been trained to detect and classify anomalies in integrated circuits;
• outputting, by the model, one or more classifications of the anomaly in the IC, wherein the model determines the one or more classifications of the anomaly based on a first subset of the features;
• identifying a second subset of the features, the second subset of features including one or more features that are not in the first subset;
• and
• further training the model by using the one or more features in the second subset and one or more labels of the one or more features.