(650b) Quantum Materials for Energy Efficient Neuromorphic Computing

Data manipulation in its many forms drives and fuels our civilization. Revolutionary developments in the past decades in hardware (principally CMOS technology) and software (such as machine-learning and artificial intelligence), has fueled the ever-increasing capabilities of modern computational machines. It is however agreed that these enhanced computational capabilities will soon slow down considerably, due to a variety of limitation, principally because of the large energy consumption expected in existing CMOS based hardware. On the other hand, nature has evolved a energy efficient computational machine (the “brain”) which has substantial advantages over conventional silicon-based computers.

I will report the effort of a large number of investigators collaborating to design and investigate the use of quantum materials to develop energy efficient neuromorphic architecture. In particular, I will describe a novel class of “thermal neuristor” based on spiking oscillators which function and communicate through thermal processes. These neuristors exhibit a wide range of electrical behavior that closely resembles that of biological neurons including: all-or-nothing law, type-II neuronal rate coding, spike-in and DC out, spike-in and spike-out, and stochastic leaky integrate-and-firing. Remarkably, inhibitory capabilities are achieved using just a single oxide device, and the transmission of cascaded information occurs solely through thermal interactions without any intricate circuits. This research provides some of the groundwork for scalable, energy-efficient, thermal neural networks, advancing the field of brain-inspired computing.

Work done in extensive collaboration within the Quantum Materials for Energy Efficient Neuromorphic Computing (Q-MEEN-C) an Energy Frontier Research Center (EFRC) funded U.S. Department of Energy, Office of Science under Award # DE-SC0019273 and the AFOSR under award number FA9550-22-1-0135