We aim to develop artifi­cial neural networks with brain-inspired circuits. Like in the brain, artifi­cial neurons and synapses are built with novel memris­tors, arranged in a cross­bar array. By combin­ing these with ultra-fast photon­ics, we seek to improve signal process­ing and matrix-vector multi­pli­ca­tions to address limita­tions from state-of-the-art archi­tec­tures. This approach aims to advance comput­ing solutions by reduc­ing energy consump­tion, compu­ta­tional time, and system complexity.

We aim to develop hardware-based artifi­cial neural networks using analogue circuits inspired by the human brain. Our approach relies on novel memris­tors that can mimic the behav­iour of neurons and synapses by single-atom movement within them. Arrang­ing these memris­tors in a cross­bar-like struc­ture allows us to create complex neuro­mor­phic circuits for various appli­ca­tions. This new neuro­mor­phic archi­tec­ture requires specialised algorithms for in-memory comput­ing, which will also be devel­oped and researched.

The innova­tion in our approach lies in combin­ing these analogue memris­tive circuits with ultra-fast integrated photon­ics, enabling us to handle data with greater speed and efficiency. Initially, we will focus on signal process­ing appli­ca­tions, where we aim to reduce system complex­ity and compu­ta­tional demands. As the project devel­ops, we plan to advance toward more sophis­ti­cated circuits for general comput­ing, address­ing the energy consump­tion bottle­neck associ­ated with the conven­tional Von Neumann computer architecture.

The devel­op­ment of photonic-neuro­mor­phic circuits for artifi­cial neural networks marks a signif­i­cant step toward rapid, low-energy, brain-inspired comput­ing. This allows for revolu­tion­is­ing signal process­ing technol­ogy in the commu­ni­ca­tion indus­try and devel­op­ing a practi­cal solution to reduce energy consump­tion through in-memory computing—a key challenge in scaling AI.