Trainable Neuromorphic Spintronic Hardware Via Analog Finite-Difference Gradient Methods
Authors/Creators
- 1. Politecnico di Bari
-
2.
International Iberian Nanotechnology Laboratory
-
3.
Istituto Nazionale di Geofisica e Vulcanologia
-
4.
Polytechnic University of Bari
-
5.
University of Messina
- 6. University of Porto
Description
This repository contains the experimental data and simulation code used in the study “Trainable Neuromorphic Spintronic Hardware Via Analog Finite-Difference Gradient Methods”.
The repository is organized into two main components: datasets (experimental measurements), simulation code (neural network training and evaluation).
The dataset includes measured current–voltage (I–V) characteristics and corresponding gradient estimations for different device configurations (Stack A and Stack B), each containing measurements for distinct magnetic states.
The code enables full reproduction of the results presented in the manuscript, including model training on IRIS and MNIST datasets, as well as knowledge distillation experiments using experimentally derived activation functions and respective gradients.
Files
code.zip
Files
(51.3 MB)
| Name | Size | Download all |
|---|---|---|
|
md5:35a92885806e9115a7c4b84ba2c98a62
|
51.2 MB | Preview Download |
|
md5:6ae6d76c27372d5171966edba69beaa0
|
123.9 kB | Preview Download |
Additional details
Software
- Programming language
- Python