Graph Convolutional Network for Netlist Testability Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep learning techniques face challenges in applying to Computer Aided Design (CAD) and Electronic Design Automation (EDA) problems, particularly in representing netlists as directed acyclic graphs, which requires complex preprocessing and loses data structure, making it difficult to train deep neural networks effectively.
Innovation Solution
A pre-processor is developed to levelize netlists, compute controllability and observability, and formulate them as collections of fan-in and fan-out nodes, followed by the application of a graph convolutional network (GCN) to analyze and improve testability by inserting test points, leveraging SCOAP algorithm and node embeddings to enhance training and inference.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If netlists are represented as directed acyclic graphs for deep learning processing, then the circuit connectivity can be analyzed, but data structure is lost and complex preprocessing is required
Solution Approach 1:
The patent introduces an intermediary preprocessing module that transforms netlists into a suitable format for deep learning without losing essential structural information. This intermediary representation serves as a bridge between the original netlist format and the deep learning model requirements, maintaining data integrity while enabling effective processing.
Solution Approach 2:
The patent applies preliminary actions by performing controllability and observability analysis before feeding data into the deep learning model. This preprocessing step prepares the data in advance, computing key metrics that guide the subsequent test point insertion decisions, thereby reducing the complexity during the main inference phase.
2Adaptability or versatility
If conventional software tools are used for CAD and EDA problems, then general purpose solutions are available, but company-specific data and internal knowledge cannot be leveraged
Solution Approach 1:
The patent changes the parameters of the analysis system by training deep learning models on company-specific historical data and internal knowledge. This allows the system to adapt to proprietary design patterns, methodologies, and data characteristics, improving both adaptability to company-specific requirements and the reliability of testability predictions through learned patterns.
3Productivity
If deep learning models are trained with preprocessed netlist data, then faster results can be obtained, but data and structure may be lost during preprocessing
Solution Approach 1:
The patent applies local quality by selectively preserving critical local structures and relationships in the netlist data during preprocessing. Instead of uniform processing, the method maintains essential local connectivity information and structural characteristics where they are most needed for testability analysis, while still enabling fast deep learning processing.
Data Source
AI summary
Techniques to improve the accuracy and speed for detection and remediation of difficult to test nodes in a circuit design netlist. The techniques utilize improved netlist representations, test point insertion, and trained neural networks.