Boosting Artificial Olfaction: Visual Cues-Enhanced Gas Classification by a Bimodal Neuromorphic Device.

Abstract

Artificial olfactory sensors have garnered significant attention in various applications, including micro-robotics, implantable medical devices, and consumer electronics. However, they still face challenges in trade-offs among high recognition accuracy, compact size, and low power consumption. Existing strategies can rely on large-scale sensor arrays (up to 10⁴ elements) to enhance gas recognition accuracy, but this substantially increases system size and power consumption. Inspired by biological multisensory synergy, we propose a visual-olfactory bimodal neuromorphic device to overcome these limitations. It emulates biological perceptual fusion, including bimodal perceptual weighting and enhancement. With a small active area of 148 µm², a static power consumption of only 3.4 µW, and a low operating voltage of 1 V, the device exhibits ppb-level sensing performance and is capable of both classifying gas types and identifying concentrations for multiple target gases. The proposed bimodal perception strategy achieves a gas classification accuracy of 98.27%, far exceeding that of the olfactory unimodal mode (52.24%), and, importantly, enables precise discrimination of mixed gases with highly overlapping sensing signatures. Our strategy not only provides a unit architecture for constructing miniaturized, low-power, and highly accurate artificial olfactory systems but also paves the way for next-generation bio-inspired multimodal neuromorphic sensing.