Answers are grounded in this paper and cite their sources.
Broadband convolution processing is critical for accurate color image recognition. However, current broadband convolution processing involves a photoresponse matrix based on continuous‐voltage modulation devices, resulting in high energy consumption. Here, we demonstrate an InSe neuromorphic device that integrates broadband convolutional processing with perception, memory, and computation functions to realize high‐accuracy recognition of color images with the potential for low latency and reduced energy consumption. A photoresponse matrix was constructed using voltage pulses for broadband convolutional processing. The device's broadband perception capability originates from the moderate bandgap of approximately 1.3 eV in InSe, while its memory function is attributed to the ferroelectric polarization effect of Pb(Zr 0.2 Ti 0.8 )O 3 (PZT). Analog computation is realized via bias‐dependent, continued nonvolatile positive and negative photoconductive responses. Moreover, the multiband non‐volatile photocurrent matrixes generated by the InSe neuromorphic devices are used for broadband convolution processing of color images. Furthermore, combining the multifunctional InSe neuromorphic device with a convolutional neural network achieves 97.4% recognition accuracy for color “S”, “D”, “U” letter images after only ten training epochs. The proposed work is promising to set the stage for next‐generation intelligent machine vision systems in imminent embodied artificial intelligence humanoid robots.
No takes yet. Share an insight, caveat, or question.
Chen et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: