FAME

FAME is a fast and low memory cost simulation framework for non-ideal ReRAM-based DNN accelerator.

Dependencies

• Python >= 3.8
• PyTorch >= 2.0.1
• Matplotlib

Usage

1. Set your dataset root path and result saving path in config.ini.
2. Simply run the following command:

   $ python mnist.py

   to execute a noise injection training sample for LeNet-5 on FAME.