All-Optically Controlled Artificial Synapses Based on Light-Induced Adsorption and Desorption for Neuromorphic Vision

Abstract

Artificial synapses with the capability of optical sensing and synaptic functions are fundamental components to construct neuromorphic visual systems. However, most reported artificial optical synapses require a combination of optical and electrical stimuli to achieve bidirectional synaptic conductance modulation, leading to an increase in the processing time and system complexity. Here, an all-optically controlled artificial synapse based on the graphene/titanium dioxide (TiO 2 ) quantum dot heterostructure is reported, whose conductance could be reversibly tuned by the effects of light-induced oxygen adsorption and desorption. Synaptic behaviors, such as excitatory and inhibitory, short-term and long-term plasticity, and learning−forgetting processes, are implemented using the device. An artificial neural network simulator based on the artificial synapse was used to train and recognize handwritten digits with a recognition rate of 92.2%. Furthermore, a 5 × 5 optical synaptic array that could simultaneously sense and memorize light stimuli was fabricated, mimicking the sensing and memory functionality of the retina. Such an all-optically controlled artificial synapse shows a promising prospect in the application of perception, learning, and memory tasks for future neuromorphic visual systems.

Tailoring Wavelength-Adaptive Visual Neuroplasticity Transitions of Synaptic Transistors Comprising Rod-Coil Block Copolymers for Dual-Mode Photoswitchable Learning/Forgetting Neural Functions

“…Table summarizes the synaptic performance of the proposed photosynapse with its counterparts in fully light-controlled artificial synapses and memory devices. Overall, the proposed BCP-based photosynapses have decent PPF indices, which can compete with their inorganic counterparts under ultralow optical pulse widths. ,,,– This is one of the few works that can exhibit long-term memory behaviors with fully light-controlled modes. ,, Notably, this is the first work to demonstrate fully light-controlled short- and long-term memory based on organic material systems. These results underline the feasibility of employing the BCP photosynapses in real-time artificial visual devices and broaden their horizons for state-of-the-art high-speed image processing/memory systems utilizing all-optical operations.…”

All‐Optically Controlled Retinomorphic Memristor for Image Processing and Stabilization

“…Compared to existing EIS systems, in which perception, memory and computational processes are performed separately, our retinomorphic memristor performs various processes on one device and thus shows promise in image stabilization systems. [ 33 ] Preprocessing steps such as image contrast enhancement can highlight the physical features in an image; [ 21 ] thus, we performed simulations to preprocess more complex images. We simulated an artificial neural network (ANN) based on the Modified National Institute of Standards and Technology (MNIST) dataset of handwritten digits.…”

“…[13,18] As a result, researchers have actively investigated alloptical synaptic devices based on oxides, 2D materials, and organics. [33][34][35][36][37] These devices enable the reversible modulation of conductance through different wavelengths of light, making them suitable for vision sensor applications. Despite significant advancements in this field, several challenges remain to be addressed.…”

“…One is that the wavelengths of light required by these devices are not always in the visible range, and another is that optical conduction cannot be modulated by the light of the same wavelength, both of which will increase the complexity of vision sensor applications. [31,33,34] To overcome these issues, it is necessary to research and develop new photosensitive materials that are well-suited for use in artificial optical synapses.…”

Biomimetic Artificial Tetrachromatic Photoreceptors Based on Fully Light‐Controlled 2D Transistors

“…[27] Besides, some devices are based on the mechanism of light-controlled surface gas molecule absorption/desorption, which requires air exposure and therefore suffers from a low instability compared to packaged solidstate devices. [5,28,29] In addition, most of the reported fully lightcontrolled synaptic devices require a strong light power density of ≈100 mW cm −2 , which is far stronger than the light power in the natural environment and the incident light power that most animal eyes can withstand (10 mW cm −2 ). [5,30,31] Even more, the photoreceptors that could mimic a tetrachromatic biological vision system not only need to achieve fully-light controlled bidirectional synaptic plasticity, but also require modulation by four specific types of lights, i.e., UV, red, green, and blue, with low power intensity of a few mW cm −2 .…”