Soda-PTA is implemented based on PyTorch and DGL, designed for online hyperparameter selection in the pseudo-transient analysis (PTA) algorithm during DC analysis in SPICE circuit simulation. Experimental results indicate that Soda-PTA consistently enhances the performance of PTA-related algorithms. For detailed design information, please refer to the reference[1].
Soda-PTA_noGNN Soda-PTA operates without a GNN optimization framework.
Soda-PTA_withGNN Soda-PTA operates with a GNN optimization framework.
- cudatoolkit=11.6.0
- dgl-cuda11.6=0.9.1
- numpy=1.23.0
- python=3.9.13
- pytorch=1.12.0=py3.9_cuda11.6_cudnn8_0- Jiatai Sun, Xiaru Zha, Chao Wang, Xiao Wu, Dan Niu, Wei W. Xing and Zhou Jin. "Pseudo Adjoint Optimization: Harnessing the Solution Curve for SPICE Acceleration", ACM/IEEE International Conference on Computer-Aided Design (ICCAD), Oct 27-31, 2024.
If you have any questions about running the code, please contact Jiatai Sun.
E-mail: [email protected]