Protocol State Graph Neural Network for IC Verification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for designing integrated circuits are time-consuming and prone to human error in generating test sequences for verification, often failing to cover all corner cases due to manual or random generation processes.
Innovation Solution
A machine learning-based approach for sequence generation that uses protocol state graph neural networks to explore and identify different states, allowing for the automatic creation of test sequences by training models to transition between states and generate sequences based on these representations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual or random test sequence generation is used, then human control and flexibility are maintained, but verification time increases and corner cases may be missed
Solution Approach 1:
The system enables self-service by allowing the neural network to automatically generate test sequences without human intervention. The neural network learns from protocol state graphs and independently produces verification sequences, eliminating the need for manual sequence generation while ensuring comprehensive corner case coverage through its ability to explore state spaces systematically.
Solution Approach 2:
The patent replaces the mechanical human-controlled process of manual test sequence generation with an intelligent neural network system. The neural network processes protocol state graphs and automatically generates verification sequences, substituting human cognitive processes with automated machine learning that can systematically explore all possible states and transitions without fatigue or oversight.
2Reliability
If manual test sequence generation is used, then human expertise can be applied, but human error and omissions occur
Solution Approach 1:
The neural network performs self-service by automatically learning from protocol specifications and generating accurate test sequences without human intervention. This eliminates human error while maintaining the complexity management through automated processes that systematically handle protocol analysis and sequence generation without requiring human expertise for each individual task.
3Productivity
If automated neural network sequence generation is used, then efficiency and completeness improve, but system complexity increases
Solution Approach 1:
The system applies preliminary action by pre-training the neural network on protocol state graphs before actual test sequence generation. This preliminary training phase enables the network to learn the protocol structure and state transitions in advance, making the actual sequence generation process efficient and automated while managing complexity through structured pre-processing of protocol specifications.
Solution Approach 2:
The patent uses protocol state graphs as an intermediary representation between the neural network and the actual test sequences. The state graphs serve as a structured intermediate format that captures protocol behavior, enabling the neural network to learn from a standardized representation and generate sequences efficiently without directly processing complex protocol specifications.
4Adaptability or versatility
If random test sequence generation is used, then simplicity is maintained, but coverage of all states is not achieved
Solution Approach 1:
The patent replaces random sequence generation with a neural network-based systematic exploration method. The neural network learns from protocol state graphs and generates sequences that systematically cover all states and transitions, substituting random sampling with intelligent exploration that guarantees comprehensive coverage while reducing the time needed to achieve complete state verification.
Data Source
AI summary
The approach disclosed herein is a new approach to sequence generation in the context of validation that relies on machine learning to explore and identify ways to achieve different states. In particular, the approach uses machine learning models to identify different states and ways to transition from one state to another. Actions are selected by machine learning models as they are being trained using reinforcement learning. This online inference also is likely to result in the discovery of not yet discovered states. Each state that has been identified is then used as a target to train a respective machine learning model. As part of this process a representation of all the states and actions or sequences of actions executed to reach those states is created. This representation, the respective machine learning models, or a combination thereof can then be used to generate different test sequences.