Machine Learning suffers from the curse of dimensionality, and requires significant computing power to process data. Tunnel diodes are highly efficient circuit components which utilise a quantum effect known as tunnelling. Here, we propose using the current–voltage (I–V) characteristic curve of a tunnel diode as a novel, physically grounded activation function for neural networks. The tunnel-diode activation function (TDAF) outperforms traditional activation functions like ReLU and Sigmoid in terms of accuracy and loss during both training and evaluation. We argue that both the increased non-linearity and the quantum tunnelling region of TDAF improve the propagation of data through the network. The potential to implement such neural networks as efficient circuits composed of tunnel diodes could pave the way for neuromorphic, quantum-inspired AI systems.
This repository accompanies our paper "Neuromorphic Quantum Neural Networks with Tunnel-Diode Activation Functions" available on ArXiv: https://www.arxiv.org/abs/2503.04978. If you utilise this code please cite
@misc{mcnaughton2025neuromorphicquantumneuralnetworks,
title={Neuromorphic Quantum Neural Networks with Tunnel-Diode Activation Functions},
author={Jake McNaughton and A. H. Abbas and Ivan S. Maksymov},
year={2025},
eprint={2503.04978},
archivePrefix={arXiv},
primaryClass={physics.app-ph},
url={https://arxiv.org/abs/2503.04978},
}