Machine Learning Classifiers for Risky Circuit Pattern Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Modern integrated circuit design and testing face challenges in identifying unreliable circuit patterns due to the complexity and computational limitations of existing simulation methods, leading to potential reliability issues that are only discovered post-fabrication.
Innovation Solution
A machine learning-based classifier is trained to recognize unreliable circuit patterns without full simulations, using a database of previously identified patterns, enabling rapid identification and update with new patterns.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If circuit simulations are performed to identify unreliable circuit patterns, then detection accuracy is improved, but processing time increases significantly
Solution Approach 1:
The system performs preliminary actions by pre-computing and storing circuit patterns that are prone to reliability problems in a database before actual design verification. When analyzing a circuit netlist, the system retrieves and compares against these pre-stored patterns rather than performing full simulations, thus achieving fast detection without sacrificing accuracy.
Solution Approach 2:
The system creates simplified copies of complex circuit behaviors by storing representative circuit patterns (topologies, device configurations, connection structures) that exhibit reliability issues. These pattern copies serve as proxies for full circuit simulations, enabling rapid comparison and identification of problematic circuits without repeating expensive simulation processes.
2Reliability
If full circuit simulations are performed on large circuits with over 1 million devices, then comprehensive reliability assessment is achieved, but computational resources required increase excessively
Solution Approach 1:
The system extracts and isolates specific circuit patterns from large-scale circuits that are known to cause reliability problems. Instead of simulating the entire 1-million-device circuit, the system extracts relevant pattern instances (specific topologies, device arrangements, connection structures) and evaluates them independently against stored reliability patterns, dramatically reducing computational complexity.
Solution Approach 2:
The system segments the large circuit into identifiable pattern instances by comparing the circuit netlist against a database of stored circuit patterns. Each matched pattern represents a segment that can be evaluated independently for reliability issues, avoiding the need to process the entire complex circuit as a single unit.
3Measurement precision
If circuit patterns are identified using traditional simulation methods, then reliability issues are detected, but the ability to simulate all input vector scenarios is lost due to time constraints
Solution Approach 1:
The stored circuit patterns in the database serve multiple functions: they represent various unreliable circuit topologies, device configurations, and connection structures that can match different instances in any circuit netlist. This universal pattern library enables the system to detect multiple types of reliability issues (electrostatic discharge, electrical overstress, aging, etc.) across diverse circuit scenarios without requiring separate simulations for each case.
Data Source
AI summary
Machine assisted systems and methods for detecting unreliable circuit patterns are described. These systems and methods can use a machine learning classifier, that has been trained to recognize such circuit patterns, to detect the unreliable circuit patterns without requiring computationally expensive simulations of a circuit netlist which can be over a million devices (e.g. over a million FETs). The classifier, once trained, can recognize unreliable circuit patterns quickly and can be updated over time as new unreliable circuit patterns are discovered from simulations or other sources.