Highly-dependable printed neuromorphic circuits based on additive manufacturing

Abstract The rapid development of emerging domains, such as the Internet Things and wearable technologies, necessitates flexible, stretchable, non-toxic devices that can be manufactured at an ultra-low cost. Printed electronics has emerged a viable solution by offering not only aforementioned features but also high degree customization, which enables personalization products facilitates low-cost product process even in small batches. In context printed electronics, neuromorphic circuits offer highly customized bespoke realization artificial neural networks to achieve desired functionality with very number hardware components. However, since analog components are utilized, performance influenced various factors. this work, we focus on three main factors perturb circuit output from designed values, namely, variations due printing errors, aging effects resistors, input originating sensing uncertainty. described approach, these taken into account during design (training) ensure dependability circuits. With expected accuracy robustness increased 27% 74%, respectively. Moreover, ablation study suggests that, effect variation may have similar networks. contrast, impact uncertainty is almost orthogonal variations.