Vacancy: Postdoc position on Material learning with dopant network processing units

Are you interested in a job at the interface of machine learning and device physics to contribute to a new generation of energy-efficient hardware for artificial intelligence? If you have a PhD with a background in semiconductor physics, electrical engineering or machine learning, you may want to apply for this postdoc position within the Horizon Europe Pathfinder project HYBRAIN.

This position is part of the Horizon Europe Pathfinder project HYBRAIN. HYBRAIN’s vision is to realize a radically new technology for ultra-fast and energy-efficient edge AI inference based on a world-first, unique, brain-inspired hybrid architecture of integrated photonics and unconventional electronics with collocated memory and processing. As the most stringent latency bottleneck in CNNs arises from the initial convolution layers, we will take advantage of the ultrahigh throughput and low latency of photonic convolutional processors (PCPs). Their output is processed using cascaded analogue electronic linear and novel nonlinear classifier layers.

In 2020 we introduced the concept of dopant network processing units (DNPUs, Nature 577, 341-345 (2020)). In this postdoc project, you will work on the development of efficient machine-learning procedures to be applied to DNPU networks in combination with photonic and analogue electronic hardware. We will move to larger multi-DNPU networks and investigate the training performance of various conventional gradient descent methods for multi-DNPU networks in simulation first.

For more information regarding this position, browse the University of Twente Careers or contact Prof. Wilfred G. van der Wiel (W.G.vanderWiel[at]utwente[dot]nl).