[论文解读] CktGNN: Circuit Graph Neural Network for Electronic Design Automation

Zehao Dong, Weidong Cao+4 more|arXiv (Cornell University)|Aug 31, 2023

Ferroelectric and Negative Capacitance Devices被引用 12

一句话总结

CktGNN 引入了一个具有预设子图基的两级 GNN，联合优化电路拓扑和器件尺寸，在 Open Circuit Benchmark (OCB) 上得到验证。

ABSTRACT

The electronic design automation of analog circuits has been a longstanding challenge in the integrated circuit field due to the huge design space and complex design trade-offs among circuit specifications. In the past decades, intensive research efforts have mostly been paid to automate the transistor sizing with a given circuit topology. By recognizing the graph nature of circuits, this paper presents a Circuit Graph Neural Network (CktGNN) that simultaneously automates the circuit topology generation and device sizing based on the encoder-dependent optimization subroutines. Particularly, CktGNN encodes circuit graphs using a two-level GNN framework (of nested GNN) where circuits are represented as combinations of subgraphs in a known subgraph basis. In this way, it significantly improves design efficiency by reducing the number of subgraphs to perform message passing. Nonetheless, another critical roadblock to advancing learning-assisted circuit design automation is a lack of public benchmarks to perform canonical assessment and reproducible research. To tackle the challenge, we introduce Open Circuit Benchmark (OCB), an open-sourced dataset that contains $10$K distinct operational amplifiers with carefully-extracted circuit specifications. OCB is also equipped with communicative circuit generation and evaluation capabilities such that it can help to generalize CktGNN to design various analog circuits by producing corresponding datasets. Experiments on OCB show the extraordinary advantages of CktGNN through representation-based optimization frameworks over other recent powerful GNN baselines and human experts' manual designs. Our work paves the way toward a learning-based open-sourced design automation for analog circuits. Our source code is available at \url{https://github.com/zehao-dong/CktGNN}.

研究动机与目标

• 自动化模拟电路设计通过联合优化拓扑和器件参数。
• 提出一个两级 GNN（CktGNN），通过预定义有序子图基编码电路以降低拓扑搜索空间。
• 创建并发布 Open Circuit Benchmark (OCB)，以实现仿真可重复的模拟电路设计自动化基准。
• 证明 CktGNN 在拓扑重构、预测准确性和真实世界电路生成方面优于基线。

提出的方法

• 将电路表示为有向无环图（DAG），并从预设计的有序子图基 B 中分解为不重叠的子图。
• 对每个子图独立应用内部 GNN 学习表示（捕获上下文信息）。
• 在转换后的 DAG 上进行外部有向消息传递，节点嵌入结合子图类型和内部子图表示。
• 使用门控求和聚合并在外部 GNN 使用基于 GRU 的更新来获得电路级嵌入。
• 将编码框架置于类似 VAE 的学习设定，以支持电路生成的拓扑和参数优化。

实验结果

研究问题

• RQ1 CktGNN 是否能比现有 DAG 编码器更有效地联合优化电路拓扑和器件尺寸？
• RQ2 具有子图基的两级 GNN 是否提升表达能力和拓扑重构相对于基线？
• RQ3 CktGNN 在电路性质的预测精度和生成有效、全新拓扑方面的表现如何？
• RQ4 Open Circuit Benchmark (OCB) 是否实现可重复基准测试并加速现实世界电路设计自动化？

主要发现

增益 RMSE	增益皮尔逊相关	带宽 RMSE	带宽皮尔逊相关	相位裕度 RMSE	相位裕度 Pearson	FoM RMSE	FoM Pearson	重构准确度
CktGNN	0.607 ± 0.003	0.791 ± 0.002	0.873 ± 0.003	0.479 ± 0.001	0.973 ± 0.002	0.217 ± 0.001	0.854 ± 0.003	0.491 ± 0.002	0.397
PACE	0.644 ± 0.003	0.762 ± 0.002	0.896 ± 0.003	0.442 ± 0.001	0.970 ± 0.003	0.226 ± 0.001	0.889 ± 0.003	0.423 ± 0.001	0.306
DAGNN	0.695 ± 0.002	0.707 ± 0.001	0.881 ± 0.002	0.453 ± 0.001	0.969 ± 0.003	0.231 ± 0.002	0.877 ± 0.003	0.442 ± 0.001	0.289
D-VAE	0.681 ± 0.003	0.739 ± 0.001	0.914 ± 0.002	0.394 ± 0.001	0.956 ± 0.003	0.301 ± 0.002	0.897 ± 0.003	0.374 ± 0.001	0.271
GCN	0.976 ± 0.003	0.140 ± 0.002	0.970 ± 0.003	0.236 ± 0.001	0.993 ± 0.002	0.171 ± 0.001	0.974 ± 0.003	0.217 ± 0.001	0.058
GIN	0.890 ± 0.003	0.352 ± 0.001	0.926 ± 0.002	0.251 ± 0.001	0.985 ± 0.004	0.187 ± 0.002	0.910 ± 0.003	0.284 ± 0.001	0.051
NGNN	0.882 ± 0.004	0.433 ± 0.002	0.933 ± 0.003	0.247 ± 0.001	0.984 ± 0.004	0.196 ± 0.002	0.926 ± 0.002	0.267 ± 0.001	0.068
Pathformer	0.816 ± 0.003	0.529 ± 0.001	0.895 ± 0.002	0.410 ± 0.001	0.967 ± 0.002	0.297 ± 0.001	0.887 ± 0.002	0.391 ± 0.001	0.081

• CktGNN 在 OCB 数据集上的 Gain、BW、PM、FoM 等指标达到最前沿预测性能和拓扑重构准确性。
• CktGNN 编码由于并行的内部 GNN 和高效的外部 DAG 处理，训练和推理速度快于基于 GRU 的编码器。
• 在真实世界的电路设计任务中，CktGNN 解码器生成更多有效 DAG 和电路，并在贝叶斯优化中达到有竞争力的 FoM。
• CktGNN 在拓扑重构和预测指标上优于若干基线（PACE、DAGNN、D-VAE、GCN、GIN、NGNN、Graphormer）。
• OCB 提供一个可扩展的开放基准，用于模拟电路拓扑生成及参数优化，促进可重复研究。