Analog computing platform based on one-memristor array efficiently processes real-time videos

As artificial intelligence models become increasingly advanced, electronics engineers have been trying to develop new hardware that is better suited for running these models, while also limiting power-consumption and boosting the speed at which they process data. Some of the most promising solutions designed to meet the needs of machine learning algorithms are platforms based on memristors.

Memristors, or memory resistors, are electrical components that can retain their resistance even in the absence of electrical power, adjusting their resistance based on the electrical charge passing through them. This means that they can simultaneously support both the storage and processing of information, which could be advantageous for running machine learning algorithms.

Memristor-based devices could be used to develop more compact and energy-efficient hardware for running AI models, including emerging distributed computing solutions referred to as edge computing systems. Despite their advantages, many existing memristor-based platforms have been found to have notable limitations, adversely impacting their reliability and endurance.

Researchers at Korea Advanced Institute of Science and Technology (KAIST) and other institutes in the Republic of Korea recently developed a new analog computing platform based on a selector-less memristor array.

This platform, introduced in a paper published in Nature Electronics, was found to run AI algorithms for the processing of real-time videos both efficiently and reliably.

“Systems based on memristor arrays face challenges implementing real-time AI algorithms with fully on-device learning due to reliability issues, such as low yield, poor uniformity and endurance problems,” wrote Hakcheon Jeong, Seungjae Han and their colleagues in their paper. “We report an analog computing platform based on a selector-less analog memristor array.”

The researchers’ platform consists of 1,024 memristors made of titanium oxide (TiO_x), arranged in a 32 x 32 grid. These memristors were found to reliably store and process the information fed to machine learning algorithms, working consistently over time.

“We use interfacial-type titanium oxide memristors with a gradual oxygen distribution that exhibit high reliability, high linearity, forming-free attribute and self-rectification,” wrote Jeong, Han and their colleagues.

“Our platform—which consists of a selector-less (one-memristor) 1 K (32 × 32) crossbar array, peripheral circuitry and digital controller—can run AI algorithms in the analog domain by self-calibration without compensation operations or pretraining,”

To assess the potential of their analog computing platform, the researchers carried out various tests, where they used it to run AI models for the real-time processing of videos employing an online training scheme. This framework was used to train the models to discern between the dynamic elements in videos and static backgrounds.

“We illustrate the capabilities of the system with real-time video foreground and background separation, achieving an average peak signal-to-noise ratio of 30.49 dB and a structural similarity index measure of 0.81; these values are similar to those of simulations for the ideal case,” wrote the researchers.

So far, the new memristor-based platform developed by Jeong, Han and their colleagues has yielded very promising results, enabling the efficient and reliable processing of real-time video streamed directly from a camera.

In the future, it could be improved further and evaluated on a broader range of real-time data processing tasks, potentially contributing to the advancement of edge computing solutions.

© 2025 Science X Network