thnguyen996/fault-injection

Implementation of "Low-Cost and Effective Fault-Tolerance Enhancement Techniques for Emerging Memories-Based Deep Neural Networks." 2021 58th ACM/IEEE Design Automation Conference (DAC).

fault-injection

This code simulates the impact of emerging memory's stuck-at faults (SAFs) on deep neural networks. It also provides different SAF-tolerance methods to increase the robustness of DNNs. Details of the proposed methods are described in: Thai-Hoang Nguyen, Muhammad Imran, Jaehyuk Choi, Joon-Sung Yang (2021). Low-Cost and Effective Fault-Tolerance Enhancement Techniques for Emerging Memories-Based Deep Neural Networks. 2021 58th ACM/IEEE Design Automation Conference (DAC).

Requirements

1. Python3.6
2. Pytorch 1.5.1
3. Cupy 7.8.0: (similar to numpy but runs on gpus)
4. torchsummary (sumarize model)
5. tensorboard (to display the results)
6. Pyinstrument (to profile the code)

USAGE EXAMPLES

1. Clone the repo git clone https://github.com/thnguyen996/fault-injection.git
2. Download pretrained weights and put it in checkpoint/ folder
3. Run the code python main.py --method method0 to simulate stuck-at faults injection

Noise injection simulation

$ python main.py --method method0

Weight mapping and encoding (Proposed method)

$ python main.py --method method2

Error correction pointer ECP

$ python main.py --method ECP 

Research papers citing

If you use this code for your research paper, please use the following citation:

@inproceedings{nguyen_imran2021,
 author = {Thai-Hoang Nguyen and Muhammad Imran and Jaehyuk Choi and Joon-Sung Yang},
 booktitle = {2021 58th ACM/IEEE Design Automation Conference (DAC)},
 doi = {10.1109/dac18074.2021.9586112},
 month = {dec},
 publisher = {IEEE},
 title = {Low-Cost and Effective Fault-Tolerance Enhancement Techniques for Emerging Memories-Based Deep Neural Networks},
 url = {https://doi.org/10.1109%2Fdac18074.2021.9586112},
 year = {2021}
}