Reinforcement Learning Agent for Digital Circuit Verification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current digital circuit design verification processes are complex and lack absolute evaluation criteria, leading to inefficiencies in ensuring quality and performance due to reliance on rule-based scenarios and the limitations of machine learning algorithms in processing large verification data sets.
Innovation Solution
A system verification method using reinforcement learning that generates and updates verification vectors based on agent actions, training the agent with a data set of actions and their coverages to improve coverage and efficiency, and providing rewards for wider state coverage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If machine learning algorithms are used for verification, then automation is improved, but processing capability deteriorates due to exceeding computation limits and large verification data set sizes
Solution Approach 1:
The verification process is segmented into multiple iterations where the verification data set is divided into training data sets and test data sets. The agent performs verification actions on portions of the data set in sequential episodes, rather than processing the entire large data set at once, thus reducing the computational burden on the machine learning algorithm.
Solution Approach 2:
The agent is trained in advance on a training data set before being evaluated on the full verification data set. This preliminary training phase allows the agent to learn verification strategies on a manageable subset of data, improving its capability to handle the complete verification task without overwhelming computational resources.
2Ease of operation
If rule based scenario verification is used, then ease of operation is maintained, but productivity deteriorates due to lack of absolute evaluation criteria and need for post-correction testing
Solution Approach 1:
The system implements feedback through coverage metrics that measure the effectiveness of verification actions. The agent receives feedback on its verification performance through the coverage calculation, which quantifies the proportion of the verification data set that has been effectively tested. This feedback mechanism enables automated evaluation and guides the agent to improve verification efficiency without requiring manual post-correction.
3Reliability
If verification coverage is expanded to ensure quality, then reliability is improved, but loss of time increases due to comprehensive testing requirements
Solution Approach 1:
The verification process is made dynamic through iterative episodes where the agent learns and adapts its verification strategy. The coverage metric dynamically updates as the agent progresses through the verification data set, allowing the system to balance thoroughness with efficiency. The agent can adjust its exploration behavior based on learned patterns, achieving high coverage without linearly increasing verification time.
Data Source
AI summary
A system verification method includes generating a first verification vector as a result of a first action of an agent, the first verification vector referring to an observation corresponding to at least one state already covered, from among states of elements of a target system, identifying a first coverage corresponding to at least one state covered by the first verification vector, from among the states of the elements, updating the observation by reflecting the first coverage in the observation, and generating a second verification vector through a second action of the agent, the second verification vector referring to the updated observation.