Printed neuromorphic hardware for artificial intelligence

We propose a project that explores the transformative potential of neuromorphic computing, a field inspired by the information processing principles of the nervous system, in the context of artificial intelligence (AI) applications.

We focus on the development of the first fully printed neuromorphic circuit incorporating analog memory, aiming to reduce the cost of deploying pre-trained machine learning (ML) models.The proposed circuit deploys a trained multilayer perceptron model to distinguish between the fumes of five common solvents presented to an electronic nose.

Our investigation will focus on three work packages: (1) the development of an ink formulation and a printing process to fabricate the first printable multi-state memory device, (2) the creation of novel reconfigurable circuits for processing and normalization of electronic nose signals, and (3) the large-scale integration of neuromorphic circuits.The ability to print custom circuits representing ML models holds promise to expand both the types of applications that can implement AI and the number of people to whom the technology is accessible.

We envision a robust ecosystem of hobbyists and entrepreneurs developing printed ML models for niche applications, facilitated by the availability of low-cost inks and modular, customizable circuits.