Light‐operated On‐chip Autonomous Vision Using Low‐dimensional Material Systems

Abstract

we can mimic human vision that employs a highly efficient imaging and recognition process? This will additionally provide a spill over opportunity to design artificial vision devices for the vision-impaired in our society.In this context, low-dimensional materials (LDMs) are a class of material systems that have their one, two, or three dimensions eliminated resulting in 2D, 1D, or 0D structures, respectively. The culling of dimensions brings to the fore exciting quantum confinement driven physics that can be exploited for a range of applications across sectors. The advent of large-area synthesis of LDMs has paved the way for integrated circuits, memory devices, and image sensors. A key feature of these materials is their ability to interact with light in unique ways. Research into 2D materials, 1D structures such as nanowires, and 0D structures such as quantum dots is continuously gaining momentum. A decisive move toward light-driven neuromorphics and machine vision is currently emerging particularly in the last 5 years and is expected to drive artificially intelligent systems to an unprecedented level in the coming years.These developments in new materials and advances in nanofabrication have coincided with strong emerging interest for light-driven cognition in devices. This has meant that there is an impetus toward understanding the physics and chemistry of these materials. The role played by defects in these material systems is not a detrimental one anymore. In fact, defects are now being readily harnessed for exciting functionalities. [25] This perspective is designed to provide a critical and futuristic assessment of optical/optoelectronic technologies that have emerged in the last 5 years to perform the collective function of a human eye, optic nerve, and neurons in the brain with an aim to cater to a broad readership interested in this research field. The author makes the case of why a combination of LDM systems and the new physics that emerges on their interaction with light could give rise to optoelectronic properties that can enable highly-efficient brain-like imaging and computing systems. This perspective will assess the different material and device design configurations that have been implemented tilldate and provide a discussion on future pathways that can overcome current technological limitations by rendering ultrafast processing speed due to high bandwidth, low parasitic crosstalk, and ultralow power consumption in self-learning miniaturized chips.Human-brain inspired machine vision can revolutionise new technologies across sectors. Monolithic devices that are able to achieve image capture, processing, and storage with ultra-low energy requirements can result in smart automation and enhance industrial output (both quality and quantity). This requires tapping into emerging novelties in materials physics, optical materials, and neuromorphic hardware. In this perspective, the author discusses the role of electrophotoactive low-dimensional materials and how their unique intrinsic properties can be...

Long Duration Persistent Photocurrent in 3 nm Thin Doped Indium Oxide for Integrated Light Sensing and In‐Sensor Neuromorphic Computation

“…[4] Hence, mimicking the human visual and memory system can be an effective pathway to create intelligent UV neuromorphic vision devices/chips capable of locally processing signals for concurrent pattern detection and recognition. This can be achieved with devices possessing continuously tuneable photoconductivity [5,6] which can realistically emulate the signal perception (photoreceptor), conversion, conduction (bipolar cells), processing, and modification (amacrine cells and ganglions) of the retina.…”

Atomically Thin Synaptic Devices for Optoelectronic Neuromorphic Vision

“…Figure 4c shows the classification accuracy where the performance of neural network drops due to nonlinearity in experimental data at the early epoch. This can be associated with the same polarity of nonlinearity value for LTP and LTD. [55,56] However, the classification accuracy exceeds 80% after 150 epoch, as after multiple training cycles the nonlinear weight update is compensated. The classification accuracy of our CNN model based on optoelectronic In 2 S 3 synaptic devices is comparable to the reported CNN models based on different 2D materials.…”

Operating Principle and Device Configuration Driven Mechanisms in Low‐Dimensional Materials for Neuromorphics

“…Apart from this, understanding of low-dimensional materials has itself not entirely been a straightforward approach because the electronic and optoelectronic properties of a material change with its physical topology such as thickness and size. [6,167] As such, analyzing every device and its associated material characteristics in its every topological form is vital to develop a correct understand. Also, there remains a lack of proper correlation between several physical parameters driving neuromorphic processes such as the switching ratio and thickness of the material and operating voltages and the speed of photo/electronic response between materials and devices, etc.…”