Artificial Leaky Integrate-and-Fire Sensory Neuron for In-Sensor Computing Neuromorphic Perception at the Edge

Neuromorphic perception and computing show great promisein termsof energy efficiency and data bandwidth compared to von Neumann'scomputing architecture. In-sensor computing allows perception informationprocessing at the edge, which is highly dependent on the functionalfusion of receptors and neurons. Here, a leaky integrate-and-fire(LIF) artificial spiking sensory neuron (ASSN) based on a NbO (x) memristor and an a-IGZO thin-film transistor(TFT) is successfully developed. The ASSN is fabricated mainly throughsimple sputter deposition processes, showing the prospect of highprocess compatibility and potential for integration fabrication. Thedevice shows excellent spike encoding ability to deliver the neuromorphicinformation through spike rate and time-to-first spike. Moreover,in the ASSN, the a-IGZO TFT not only provides the fundamental spikesignal computing function of the artificial neuron but also has NO2 gas and ultraviolet (UV) light dual sensitivity to introducethe neuromorphic perception capability. As a result, the ASSN successfullyexhibits an inhibitory property under NO2 stimulation whileexhibiting an excitatory state under UV light stimulation. Futhermore,self-adaption and lateral regulation circuits between different ASSNsare proposed at the edge in mimicking biological neurons' richinterconnection and feedback mechanisms. The ASSNs successfully achieveself-regulation after a huge response during a burst stimulus. Inaddition, the neuron transmits a more obvious output when the target-sensitiveevents occur through the edge internal regulation. The self-adaptionand lateral regulation demonstrated in ASSN move an important stepforward to in-sensor computing, which provides the potential for amultiscene perception in complex environments.

Automated summary

This article introduces an artificial spiking sensory neuron based on a NbO(x) memristor and an a-IGZO TFT. The neuron has excellent spike encoding ability and dual sensitivity to NO2 gas and UV light. Additionally, the proposed self-adaption and lateral regulation circuits mimic the rich interconnection and feedback mechanisms of biological neurons.