Power Rail Anomaly Detection for Low-Current IC Faults
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing anomaly detection systems in electronic systems, particularly in ICs and SoCs, struggle to accurately detect low-current anomalies such as micro-SELs and non-radiation-induced faults due to limitations in current monitoring and simplistic machine-learning algorithms, leading to reliability issues.
Innovation Solution
A system-level anomaly detection apparatus and method that senses current and voltage characteristics across multiple power rails and pins in an electronic system, utilizing signal processing and machine learning to identify anomalies by correlating pre-characterized signatures with real-time measurements, and power-cycles affected components to mitigate anomalies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If current monitoring threshold is set low to detect micro-SELs, then detection sensitivity improves, but false detection increases due to normal operating current variations
Solution Approach 1:
The power rail monitoring is segmented into multiple distributed power rails, each monitored separately. This allows the system to detect anomalies in specific regions without being affected by normal current variations in other regions, thereby improving detection sensitivity while reducing false positives.
Solution Approach 2:
The system transitions from monitoring a single power rail to monitoring multiple distributed power rails simultaneously. This dimensional expansion allows correlation analysis across multiple points, enabling differentiation between genuine anomalies and normal operating variations through spatial pattern recognition.
2Measurement precision
If machine learning algorithm complexity is increased to improve anomaly detection accuracy, then detection precision improves, but computational overhead and system complexity increases
Solution Approach 1:
Anomaly signatures are pre-characterized and stored in a database before runtime. During operation, the system performs pattern matching against these pre-computed signatures rather than performing complex real-time analysis, significantly reducing computational overhead while maintaining high detection accuracy.
Solution Approach 2:
Instead of implementing complex machine learning models, the system creates simplified representations (signatures) of anomaly patterns based on pre-collected data. These signature copies enable fast comparison and detection without requiring complex computational algorithms during runtime.
3Reliability
If power cycling is activated frequently to remove detected anomalies, then anomaly mitigation effectiveness improves, but system availability and performance deteriorates
Solution Approach 1:
The system implements feedback control by continuously monitoring power rail characteristics and dynamically adjusting the anomaly response threshold. When anomaly signatures are detected with high confidence, power cycling is activated; when confidence is low, the system continues monitoring without intervention, optimizing the balance between reliability and productivity.
Solution Approach 2:
The system changes the response parameter (power cycling activation) based on the confidence level of anomaly detection. By adjusting the threshold for triggering power cycling based on signature matching confidence, the system ensures that power cycling is performed only when necessary, maintaining system availability while effectively mitigating anomalies.
Data Source
AI summary
An apparatus for detecting an anomaly in an electronic system embodying at least two integrated circuits, and where necessary, removing/mitigating the anomaly. The anomaly detection is based on sensing the characteristics of either the current, the voltage, or both the current and voltage of the supply rail connected to the at least two integrated circuits. When an anomaly occurs, the anomaly is detected by one sensing circuit sensing that the characteristics are different from that when the electronic system is functioning normally.