Computing hardware is facing the urgent challenge of processing massive amounts of data with low power consumption and low latency, in particular because of the rapidly increasing demands from Artificial Intelligence (AI). Our vision is to develop a radically new… Read more hardware platform for ultra-fast (~1 μs) and energy-efficient (~ 1 μW) edge-AI inference based on a world-first, unique, braininspired hybrid architecture of analogue in-memory computing, realised with nanoscale dopant network processing units (DNPUs) in silicon integrated with silicon/silicon nitride memristors. We apply the principle of material learning to exploit the intrinsic nonlinearity and tunability of a network of dopants in silicon without the need to design tailor-made circuitry for desired functionality. We integrate complex tunable nonlinear functionality with embedded memory to realise colocation of processing and memory, addressing the von Neumann bottleneck. A new unconventional computing paradigm is expected by combining in-materio and neuromorphic computing approaches. Our technology will be validated for all-hardware systems via machine-learning benchmarks and progressively more complex tasks with the MNIST handwritten digits, CIFAR10 and federated learning. Importantly, the silicon-based platform is an innovative, unconventional computing architecture that can nevertheless be realised with conventional CMOS technology. Low power consumption, low latency, and a small footprint make our technology uniquely suitable for edge computing, with unprecedented performance in areas such as autonomous AI and image processing for health and security. DN4UC will thus establish an energy-efficient, fast, nanoscale processing platform to handle societal challenges ranging from personalised health, climate, energy, to security, contributing to the digital and green transition. Moreover, we expect DN4UC to contribute to securing and maintaining the EU’s global leadership in AI.