NEW Prof. Seo Hyung-tak Develops AI-based Photovoltaic Memory Device that Imitates Human Vision
A research team led by Ajou University Professor Seo Hyung-tak has successfully developed an AI-based photovoltaic memory device that can imitate human visual capacity and store images and distinguish data. It is expected to find use in the development of artificial neural visual devices and applications in a variety of fields, including intelligent photovoltaic sensors, data processing, and robot engineering.
Prof. Seo Hyung-tak (Department of Materials Science and Engineering, Graduate School of Energy Systems, center in photo) announced the development of an optical memory device that can detect optical image signals and store them in the form of non-volatile data, and conduct image storage programming according to input signals. This device is highly transparent and can be independently powered.
The results were published in a thesis entitled, “Highly transparent reconfigurable non-volatile multilevel optoelectronic memory for integrated self-powered brain-inspired perception” in the August 27, 2021 online edition of Nano Energy (IF=17.881), an international journal in the field of nano technology. Ajou University’s Professor Mohit Kumar (right in photo) and master’s degree student Lim Jaeseong (left in photo) also took part in the research.
The human visual perception system can detect optical signals in real time, and recognizes them differently according to intensity or repetition. Not only does it selectively save the received information in long-term or short-term memory, but it can also discern its importance. Human vision grasps optical information, encodes it into electrical spikes of appropriate sizes through photovoltaic transformation, and delivers the information to the visual cortex in the brain and stores it in the bio synapse network.
As such, if the bio visual perception integrating optical signals detection and information storage can be imitated and made into a device, it can process intelligent visual information. With current technology, the optical sensor is separated from an information storage memory device. In addition, signal treatment devices are also needed to identify the meaning of the optical signals, and high-level video processing, primarily intelligent CCTVs, essentially require the composition of complicated hardware. An effective, simple humanoid photoelectric circuit design that can ultimately replace existing complicated circuits is needed to achieve a technological breakthrough. To simplify such circuits, signal processing and memory storage must be integrated so that the device can adjust to diverse environments, discriminate between information and store it.
Ajou University’s research team developed Nano film (optical sensor) using high quality titanium dioxide (TiO2), and created a core-shell Nano column array (memory) combined with nickel and nickel oxide (NiO) to overcome this limitation and integrate non-volatile memory and intelligent optical sensor functions. Then, it developed the first-ever heterojunction device that evenly allocates nickel oxide film (optical switch) and silver Nano wire, and placed it in the top layer. The nickel oxide Nano column developed by the research team can securely collect and store photoelectric charges generated by the photoelectric current, serving as the core material for intelligent non-memory.
The research team used Nano columns, combined with nickel and nickel oxide, to control the distribution of oxygen vacancy (vacancy created by the lack of oxygen in a metal oxide crystal), thereby ensuring safety. The width of a NiO Nano column is around 20-30 nanometers. In the measurement of photoelectric currents using atomic force microscopy (AFM), optical switching features seemed stable in an area of 30 nanometers, which means it can produce pixel intensity of 716GB per 1-inch area.
The research team succeeded in imitating several functions of bio visual perception using the photovoltaic device they developed. It confirmed that the ultraviolet optical pulse signal that is input every 0.1 second can be strengthened or weakened in the long term to create intelligent non-volatile memory storage and programming. The intensity of optical signals, and the polarity and size can be used as variables to store and discern optical signals at multiple levels.
The research team’s photovoltaic device can strengthen or weaken data by converting voltage even if the optical pulse signals are of the same intensity and number. This signifies it has the same effect as how visual data repeatedly perceived in a human’s visual perception remains in memory for a long time, whereas a one-time bit of visual data is quickly forgotten. The research team created a 5x5 pixel array with the developed devices and demonstrated how it can discern short messages.
“The human visual perception system has an intensely complicated structure that cannot be made with integrated circuit devices using current technology,” said Prof. Seo. “Through this research, we managed to develop a simplified memory integrated optical sensor at the device level, which will be utilized in the future for development of an AI optical perception system.”
Prof. Seo added, “With additional research and development, it will ultimately have broad applications, including intelligent CCTV, data processing, and robot engineering.”
This research was sponsored by the General and Mid-Career Researcher Support Programs [KHTC1]hosted by the Korean Ministry of Science and ICT (MSIT) and the National Research Foundation of Korea (NRF).
[KHTC1]“중견·기본 기초연구지원사업” 임의번역하였습니다.