Professors Park Young-ran, left, and Wang Gun-uk of the KU-KIST Graduate School of Converging Science and Technology / Courtesy of Korea University
Researchers at Korea University have developed a dual-output artificial synapse designed to enhance the power effectivity of multitasking artificial intelligence (AI) techniques, the college stated Monday.
The analysis was led by professors Wang Gun-uk and Park Young-ran of the KU-KIST Graduate School of Converging Science and Technology, a joint program established by Korea University and the Korea Institute of Science and Technology.
The human mind carries out many duties without delay, processing a stream of indicators in parallel with exceptional effectivity. Conventional AI chips, in distinction, are sometimes constructed for narrowly outlined features. As a outcome, juggling a number of operations typically requires breaking computations into separate elements or working them in sequence — a design that may sharply enhance energy consumption.
Replicating the mind’s functionality has grow to be a rising focus in AI analysis as techniques require rising quantities of computing energy and electrical energy.
The KU-KIST group created a brain-inspired artificial synapse that emits each electrical and optical indicators concurrently, enabling AI techniques to deal with a number of duties in parallel on a single chip. The gadget demonstrated secure studying conduct throughout about 1,000 distinct states, in line with the researchers.
In testing, the gadget improved computational velocity by as much as 47 p.c and diminished power consumption by as a lot as 32 occasions in contrast with typical GPU-based {hardware} accelerators.
“This achievement presents a new hardware architecture for multitasking AI through an artificial synapse that simultaneously utilizes electrical and optical signals,” Wang stated. “It could be further expanded to high-speed, low-power AI systems in fields requiring complex decision-making, such as robotics, medical and health care applications, and autonomous driving.”
The examine was printed Friday in Science Advances, a journal of the American Association for the Advancement of Science.