The Korea Advanced Institute of Science and Technology (KAIST) has developed a self-learning memristor that replicates human brain synapses, advancing AI computing efficiency and local processing.
Memristors, or “memory resistors,” are touted as the best candidates for mimicking synapses in neuromorphic computers. KAIST’s latest development surpasses previous attempts, offering enhanced synapse replication. This breakthrough could enable AI to operate locally, boosting energy efficiency and task improvement over time.
In 1971, Leon Chua theorized the existence of a fourth fundamental computing element—a memristor. This component could store data even when turned off, forming the bedrock of neuromorphic computing. Memristors can handle data storage and computation simultaneously, akin to the human brain. Since their discovery in 2008, researchers worldwide have been refining memristor capabilities to create brain-like computers.
In January 2025, KAIST announced a memristor that corrects errors and learns from them, solving previously challenging neuromorphic tasks. For instance, this chip can separate moving images from backgrounds during video processing and improves over time. The breakthrough was detailed in Nature Electronics.
This AI learns your body and outsmarts your diabetes
KAIST claims this memristor allows for local AI processing, bypassing reliance on cloud servers and enhancing privacy and energy efficiency. Researchers Hakcheon Jeong and Seungjae Han likened this system to a smart workspace, where all tasks occur in a single, efficient location.
KAIST has also developed the first AI superconductor chip, which runs at ultra-high speeds with minimal power consumption. Mimicking the brain’s efficiency, this chip performs a billion-billion operations per second using just 20 watts of power.
Enhanced memristors move us toward a brain-on-a-chip, accelerating AI development and potentially nearing the technological singularity. However, achieving true human-like intelligence in AI remains a complex challenge.
While the hype around the technological singularity may be a bit overblown, KAIST’s AI superconductor chip hyper-focuses on energy efficiency and speed, showcasing a practicality that could drive real-world applications long before we hit Skynet territory.
Under the cherry-roof of memristor hype, KAIST’s ability operates locally, which sways the ethical pendulum away from central cloud control. If this becomes the norm, companies needing to leverage AI could bifurcate their operations away from the control of Big Clouds.
The real meat lies in separating moving images from static backgrounds—a minor task on paper, but crucial in real-world applications like vehicle and drone navigation. This task used to be a stringent test for neuromorphic chips, but KAIST’s newest technique proves it’s ready for the main stage.
