Nanotechnology By The Experts

Brain-inspired computing hardware is emerging as an alternative to silicon complementary metal-oxide semiconductor chips in order to address the energy crisis of rapidly increasing digital data generation and processing. Conventional non-volatile memories can realize highly parallelized in-memory computing in neural networks, but they lack the adaptability and reconfigurability that are the key attributes of low-energy biological systems. To this end, neuromorphic devices based on low-dimensional nanomaterials embody bio-realistic tunable learning, coupled state variables, non-linear responses, and multi-terminal architectures of synapses. In this Distinghuished Lecture, Sangwan makes the case for two-dimensional (2D) nanomaterials for neuromorphic hardware by drawing parallels between fundamental properties, form factors, and required functionalities in artificial synapses, neurons, and network architectures. As a few examples, MoS₂memtransistors show bio-realistic synaptic learning in scalable crossbar arrays for higher-dimensional dynamic neural networks. Anomalous ferroelectricity in bilayer graphene moiré synaptic transistors achieves unprecedented adaptive learning. Dual-gated, self-aligned, mixed-dimensional Gaussian heterojunction transistors not only realize complex spiking behavior but also enable low-power hardware for machine learning algorithms in edge computing.