Seoul National University Pioneers Ultra-Low Power Neuromorphic Hardware for Advanced AI
A research team from the College of Engineering at Seoul National University (SNU), headed by Professor Ho Won Jang, has achieved a breakthrough in ultra-low power neuromorphic hardware designed for artificial intelligence (AI) computation. The team's innovative hardware, which significantly reduces power consumption, is set to advance AI technology across various industries.
This research addresses core issues in semiconductor materials and neuromorphic devices, showing potential for large-scale, array-level technology applications. The findings were recently published in Nature Nanotechnology, a leading journal in the field, marking a major achievement for the research community.
As AI-powered applications like the Internet of Things (IoT), generative AI, and autonomous vehicles increasingly rely on massive data processing, current silicon-based computing faces challenges such as high energy demands and processing limitations. This makes the development of next-generation neuromorphic hardware—a technology that emulates the human brain’s neural processing—a critical step forward. Unlike traditional computing, neuromorphic systems use synapse-like components to perform complex computations with energy efficiency and accuracy.
The SNU team focused on halide perovskite materials, previously notable for their role in solar cells and LEDs, to develop neuromorphic devices with unprecedentedly uniform ion distribution. These advanced materials, designed with hybrid organic-inorganic structures, enabled the team to achieve ultra-linear and symmetric synaptic weight control, enhancing computation accuracy and efficiency.
Performance testing revealed that the device could handle large datasets with an error margin below 0.08%, achieving high precision on tasks ranging from basic image recognition to complex AI inference. Collaborative studies with the University of Southern California further demonstrated that the technology could operate on ultra-low power, both at the individual device level and in arrays, reinforcing its scalability and potential impact.
This innovative neuromorphic hardware offers a promising solution to the escalating energy demands of AI computation. The technology is anticipated to have broad applications in fields like autonomous driving, medical diagnostics, and AI-driven industries, while fostering advancements in AI hardware and semiconductor innovations.
The research builds on a previous study by Dr. Seung Ju Kim and Prof. Ho Won Jang, published in Materials Today, with patent applications underway in South Korea and the U.S. Dr. Kim, a distinguished researcher and Seoul National University alumnus, is currently advancing this technology at the University of Southern California, working alongside American research labs to develop intelligent semiconductors for aerospace applications.
Seoul National University, established in 1946, is South Korea’s premier national university and a key player in global engineering advancements, with the College of Engineering driving innovation through cutting-edge research and international collaborations.