A TaOx-based self-rectifying memristor for a highly compact thermal in-sensor computing system

The number of sensory nodes has grown rapidly with the development of Artificial Intelligence of Things (AIoT), bringing about challenges in terms of hardware integration and redundant data transfer. Here, to address these problems, we propose a highly compact thermal in-sensor computing system based on a sensory crossbar array and leaky integrate-and-fire (LIF) neurons, both constructed using TaOx-based memristors. Benefiting from the self-rectifying capabilities (>10(4)), high uniformity (3.77%) and switching mechanism-based temperature-sensing characteristics (298-418 K) of this device, it can be integrated into a crossbar array to execute both sensing and preprocessing functions simultaneously and efficiently. Moreover, this TaOx-based memristor shows the ability to integrate spatial-temporal information like biological dendrite neurons. Thus, we combine our device in series with a threshold switching unit to realize LIF neuron function. Combining the sensory crossbar array, LIF neurons and a three-layer artificial neural network, a thermal in-sensor computing system for fingerprint recognition was constructed finally and achieved 100% accuracy based on the FVC 2004 database in simulation. This work provides a new design concept for a highly compact and efficient thermal in-sensor computing system.