DepthGraphNet: Circuit Graph Isomorphism Detection via Siamese-Graph Neural Networks
DepthGraphNet: Circuit Graph Isomorphism Detection via Siamese-Graph Neural Networks
复制标题
DOI：
10.1109/mlcad58807.2023.10299839
复制
发表时间：
2023-09
期刊：
影响因子：
--
通讯作者：
Fin Amin;Soumyadeep Chatterjee;P. Franzon
中科院分区：
文献类型：
--
作者：
Fin Amin;Soumyadeep Chatterjee;P. Franzon
电路图同构（CGI）问题是EDA中的一个基本计算问题，涉及到确定两个电路图的等价性。这个问题在电路的设计流程中具有重要意义，并且与知识产权（IP）保护工作一样重要-解决这个问题将对设计过程产生重大影响，从而实现更快、更高效的设计流程。在这项工作中，我们介绍了DepthGraphNet，一个用于CGI问题的连体图神经网络（SGNN）。此外，我们表明，我们提出的方法运行速度大大快于经典的方法，同时保持高精度的真实世界的电路数据集。此外，我们提供的定理，这有助于处理耗时的SGNN架构设计过程。
The circuit graph isomorphism (CGI) problem is a fundamental computational problem in EDA that involves determining the equivalence of two circuit graphs. This issue carries great significance in the design flow of circuits, as well as with intellectual property (IP) protection efforts–a solution to this issue would have significant implications for the design process, enabling a faster and more efficient design flow. In this work, we introduce DepthGraphNet, a Siamese-Graph Neural Network (SGNN) for the CGI problem. Additionally, we show that our proposed approach runs considerably faster than classical approaches while maintaining high accuracy on a real-world circuit dataset. Furthermore, we provide theorems which help deal with the time-consuming SGNN architecture design process.