Percolation with plasticity for neuromorphic computing

TL;DR

This paper introduces percolation with plasticity systems that mimic neural functionalities like memory and learning, offering a new neuromorphic computing approach inspired by brain-like networks.

Contribution

It presents a novel PWP system with multiple interfaces and resistances, enabling neuromorphic functions and offering advantages over traditional neural architectures.

Findings

PWP systems can perform matrix-vector multiplication.

They exhibit multi-valued memory and associative learning.

They have high parallelism and brain-like topology.

Abstract

We introduce the percolation with plasticity (PWP) systems that exhibit neuromorphic functionalities including multi-valued memory, random number generation, matrix-vector multiplication, and associative learning. PWP systems have multiple (N >> 1) interfaces with external circuitry (electrodes) allowing N! >> 1 measurable interelectrode resistances. Due to the underlying material properties, they undergo successive nonvolatile modifications in response to electric pulses. PWP networks offer some advantages over the existing neural network architectures. Overall, random self-tuning PWP systems with high degree of parallelism, multiple inputs and outputs present close similarities to the cortex of mammalian brain. Understanding their topology, electrodynamics, and statistics opens a field of its own calling upon new theoretical and experimental insights.