Machine Learning Debugging for Electronic Design Verification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current post-process debugging methods for electronic design verification struggle to efficiently identify the root cause of failures in simulation runs on a design under test (DUT), as they rely on manual analysis of recorded execution events, which can be time-consuming and prone to errors.
Innovation Solution
A method using a processor to analyze recorded simulation runs, construct trees of causes for suspected errors, perform driver tracing, and train a machine learning algorithm to distinguish between failed and successfully passed runs, extracting significant features that predict simulation outcomes and output them for user analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual analysis of recorded execution events is used for debugging, then comprehensive review of all execution events is possible, but debugging time and effort increase significantly
Solution Approach 1:
A machine learning classifier is introduced as an intermediary between the recorded execution events and the debugger. The classifier automatically analyzes execution events and identifies potential failure causes, presenting only relevant information to the user. This mediator filters through the comprehensive data, maintaining debugging accuracy while dramatically reducing the time required for analysis.
Solution Approach 2:
The debugging system performs self-service by automatically analyzing execution events and generating failure cause predictions without requiring manual review of all events. The machine learning model autonomously processes the recorded data, identifies patterns, and highlights probable failure causes, allowing the system to serve its own debugging needs efficiently.
2Loss of information
If all execution events are recorded and saved for analysis, then complete information is available for debugging, but data storage and processing requirements increase
Solution Approach 1:
The system extracts only the most relevant features and execution events that contribute to failure causes, rather than processing all recorded data uniformly. The machine learning model identifies and extracts key signals and events that are most indicative of failures, separating essential information from redundant data, thus reducing processing requirements while maintaining diagnostic accuracy.
3Difficulty of detecting and measuring
If traditional debugging tools are used to review execution events, then detailed analysis is possible, but automation level remains low
Solution Approach 1:
The manual mechanical process of reviewing execution events is replaced with an automated machine learning-based system. Instead of manually examining recorded events, the system uses trained classifiers to automatically detect failure causes by analyzing execution patterns. This substitution maintains detailed analysis capability while dramatically increasing automation, transforming debugging from a manual task to an automated process.
Data Source
AI summary
A method for identifying a potential cause of a failure in simulation runs on a design under test (DUT) using machine learning is disclosed.