Hardware Description Graph Attention for IC Metric Feedback
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing semiconductor integrated circuit design methods lack a specific method for providing feedback on the design, making it difficult to estimate which statements in a hardware description require improvement to enhance the quality metric.
Innovation Solution
A semiconductor integrated circuit design assistance method that converts a hardware description into a graph object using a control data flow graph (CDFG) format and employs a neural network represented by a graph attention network (GAT) to infer a physical metric, identifying key statements influencing the metric and extracting an attention part for feedback.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If a quality metric is inferred at the initial stage without performing all design stages, then the time required for design is shortened, but it becomes difficult to provide specific feedback on which statements in the hardware description need improvement
Solution Approach 1:
The hardware description is segmented into multiple statements, each represented as a node in a control data flow graph. This segmentation allows the system to analyze individual statements independently and provide targeted feedback on specific portions of the hardware description that need improvement.
Solution Approach 2:
A control data flow graph is introduced as an intermediary representation between the hardware description and the quality metric inference. This graph structure enables the system to process the hardware description into a format suitable for neural network analysis while preserving the relationship between individual statements and the overall quality metric.
2Manufacturing precision
If all stages of the design are performed to obtain accurate physical metrics, then the manufacturing precision is improved, but the productivity is reduced due to the time-consuming process
Solution Approach 1:
The system performs preliminary conversion of the hardware description into a control data flow graph and extraction of key statements before the full design process is completed. This preliminary analysis enables accurate physical metric estimation to be obtained early in the design process, allowing for iterative improvements without requiring completion of all design stages.
Solution Approach 2:
The patent replaces traditional mechanical design iteration processes with a neural network-based inference system. The neural network models the physical metrics by learning from training data, substituting the need for actual physical simulation or measurement with computational inference that achieves comparable accuracy much faster.
Data Source
AI summary
A semiconductor integrated circuit design assistance method includes: converting a hardware description into a graph object; inferring an inferred physical metric of a semiconductor integrated circuit by inputting the graph object into a neural network trained in a GAT method; and extracting an attention part of the hardware description based on the inferred physical metric. The graph object includes a node and an edge. The node includes a first node including description position information that indicates a position in the hardware description. The neural network includes, as a weight, an attention coefficient that indicates a degree of influence that the edge has on the inferred physical metric. In the extracting, a first edge is identified based on the attention coefficient, and a part of the hardware description is extracted as the attention part based on the description position information of the first node that corresponds to the first edge.