Erika Covi1
NaMLab1
As the world is becoming more and more interconnected, systems increasingly operate on the edge – we have many wearable and implantable devices which monitor our health or our daily routine, give us suggestions on how to improve our wellbeing, or even save our lives, e.g., by preventing heart failure. To work properly, these devices need to be able to process the data generated and use them to learn, adapt, and interact with the external environment [1]. In this scenario, the assumption of fungibility, i.e., all the machines in a network have the same capabilities and can accomplish all tasks, which is the basis of the paradigm of powerful mainframes we have used so far, does not hold anymore, since the quality of data that our devices have to process is very diverse and has to be taken into consideration. Designing intelligent systems capable of local edge computing is therefore becoming a strategic goal in next generation smart systems. Event-based neuromorphic systems provide a low-power solution by using artificial neurons and synapses to process data asynchronously in the form of spikes [2]. However, the tight memory and power constraints in edge systems call for the design of novel architectures that use emerging technologies based on new concepts and materials to extend the functionality of state-of-the-art Complementary Metal Oxide Semiconductor (CMOS) technology [3].<br/>Thanks to its CMOS-compatible fabrication process and extreme energy efficiency, Hf-based ferroelectric devices are rapidly affirming themselves as one of the most promising technologies for neuromorphic computing [4]. Here, we present fundamental building blocks, i.e., neurons and synapses, for edge computing neuromorphic networks that combine ferroelectric and CMOS technologies. We demonstrate electrically tuneable neural and synaptic dynamics achievable by tuning the switching of the Hf-based ferroelectric devices. Finally, the main challenges to achieve a neuromorphic-ferroelectric hardware are presented, particularly in the context of optimising such systems for applications on the edge.<br/><br/><i>References</i>:<br/>[1] Covi, E., Donati, E., Liang, X., Kappel, D., Heidari, H., Payvand, M., and Wang, W. (2021). “Adaptive extreme edge computing for wearable devices”. <i>Frontiers in Neuroscience</i>, 429.<br/>[2] Indiveri, G., and Liu, S. C. (2015). “Memory and information processing in neuromorphic systems”. <i>Proceedings of the IEEE</i>, <i>103</i>(8), 1379-1397.<br/>[3] Chicca, E., and Indiveri, G. (2020). “A recipe for creating ideal hybrid memristive-CMOS neuromorphic processing systems”. <i>Applied Physics Letters</i>, <i>116</i>(12), 120501.<br/>[4] Covi, E., Mulaosmanovic, H., Max, B., Slesazeck, S., and Mikolajick, T. (2022). “Ferroelectric-based synapses and neurons for neuromorphic computing”. <i>Neuromorphic Computing and Engineering</i>, <i>2</i>(1), 012002.