This talk presents PolHEEPo, an edge-computing chip designed to support efficient and secure workloads in the cloud–edge continuum. The chip integrates heterogeneous hardware accelerators for machine learning and cryptographic functions, namely ROGUE, PUFFO, KECCAK, and ASCON, enabling energy-efficient neural network inference together with secure data processing on resource-constrained edge devices. The presentation will first introduce the PolHEEPo architecture and then discuss the accelerators in detail, focusing on the design choices adopted to improve performance, efficiency, and trustworthiness at the edge.

In the second part of the talk, PLINIO, a deployment tool for optimising neural networks for edge inference, will be presented. The tool supports the adaptation of models to the target platform through an automated optimisation flow, simplifying the deployment of edge AI applications. The talk will conclude with a live demonstrator showing how PLINIO processes and optimises a neural network for execution on PolHEEPo.