SNE, the sparse neural engine, is a fully digital spiking neural network accelerator that can handle sparse convolutions efficiently. This repository contains the RTL description of the accelerator, as well as a basic test to simulate the accelerator standalone.
The SNE repository is structured as follows:
rtlcontains the RTL code and Verilog testbench for CUTIEsimcontains the files required to start the simulationvsne_convcontains a high level python model of the SNE acceleratorutilscontains several simulation utility script
This repo uses Bender (https://github.com/pulp-platform/bender) to manage its dependencies and generate compilation scripts. For this reason, the build process of this project will download a current version of the Bender binary.
to fetch all the needed dependencies, execute:
make checkout
To build the RTL platform, execute the following command.
source config.sh
And then run:
make build opt
This will generate the required scripts and build the RTL. Then, to run a basic convolutional regression test, run:
make -C sim all
Currently, Modelsim is the only supported simulation platform.
SNE is released under permissive open source licenses. SNE's source code is released under the Solderpad v0.51 (SHL-0.51) license see LICENSE. The code in vsne_conv is released under the Apache License 2.0 (Apache-2.0) see vsne_conv/LICENSE.
If you find SNE useful in your research, you can cite us:
@INPROCEEDINGS{9774552,
author={Di Mauro, Alfio and Prasad, Arpan Suravi and Huang, Zhikai and Spallanzani, Matteo and Conti, Francesco and Benini, Luca},
booktitle={2022 Design, Automation & Test in Europe Conference & Exhibition (DATE)},
title={SNE: an Energy-Proportional Digital Accelerator for Sparse Event-Based Convolutions},
year={2022},
volume={},
number={},
pages={825-830},
doi={10.23919/DATE54114.2022.9774552}}