Verification Regression Optimization via Machine Learning Correlation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current verification regression processes are inefficient, often requiring numerous redundant test runs and being sensitive to changes in the verification environment, which leads to poor throughput and inability to effectively achieve verification targets beyond coverage metrics.
Innovation Solution
A method and system that utilize machine learning to analyze data from previous verification regression sessions, identifying correlations between control knobs and metrics to generate control conditions that optimize verification regression sessions, allowing for reduced test runs and improved target achievement with maintained randomness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If numerous test runs are conducted to achieve verification targets, then verification coverage is improved, but the number of redundant test runs increases and throughput decreases
Solution Approach 1:
The system performs preliminary analysis of verification data from previous test runs to identify patterns and correlations between control knobs and verification metrics. This preliminary action enables the system to generate optimized control conditions that predict which test runs are most effective, eliminating the need to conduct all possible test runs and thereby improving throughput while maintaining verification coverage.
Solution Approach 2:
The system continuously collects verification data from test runs and uses machine learning algorithms to analyze correlations between control knobs and metrics. This feedback loop allows the system to refine its understanding of what drives verification coverage and generate increasingly optimized control conditions, reducing redundant test runs over time while maintaining or improving coverage.
2Reliability
If verification regression is conducted with current methods, then verification targets are achieved, but the system becomes sensitive to changes in verification environment
Solution Approach 1:
The system changes the parameters of control conditions based on learned correlations between control knobs and verification metrics. By adjusting these parameters dynamically, the system can adapt to changes in the verification environment while maintaining focus on what drives verification target achievement, thereby reducing sensitivity to environmental changes.
Solution Approach 2:
The system transitions from static verification regresses to dynamic control condition generation. The machine learning model continuously adapts control conditions based on observed correlations, making the verification process flexible and resilient to environmental changes while maintaining reliable achievement of verification targets.
3Measurement precision
If traditional verification regression is used, then coverage metrics are measured, but redundancy in test runs increases and computing resources are wasted
Solution Approach 1:
The system extracts only the essential information needed for verification from the full set of possible test runs. By using machine learning to identify which control knob combinations produce the most valuable verification coverage, the system extracts and executes only those critical test runs, eliminating redundant computing resource consumption while maintaining accurate coverage measurement.
Solution Approach 2:
The verification system serves itself by using its own collected data to generate optimized control conditions. The machine learning model trains on verification data and automatically produces improved control conditions that reduce redundancy, making the system self-optimizing and eliminating the need for external intervention to reduce computing resource waste.
Data Source
AI summary
A method for optimizing a verification regression includes obtaining data, by a processor, of previously executed runs of at least one verification regression session; extracting from the data, by the processor, values of one or a plurality of control knobs and values of one or a plurality verification metrics that were recorded during the execution for each of the previously executed runs of said at least one verification regression; finding, by the processor, correlation between said one or a plurality of the control knobs and each said one or a plurality of verification metrics, and generating a set of one or a plurality of control conditions based on the found correlation; and applying, by the processor, the generated set of one or a plurality of control conditions on the verification environment or on the DUT, or on both, to obtain a new verification regression session.