Modern computers struggle to match the efficiency of the human brain. Even recognizing handwritten digits can consume significant energy, largely because conventional computers separate memory from processing. Biological neurons, by contrast, sense, compute, and store information in the same physical structure with remarkably little power.
In International Journal of Extreme Manufacturing , researchers report a neuromorphic chip that takes a very different approach: it processes and learns information using light and electronics tightly integrated on a single platform, much like biological neurons combine sensing, communication, and memory in one physical structure.
Their bio-inspired neuron platform can operate at low voltages (~±1 V), with high on/off ratios (up to 10⁶) and extremely low subthreshold swings (~78 mV·dec⁻¹), while maintaining excellent device stability over 1,000 switching cycles, and can achieve an image recognition accuracy of 92.02% in simulations.
"Our goal was to move away from fragmented architectures and toward a system where signal generation, transmission, and learning all happen together," the corresponding author Prof. Jianwen Zhao at Suzhou Institute of Nano-Tech and Nano-Bionics, Chinese Academy of Sciences, explained. "This is much closer to how real neural systems operate."
The core of their platform is a monolithic system built from single-walled carbon nanotube thin-film transistors coupled with miniature light-emitting diodes (Mini-LEDs). Together, these components form a closed electrical–optical–electrical loop: electrical signals generate light, light modifies electronic behavior, and the resulting electrical signals carry information forward, much like communication between biological neurons.
Notably, the same carbon-nanotube transistor performs two critical roles at once. It acts as a driver for the LEDs, controlling light with low voltages of around 1 volt, and it also behaves as an artificial synapse, responding to the light it produces to simulate learning behaviors. "Combining these functions in a single device allows us to greatly simplify the circuit while improving efficiency," Prof. Zhao noted.
Unlike many optoelectronic neuromorphic systems, which rely on bulky external light sources, this platform generates and processes optical signals entirely on chip. Their devices were fabricated at wafer scale using semiconductor-compatible processes, with micrometer-sized features and operating voltages around one volt. When paired with an organic semiconductor layer, the transistors respond across most of the visible spectrum, enabling them to process optically encoded signals generated directly on the chip.
Using this integrated system, Prof. Zhao's team demonstrated key synaptic behaviors found in biological learning, including short-term and long-term changes in signal strength. Based on the measured device characteristics, they simulated a five-layer convolutional neural network that achieved over 92% accuracy on standard image-recognition tasks and around 86% on handwritten digits, without relying on external memory or computing units.
While the system is not yet a complete artificial brain, it points toward a future in which sensing, computing, and memory are deeply unified. With further integration and scaling, this approach could enable compact, energy-efficient hardware for neuromorphic computing, adaptive electronics, and next-generation human-machine interfaces that learn and respond quickly, efficiently, and with minimal power, more like the brain itself.
International Journal of Extreme Manufacturing (IJEM, IF: 21.3) is dedicated to publishing the best research related to the science and technology of manufacturing functional devices and systems with extreme dimensions (extremely large or small) and/or extreme functionalities
- Maintained #1 in Engineering, Manufacturing for consecutive years
- Average time to First Decision after Peer Review: 34 days
- Open Access Publishing after Double-Anonymous Peer Review
Visit our webpage , like us on Facebook , and follow us on Twitter
