Additive manufacturing of neuromorphic devices and neural network architectures

The focus of the project is to research the design, additive manufacturing and characterisation of neuromorphic and neural network architectures.

Introduction

The brain is a complex and efficient supercomputer capable of real-time processing and learning. The enhanced operation of the brain can be attributed to the nature of the brain’s base unit – the synapse, and specifically its use of ionic fluxes for communication and the complex 3D connectome structure of the brain. By mimicking the synapse, neuromorphic devices aim to achieve efficient real-time processing and learning.  Current neuromorphic architectures being developed are fabricated on rigid silicon substrates using traditional lithography techniques which only allow planar device fabrication. As a result, there is very little architectural and mechanical design flexibility.

Aim

The aim of this research is to design, additively manufacture and characterise neuromorphic and neural network architectures made based on organic semiconducting polymers such as PEDOT:PSS. We want to use additive manufacturing to overcome the limits traditional approaches impose on both the applications and also the level of achievable complexity of connectome-like structures. The project initially will focus on existing models for neuromorphic devices (e.g. de Burgt et al. Nature Materials,16, 414, 2017) and will examine how these devices can be created by additive, digitally enabled techniques (e.g. 3D printing, inkjet printing).

Once made, the neuromorphic behaviour during use will be studied, followed by a study into the evolution of the device geometry and chemistry with time. This will help us build an understanding of the evolution of the synaptic-like materials pand the device neuromorphic behaviours. After research on the performance of the integral material is carried out, more complex neural network architectures are to be designed and fabricated. In parallel, other novel semiconducting polymers currently being developed will be characterised in both the neuromorphic devices and neural network architectures

Current research

Proof-of-concept devices have been fabricated and their neuromorphic behaviour are currently being studied.