by Lillian Pentecost and Alexander Hankin, 2021
This is a beta version of a broad design space exploration framework for evaluating and comparing different on-chip memory solutions (including embeddable non-volatile memory technologies such as RRAM, PCM, STT, FeFET, and CTT devices) with system-level constraints and application-level impacts in-the-loop.
Users can configure experiments with customized cell-level parameters, specific memory array design priorities and constraints, and analytically evaluate the performance and efficiency of memory solutions for specific application traffic patterns. Documentation is a work-in-progress, but templates for various experiments, example cell confiugrations, and example application traffic are provided with sample configurations in the config
directory. Additionally, fault injection experiments for different NVM configurations and fault models can be developed and run within nvmexplorer_src/nvmFI
, and surveyed cell-level parameters for various technologies are provided in output/NVM_data
and leveraged in example studies (see sample configs for details).
Please see https://nvmexplorer.seas.harvard.edu for additional documentation and details, and get started using the instructions below.
For example studies or to cite this work, please see https://arxiv.org/abs/2109.01188
NVMExplorer relies on an extended and modified version of NVSim
Be sure to clone this repository with ``git clone --recurse-submodules https://github.com/lpentecost/NVMExplorer''
After cloning, compile nvsim_src:
cd nvmexplorer_src/nvsim_src
make
Prior to running NVMExplorer, please verify you are using Python 3.X and have the following packages available:
- pandas
- numpy
Please see additional documentation for tutorials, fault injection setup, and more information
python run.py config/[config name].json
http://www.nvmexplorer.seas.harvard.edu
Questions, comments, or feature requests? Please reach out directly:
If you use or expand on this work, please cite NVMExplorer [https://arxiv.org/abs/2109.01188]