Welcome to the CrossNAS framework!
This is the GitHub repository for the paper CrossNAS: A Cross-Layer Neural Architecture Search Framework for PIM Systems.
The CrossNAS framework has two segments.
- NN_Architecture_Search: Performs weight-sharing based supernet training and subnet search for the neural network architecture
- Mixed_Precision_Quantization_and_PIM_Search: Finds the efficient mixed-precision quantization map and PIM configuration parameters for the searched NN architecture
You need to pull the MNSIM-2.0 PIM simulation framework in both of the directories before initiating simulations.
Please cite the following paper if you use the framework. Md Hasibul Amin, Mohammadreza Mohammadi, Jason D. Bakos, and Ramtin Zand. 2025. CrossNAS: A Cross-Layer Neural Architecture Search Framework for PIM Systems. In Proceedings of the Great Lakes Symposium on VLSI 2025 (GLSVLSI '25). Association for Computing Machinery, New York, NY, USA, 334–340. https://doi.org/10.1145/3716368.3735178