University of Seoul Team Develops Flexible Hybrid Synaptic Device for Next-Gen On-Device AI

Researchers at the University of Seoul have successfully developed a "flexible organic-inorganic hybrid charge-trap synaptic device" poised to advance next-generation wearable and bio-integrated computing. Led by Professors Park Dong-wook and Kim Yun from the Department of Electronic, Electrical, and Computer Engineering, the breakthrough device effectively mimics the core functions of biological synapses.

The study, published in the esteemed journal 'npj Flexible Electronics' (IF 15.4, top 1.8% JCR), demonstrates the device's superior performance in processing time-series data via physical reservoir computing (PRC). To overcome the flexibility and stability issues plaguing current neuromorphic devices, the team employed a hybrid structure. They integrated an inorganic charge-trap layer (HfOx) with an organic semiconductor channel (DPP-DTT) on a flexible, biocompatible substrate (Parylene-C). This design harnesses the flexibility of organic materials and the stability of inorganic components, creating abundant charge traps at the interface.

“

The hybrid device we developed combines the excellent flexibility of organic materials with the reliability of inorganic materials, presenting a low-power, high-efficiency solution that can overcome the Von Neumann bottleneck of existing computing methods.

This structure enables a transient charge trapping/detrapping mechanism, mirroring the temporary conductivity changes in brain synapses. This process successfully replicates short-term plasticity (STP), a key characteristic of biological synapses. Building upon this, the research team constructed a PRC system and applied it to process surface electromyography (sEMG) signals. The system achieved high accuracy in classifying complex hand movements, including finger flexion (87.0% for the ring finger and 87.4% for the little finger), and also showed strong performance in MNIST digit classification.

Kim Kyung-bin, a researcher and co-first author, managed the device's design and fabrication, alongside performing STP measurements and physical analysis. Dr. Kim Bo-ram, also a co-first author, led the design of the PRC system based on the device's nonlinear dynamic characteristics and conducted simulations for MNIST and sEMG-based gesture recognition. Professor Park Dong-wook described the hybrid device as a "low-power, high-efficiency solution" that combines the strengths of organic and inorganic materials, potentially overcoming the Von Neumann bottleneck in current computing architectures. Professor Kim Yun expressed optimism, stating, "Having demonstrated the ability to effectively process complex, real-time biological signals, this will become a core foundational technology for next-generation flexible neuromorphic hardware in areas like edge computing, real-time signal analysis, and adaptive sensing systems."

“

Having demonstrated the ability to effectively process complex, real-time biological signals, this will become a core foundational technology for next-generation flexible neuromorphic hardware in areas like edge computing, real-time signal analysis, and adaptive sensing systems.