In-situ IC Anomaly Detection via On-Chip Machine Learning Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques for detecting anomalies in integrated circuits (ICs) are limited by their reliance on pre-calibrated thresholds, which can lead to inaccurate and inefficient detection, and require external systems for processing, resulting in delayed anomaly mitigation and increased computational resource consumption.
Innovation Solution
The integration of sensors and a machine learning model within the IC for real-time anomaly detection, where sensors generate datasets, a processing unit extracts features, and a diagnostics model, trained using machine learning techniques, classifies anomalies and mitigates them in-situ.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If pre-calibrated threshold techniques are used for anomaly detection, then the system is simple to implement, but the detection accuracy is poor and the system is brittle to operating condition changes
Solution Approach 1:
The patent transforms the detection approach from using fixed pre-calibrated thresholds to using dynamic machine learning models that adapt to different operating conditions. The ML models learn optimal detection parameters from training data, enabling accurate anomaly detection across varying conditions without manual recalibration of thresholds.
Solution Approach 2:
The system implements self-service through automated machine learning model training and deployment. The ML models automatically learn from sensor data and improve detection accuracy over time without requiring manual intervention or external calibration, making the system adaptive to changing operating conditions.
2Loss of time
If external systems are used for sensor data processing, then the IC design is simplified, but the anomaly mitigation is delayed and computational resources are consumed externally
Solution Approach 1:
The patent merges the anomaly detection functionality directly into the IC by integrating machine learning models within the chip architecture. This combination of sensors, processing units, and ML models in-situ enables real-time anomaly detection and mitigation without external system dependency, reducing latency and keeping computational resources local.
Solution Approach 2:
The patent introduces machine learning models as intermediaries between sensor data collection and anomaly mitigation actions. These ML models process sensor data locally within the IC, serving as an intelligent mediator that enables fast decision-making and reduces the need for external processing systems.
3Measurement precision
If more features are used for anomaly classification, then the detection accuracy improves, but the computational resources and processing time increase
Solution Approach 1:
The patent applies partial action by selecting and using only the most relevant features for anomaly classification rather than processing all available sensor data. The machine learning models identify and focus on critical features that provide sufficient detection accuracy while minimizing computational resource consumption and processing time.
Data Source
AI summary
An integrated circuit (IC) is provided for in-situ anomaly detection. Sensors in the IC generates sensor datasets including information indicating conditions in the IC. A processing unit in the IC uses a sensor dataset and a model to detect and classify the anomaly. The processing unit may filter the sensor dataset, extract features from the filtered sensor dataset, and input the features into the model. The model outputs one or more classifications of the anomaly. A feature may be a distance vector that represents a difference between a data value in the filtered sensor dataset from a reference data value. The model may be a network of bit-cells in the IC. The model may be continuously trained in-situ, i.e., on the IC. The processing unit may provide the classifications to another processing unit in the IC. The other processing unit may mitigate the anomaly based on the classifications.