We are hiring!

Printed electronics is increasingly recognized as a key enabler for the Internet of Things as part of the “Fourth Industrial Revolution”, whose core technology advances are functionality and low-cost. Once stretchability, flexibility, porosity, and non-toxicity benefits of printed circuits are considered, printed electronics become compelling for several applications and form the core enabler towards a true pervasive computing. Despite their attractive features, their large feature sizes lead to high device latencies and low integration density. Hence, integrating Machine-Learning (ML) classification in printed circuits–mandatory in printed areas such as healthcare and medical devices, smart packaging, etc.– is extremely challenging and even questionable. In this project, we address this limitation with AutoPNN, an automated framework for generating Printed Neural Network (PNN) circuits. The first main objective of AutoPNN is to enable battery-powered complex ML classification. To achieve this we employ non-conventional computing approaches such as bespoke design, combine it with approximate computing, and apply optimizations/approximations across the entire design stack from the neural architecture down to the circuit level. Specifically, we implement a hardware-aware neural architecture search and neural minimization. The obtained PNN circuit is further optimized with approximate hardware arithmetic and finally the voltage is scaled to even near-threshold values. The second main objective of AutoPNN is to abstract the design overhead from the user and enable researchers/start-ups to join the printed electronics ecosystem. To that end, all the procedures in AutoPNN are fully automated. Area and accuracy models will be developed and used in a genetic multi-objective optimization that will search the software-hardware approximation codesign space to extract the most area-efficient circuit that satisfies user-defined accuracy and battery constraints.

This project is selected for funding by the Hellenic Foundation for Research & Innovation (HFRI) under the call “Basic Research Financing (Horizontal support for all Sciences), National Recovery and Resilience Plan (Greece 2.0)”.