Advancements in artificial intelligence are driving progress in wearable visual bionic devices. However, traditional silicon vision chips suffer from high energy consumption and limitations in mimicking biological systems. Birds possess exceptional visual capabilities, making them attractive models for bionic visual enhancement. Current vision architectures separate sensing, memory, and processing, leading to inefficiencies. In-memory computing, which integrates these functions, offers a solution, particularly when using flexible, biocompatible organic materials. This research aims to create a highly efficient, biomimetic optoelectronic neuromorphic system by designing a novel material and device architecture inspired by avian vision, overcoming limitations of existing organic semiconductor-based systems which often lack non-volatile properties or broad wavelength response.

Previous studies have explored in-memory sensing and computing frameworks using organic materials to emulate the human retina. Chen et al. demonstrated an organic electrochemical optoelectronic synapse with significant photocurrent at low voltages, enabling non-volatile conductance states for neuromorphic computing across the visible spectrum. Huang's group developed a material-algorithm co-design for in-situ sensing and optical input pre-processing (contrast enhancement, noise reduction). Liu et al. presented a semi-crystalline macromolecule for ion-based organic memristive devices. However, existing optoelectronic artificial visual devices based on these systems often lack non-volatile properties or sufficient broadband response (solar-blind to near-infrared). Furthermore, large-scale integration of 2D materials remains a challenge.

This study introduces a dual-junctional enhanced birdlike broadband neuromorphic visual sensor (BBNVS) array. GaAs nanowires (NWs) arrays were synthesized via catalytic solid-source chemical vapor deposition (CVD) and transferred onto a self-assembled poly(3-hexylthiophene-2,5-diyl) (P3HT) organic film to create a P-N junction. Ag electrodes formed a Schottky junction. The meticulously ordered structure of the self-assembled P3HT, combined with Schottky barriers, enables broadband in-memory sensing and computing. The BBNVS arrays were fabricated on various substrates (glass, PET, PDMS, PI, acrylic). A multi-tasking reservoir computing (RC) system was used for multimodal recognition (shape, motion, color, UV grayscale). Characterizations included scanning electron microscopy (SEM), high-resolution transmission electron microscopy (HRTEM), atomic force microscopy (AFM), Raman spectroscopy, photoluminescence (PL) spectroscopy, X-ray photoelectron spectroscopy (XPS), and synchrotron grazing-incidence wide-angle X-ray scattering (GIWAXS).

The BBNVS array demonstrated over 5 bits of in-memory sensing and computing with both positive and negative photoconductivity across visible light wavelengths (blue, green, red). Over 128 memory states were achieved in the solar-blind range. A near-zero power consumption operating mode was realized, mimicking avian energy-saving behaviors. The device showed excellent bending resistance on various flexible substrates. The RC system achieved up to 94% accuracy for color recognition and successfully extracted motion and UV grayscale information (with sunscreen), leading to fusion visual imaging. Analysis using synchrotron GIWAXS revealed a highly ordered edge-on arrangement of P3HT molecules, crucial for efficient charge transport and non-volatile storage. The dual junctions (P-N and Schottky) synergistically enhance carrier separation, injection, and retention. The 5x5 array demonstrated stable, non-volatile optical signal storage.

The findings address the research question by demonstrating a novel BBNVS array that surpasses limitations of current organic semiconductor-based systems. The biomimetic design, inspired by avian vision, enables superior performance in terms of broadband response, non-volatile storage, low power consumption, and flexibility. The integration of the BBNVS with a reservoir computing system showcases the potential for efficient, multi-task in-memory sensing and computing. The ability to perform fusion imaging (visible and UV) opens up new possibilities for applications requiring comprehensive visual information. The success of the 5x5 array suggests scalability and feasibility for larger arrays.

This work successfully demonstrated a birdlike broadband neuromorphic visual sensor (BBNVS) array with enhanced visual perception capabilities, including non-volatile broadband optical signal storage, low-light perception, and near-zero power consumption. The combination with a reservoir computing system achieved high-accuracy color recognition and fusion visual imaging. The flexible design and demonstrated array integration show great potential for wearable applications. Future research could focus on expanding the array size, exploring alternative materials, and integrating this technology into complex visual processing systems.

While the 5x5 array demonstrates the potential for large-scale integration, further research is needed to optimize the scalability and yield of larger arrays. The current study focused on specific color and motion recognition tasks; additional research is needed to evaluate performance across a wider range of tasks and more complex scenarios. The use of sunscreen as a UV pattern might limit the applicability to real-world scenarios where the patterns are not pre-defined.