Analog Circuit Anomaly Detection via Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing complexity of automotive electronic systems and the lack of a widely accepted analog fault model pose challenges in achieving high Automotive Safety Integrity Levels (ASIL) for analog circuits, particularly in detecting parametric faults and environmental stress-induced failures, which can lead to catastrophic system failures.
Innovation Solution
A machine learning-based anomaly detection system that utilizes dynamic in-field time-series data to predict imminent failures in analog circuits, leveraging design-for-x features and federated learning to improve functional safety with minimal hardware overhead, enabling early detection and proactive action.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional fault detection methods are used in analog circuits, then the system structure remains simple, but the ability to detect parametric faults and environmental stress-induced failures is insufficient
Solution Approach 1:
The patent implements preliminary action by continuously monitoring analog circuits for anomalies before actual failures occur. The anomaly detection system identifies degradation trends and predicts potential failures in advance, allowing proactive maintenance actions to be taken before the circuit fails, thus improving functional safety without requiring complex real-time intervention systems
Solution Approach 2:
The patent uses machine learning models as intermediaries between raw circuit data and fault detection decisions. These models process and interpret complex analog signal patterns, transforming difficult-to-analyze raw data into actionable anomaly predictions, thereby enhancing detection capability while managing system complexity through intelligent abstraction
2Measurement precision
If more monitoring and detection mechanisms are added to improve fault detection capability, then the detection accuracy improves, but the hardware overhead increases
Solution Approach 1:
The patent applies self-service by implementing built-in self-test (BIST) mechanisms within the analog circuits themselves. The circuits monitor their own operational parameters and generate test data internally without requiring external test equipment, enabling accurate anomaly detection while minimizing additional hardware overhead through resource-efficient self-diagnosis
Solution Approach 2:
The patent achieves universality by designing monitoring mechanisms that serve multiple functions: they simultaneously perform normal circuit operation, collect operational data for training, execute anomaly detection, and support predictive maintenance. This multi-functionality allows high detection accuracy to be achieved without proportionally increasing hardware complexity, as the same hardware resources serve multiple purposes
3Reliability
If corrective actions are taken only after failure occurs, then the response time is minimized, but the safety mechanism fails to prevent catastrophic results
Solution Approach 1:
The patent implements preliminary action by detecting anomalies and predicting failures before they actually occur. The system identifies degradation patterns and issues warnings in advance, allowing maintenance actions to be scheduled and executed before catastrophic failure, thereby improving safety effectiveness while actually reducing total fault handling time through proactive rather than reactive management
Data Source
AI summary
In some examples, systems and methods may be used to improve functional safety of analog or mixed-signal circuits, and, more specifically, to anomaly detection to help predict failures for mitigating catastrophic results of circuit failures. An example may include using a machine learning model trained to identify point anomalies, contextual or conditional anomalies, or collective anomalies in a set of time-series data collected from in-field detectors of the circuit. The machine learning models may be trained with data that has only normal data or has some anomalous data included in the data set. In an example, the data may include functional or design-for-feature (DFx) signal data received from an in-field detector on an analog component. A functional safety action may be triggered based on analysis of the functional or DFx signal data.