Perovskite Spiking Neurons for Intelligent Networks

Introduction

A brain is a complex structure where computing and memory are tightly intertwined at a very low power cost of operation, by analog signals across vast quantities of synapse-connected spiking neurons. Animal brains react intelligently to environmental events and perceptions.

Objective

By developing similar Spiking Neural Networks (SNN), we can realize neuromorphic computation systems that are excellent for dealing with large amounts of noisy data and stimuli, and are very well suited for perception, cognition, and motor tasks. However, the current CMOS technologies perform very poorly for emulating biological brains, and their power consumption is large.

Challenges

Currently, we cannot replicate biological neuron behaviors with existing design and manufacturing technology. This project aims to develop compact miniature material elements that will closely emulate the complex dynamic behavior of neurons and synapses, to form SNNs with substantial reductions in footprint, complexity, and energy cost for perception, learning, and computation.

Materials and Methods

We investigate the properties of metal halide perovskite, which have produced excellent photovoltaic devices in the last decade. These perovskites exhibit:

• Ionic/electronic conduction
• Hysteresis
• Memory effect
• Switchable and nonlinear behavior

These characteristics make them ideally suited for the realization of devices that closely mimic biological electrochemically gated membranes in neurons and information-tracking synapses.

Methodology

We will use the methodology of impedance spectroscopy and equivalent circuit analysis to fabricate devices with dynamic responses that emulate natural neuronal coupling and synchronization. This method will produce the hardware needed for a preferred spiking computational model, incorporating time, analog physical elements, and dynamical complexity as computational tools.

Illustration

As an illustration, we will show visual object recognition from spiking data provided by a spiking retina using advanced neuristors and dynamic synapses.