Sungkyunkwan University researchers develop technology for brain-like balanced learning

Professors Yang Woo-seok, left, and Jo Sae-byeok / Courtesy of Sungkyunkwan University

A Sungkyunkwan University research team has developed a technology that can strengthen or weaken the memory of an artificial intelligence (AI) semiconductor device simply by changing the color of light.

The university said Thursday that the team has devised a next-generation synapse for artificial neural networks by utilizing the “disorder” and “defect” characteristics — long regarded as inherent challenges of semiconductor materials — as a means of maintaining memory homeostasis.

It added that the technology mimics the human brain by allowing important information to be retained for a long time, while weakening access to unnecessary information.

The team, led by professors Jo Sae-byeok and Yang Woo-seok at the university’s School of Chemical Engineering, said the technology is expected to be applied to next-generation artificial intelligence chips to consume less power, as well as “see-and-remember” artificial eyes.

The findings were published May 18 in the international science journal Nature Communications under the title, “Disorder-mediated Non-equilibrium Photocurrent Redistribution Enables Homeostatic Synaptic Conditioning in AgBiS2 Heterostructure.”

By changing the color of light, a single disorder-engineered synapse selectively strengthens or weakens its memory, enabling brain-like homeostatic learning. Courtesy of Sungkyunkwan University

The university said the human brain actively keeps “learning” in balance by holding on to what matters and letting go of what does not.

It noted that the research team reproduced the human brain’s ability in a semiconductor device, using the color of light to strengthen (remember) or weaken (forget) an artificial synapse's memory.

“Knowing how to forget is as important as knowing how to remember. The essence of this work is that we separated those two functions by the color of light, and revived what was considered a defect into a self-balancing learning function for AI hardware,” Professor Jo said.

The approach is not limited to one material, and all processing uses low-temperature, ink-based solution methods compatible with existing semiconductor lines.

Jo expects the technology to contribute to light-based neuromorphic computing, low-power AI accelerators, in-sensor computing, and machine-vision systems for artificial retinas that can see and remember.

The research was supported by the Ministry of Science and ICT and the Ministry of Education.