Dongguk University Researchers Develop Battery-Free Artificial Synapse for Flexible Electronics

Researchers in South Korea have developed a flexible, self-powered neuromorphic device that combines tactile sensing and memory functions on a single substrate, offering a potential path toward battery-free wearable electronics. The research, published in Advanced Materials and reported by TechXplore (https://techxplore.com/news/2026-09-powered-artificial-synapse-combines-memory.html), couples a graphene-channel ion-gel-gated transistor with motion-driven nanogenerators to eliminate the need for an external power source.

Neuromorphic devices aim to emulate biological neural networks for low-power edge computing. While graphene-channel ion-gel-gated transistors (g-IGTs) operate at low voltages and can modulate synaptic weights to mimic biological synapses, conventional configurations require external electrical power. That constraint has limited their deployment in continuous, standalone wearable monitoring.

To overcome that limitation, a research team led by Professor Sejoon Lee in the Department of System Semiconductor at Dongguk University integrated the g-IGT with triboelectric nanogenerators (TENGs). TENGs harvest mechanical energy from physical touch, motion, or vibration and convert it directly into electrical spikes.

The system mimics human tactile mechanoreceptors, which translate skin deformation into nerve impulses processed at synapses. In the Dongguk team's design, two TENGs connect to a single g-IGT: a gate-side TENG delivers presynaptic spikes, while a drain-side TENG provides postsynaptic spikes. Voltage pulses generated during physical contact regulate the transistor's synaptic behavior entirely on harvested energy.

In laboratory testing, the device demonstrated biological-style hierarchical memory retention. It exhibited sensory memory with a decay time of roughly 70 milliseconds and short-term memory lasting 0.2 to 0.45 seconds. Repeated mechanical stimulation shifted the state into long-term memory, extending retention past 2 seconds. The transistor also exhibited spike-rate-dependent plasticity—adjusting synaptic connection strength according to spike frequency—and maintained that capability under mechanical bending.

The researchers evaluated the system's learning capability by feeding experimentally measured synaptic responses into a single-layer artificial neural network configured for human activity recognition. Using public motion datasets, the bent, TENG-driven system classified six activities—walking, sitting, standing, lying, climbing stairs, and descending stairs—with 88.05% accuracy. Under high-noise test conditions, classification accuracy remained above 75%, though extreme distortion reduced performance.

Potential applications include self-powered electronic skin, smart prosthetics, continuous health monitors, and human-machine interfaces. Lee noted that the architecture demonstrates that sensing, memory, learning, and processing can be united on a flexible, battery-free platform.