A joint Korea-U.S. research team has developed a neuromorphic semiconductor device that combines information storage and leakage current blocking in one.
Professor Lee Sang-han of GIST's department of materials science and engineering, Professor Jang Ho-won of Seoul National University's department of materials science and engineering, and Professor Joshua Yang of the University of Southern California (USC) department of electrical and computer engineering said on the 30th that their team developed a neuromorphic semiconductor device that stores information within a fixed width in response to repeated electrical stimulation while maintaining the stored state stably. The results were published in March in the international journal "Nature Communications."
Neuromorphic Computing is a technology that, like the human brain, processes information storage and computation simultaneously. Using a crossbar structure that arranges memory devices in rows and columns, it can process large amounts of data in parallel, but there have been problems of information accumulating irregularly or current leaking to adjacent devices.
The team added barium to the ferroelectric bismuth ferrite to control the movement of oxygen vacancies. Oxygen vacancies are empty sites created when oxygen atoms are missing inside the oxide, and they affect current flow and information storage characteristics.
By stabilizing the movement of oxygen vacancies through barium addition, the team realized a linear characteristic in which the stored amount increases uniformly even under repeated electrical stimulation. At the same time, they applied a self-rectifying function to the device itself to make current flow only in one direction, reducing leakage current without a separate selector device.
In experiments, the device maintained performance after more than 10 million write-and-erase operations. The rectification ratio, which indicates the directionality of current, was measured at 1,000,000 to 1 or higher.
The researchers also fabricated an 11×11 crossbar array consisting of 121 memory cells and conducted experiments writing and deleting text and image patterns. As a result, they confirmed that information in each cell can be stored and controlled stably even in a structure without a selector device.
Lee said, "We solved both the nonlinearity of information accumulation and the leakage current problem within a single device," adding, "It could be used to simplify neuromorphic semiconductor circuits and processes."
References
Nature Communications (2026), DOI: https://doi.org/10.1038/s41467-026-70727-2