Graph Convolutional Network for Netlist Test Point Insertion
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, where preprocessing requires human expert considerations and results in data loss, making it difficult to identify difficult-to-test nodes.
Innovation Solution
A pre-processor is developed to levelize netlists, compute controllability and observability metrics, and apply a SCOAP algorithm to encode netlist connectivity, followed by a graph convolutional network (GCN) to predict difficult-to-test nodes and insert test points for improved testability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If netlists are preprocessed using traditional methods, then data structure is preserved for processing, but data loss occurs and human expert consideration is required
Solution Approach 1:
The patent replaces traditional mechanical preprocessing methods with a graph convolutional network (GCN) based deep learning approach. The GCN automatically learns to preserve essential netlist characteristics while transforming the data into a format suitable for processing, eliminating manual expert intervention and reducing information loss through automated feature extraction and preservation mechanisms.
2Productivity
If deep learning techniques are applied to netlists, then processing speed is improved, but difficulty in representing netlists as DAGs increases
Solution Approach 1:
The patent segments the netlist representation into a directed acyclic graph (DAG) structure where nodes represent logic blocks and edges represent connections. This segmentation transforms the complex netlist data into a hierarchical structure that deep learning models can process efficiently, maintaining processing speed while managing representation complexity through structured decomposition.
Solution Approach 2:
The patent introduces a graph convolutional network as an intermediary between the netlist input and the deep learning processing pipeline. The GCN serves as a mediator that automatically handles the transformation and representation of netlists as DAGs, reducing the complexity burden on subsequent processing stages while maintaining high processing speed through optimized graph-based operations.
3Reliability
If test points are inserted to improve testability, then observability and controllability are improved, but circuit complexity increases
Solution Approach 1:
The patent applies partial action by selectively inserting test points only at critical locations identified by the deep learning model, rather than uniformly across the entire circuit. The GCN analyzes the netlist and identifies specific nodes where test point insertion will maximally improve observability and controllability, thereby enhancing testability while minimizing the increase in circuit complexity through targeted rather than comprehensive test point placement.
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.