Neuromorphic computing Enabled by Heavily doped semiconductor Optics

Introduction

NEHO will develop a novel photonic integrated circuit platform that enables ultrafast and low-energy consumption neuromorphic information processes by means of a newly developed nonlinear photon-plasmon semiconductor technology at mid-infrared wavelengths (8-12 µm).

Vision

NEHO's vision will be achieved by the unconventional use of semiconductors to optimize and control plasmonic effects that will provide the optical nonlinearity required to implement the functionalities of an artificial neuron. NEHO's optical neuron will be the building block for the realization of ultrafast optical neural networks.

Technology Integration

We will combine the flexibility of field-effect devices realized on semiconductors with the nanoscale nature of plasmonic processes to enable the reconfigurability of the nonlinear optical coefficient at each node of the network. This will be simply obtained by controlling DC electric potential levels.

Core Concept

At the heart of NEHO is the idea of exploiting the rich electron dynamics of semiconductors. Doped semiconductors undergo an interesting transition from the size-quantization regime to the classical regime of plasmon oscillations.

Nonlocal and Nonlinear Optical Response

This transition region can exhibit strong nonlocal and nonlinear optical response due to a large variety of electron-electron interactions. The decrease in electron density induced on the semiconductor surface by an external bias can be used to modulate the nonlinear response strength.

Applications

This unprecedented feature will be used to leverage the hardware implementation of a neural network into the development of new machine learning optimization techniques, including:

1. The optimization of the nonlinear activation function to different tasks.
2. Enhancing the performance of various machine learning applications.

This extra degree of freedom will offer tremendous benefits for a large variety of machine learning applications.