Edge computing systems, particularly in mobile and security applications, face growing demands for real-time data processing at the device level. Traditional approaches require multiple discrete components and rely on frequent data transfers between sensors and processors, leading to latency, power consumption, and integration inefficiencies.
Memristor-based in-sensor computing is a potential solution for supporting signal processing and memory in the same array, but it is limited by high voltage requirements, large fully connected networks, and the need for analog-to-digital converters.
Reservoir computing, which uses nonlinear transformations to extract temporal features while limiting training to a lightweight readout layer, offers a more compact solution. However, many reservoir computing systems require different types of memristors for the reservoir and output layers, causing greater design complexity. Existing organic optoelectronic memristors can assist with low-power computing tasks but often depend on heterojunctions, which limit their scalability and reconfigurability.
These obstacles can be avoided with organic polymers. Their tunable photoelectric properties, based on adjustable crystallinity and molecular stacking, make them ideal candidates for building low-power, flexible devices.
The Forming-Free Optoelectronic Memristor
In this study, the research team fabricated a forming-free optoelectronic polymer memristor with a PTB7-Th solution, which can be electrically and optically tuned at millivolt-level voltages. The device has dynamic, on-demand control and a simplified device architecture that combines sensing, reservoir processing, and classification within a single unit.
The memristor was fabricated from indium tin oxide-coated glass substrates, which were cleaned, UV-ozone treated, and coated with a spin-cast layer of PTB7-Th. After annealing, thermal deposition of 25 nm silver top electrodes created a crossbar array with different cell areas. Silica, copper, and gold electrodes were created via the same method.
Structural and surface analysis was performed using scanning electron microscopy, stylus profilometry, and atomic force microscopy. The researchers also used Kelvin probe force microscopy and conductive AFM to investigate charge distribution and conductivity at the nanoscale.
Optical characterization included UV-visible spectroscopy, ultraviolet photoelectron spectroscopy, fluorescence emission measurements, and time-resolved photoluminescence. Electrical testing was conducted under ambient and vacuum conditions using a Keithley semiconductor parameter analyzer, with photoresponse evaluated under a Xenon lamp across 320 to 670 nm using pulsed illumination.
The device achieved a fingerprint recognition accuracy of 97.15 % while operating at ultra-low power, maintaining a compact reservoir size, and eliminating the need for additional ADCs.
Photocurrent could be dynamically controlled through simple voltage adjustments, allowing the same memristor to switch between linear and nonlinear response regimes depending on operational needs. Analog signal transmission was maintained throughout, reducing latency and power consumption.
Download your PDF now!
Impact and Implications
The memristors were manufactured using a solution-processable, low-temperature method that is compatible with flexible substrates and also scalable for mass production. They show potential for integration with organic photovoltaic arrays, revealing greater prospects for self-powered, on-chip computation.
The study suggests that their devices and similar memristors may be especially well-suited to future biometric systems, including under-display fingerprint sensors, as well as smart, portable electronics that require energy-efficient, on-device intelligence.
By demonstrating a fully analogue, forming-free, voltage- and light-controlled memristor capable of handling sensing, computing, and memory functions, the researchers provide a new framework for in-sensor edge computing.
Journal Reference
Zhou, J. et al. (2025). Optoelectronic polymer memristors with dynamic control for power-efficient in-sensor edge computing. Light: Science & Applications, 14(1), 1-11. DOI: 10.1038/s41377-025-01986-9, https://www.nature.com/articles/s41377-025-01986-9
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.