Multi-Order Complexity Spiking Neurons Enabled by Printed MoS₂ Memristive Nanosheet Networks

The recent explosion in data-centric workloads has highlighted the shortcomings of traditional von Neumann computing and necessitated the development of brain-inspired hardware for neuromorphic computation, biohybrid systems, and smart sensing. Mimicking the functionality of biological neurons is critical to realizing the full potential of these neuromorphic applications. While silicon-based circuits using traditional transistors have been used to build neuromorphic systems, the lack of emergent phenomena and biological features such as local activity, adaptation, and functional complexity in transistors necessitates the use of elaborate circuits to realize simple biological functionalities. Hence, devices that make use of elementary physical phenomena as computational primitives are crucial to efficiently implement biomimetic computation. In this regard, dynamical devices, where intrinsic device dynamics are used to replace elaborate digital circuits, are promising building blocks for neuromorphic systems. The dynamical nature of these devices is represented in terms of the order of complexity, which is the number of first-order differential equations needed to describe the system. Currently available memristive neurons have not yet shown simultaneous first-order, second-order, and third-order spiking, which is required to emulate the full functionalities of biological neurons. Similarly, biologically-relevant spike characteristics for brain-machine interfaces have also not been achieved. Moreover, material systems amenable to scalable, low-cost solution-based fabrication of memristive neurons have not been experimentally realized. These unmet needs for the development of artificial neurons have hindered the development of neuronal systems for biomimetic spiking neural networks.<br/><br/>In this work, we use printed MoS₂ memristive nanosheet networks to demonstrate multi-order complexity spiking neurons with biologically-relevant spiking characteristics. In particular, we use the dynamical local activity arising from thermally-activated transport in MoS₂ to show the first observation of current-controlled snap-back negative differential resistance (NDR) in percolating nanosheet networks. This snap-back NDR, which leads to volatile threshold switching memristive behavior, enables artificial oscillatory and spiking neurons using simple neuristor circuits. We use the resulting oscillatory neuristor to demonstrate retina-inspired artificial sensory neurons that can be triggered and modulated using light. Our spiking neuristor also achieves simultaneous first-order, second-order, and third-order complexity spiking. Finally, we confirm the biological relevance of our spiking neuristor by activating the neural circuitry in a mouse cerebellum using in vitro cell stimulation. The results demonstrate that neuristor waveforms can be used to effectively activate neuronal circuits, inducing action potential firing and synaptic release. These printed snap-back memristors have broad implications for neuromorphic computation, biohybrid systems, and smart sensing.