Chinese scientists from Fudan University have developed artificial neural cells that closely mimic living brain cells.

Their device combines an ultra-thin monolayer of molybdenum disulfide (MoS₂) with dynamic random-access memory (DRAM), allowing not only the simulation of neural signal transmission but also adaptive plasticity, meaning the ability of neural connections to adjust their strength based on experience.

This innovation comes amid increasing demand for devices with higher performance and energy efficiency due to the rapid development of artificial intelligence and machine learning systems. Since machine learning algorithms are inspired by biological neural networks, engineers aim to replicate the structure and function of the human brain, making these components based on neural principles worthy of the term “artificial neural cells.” They are connected by dynamic links whose strength changes over time, simulating the brain’s ability to learn and adapt.

The innovation relies on two main elements:

• DRAM cells: to store electric charge that mimics the membrane potential of biological neurons.
• The inverter: to switch signals and create pulses resembling “neural spikes.”

To test the innovation’s potential, researchers built a small network of 3×3 artificial neural cells and evaluated its ability to adapt its response to input data based on incident light, simulating human vision under different lighting conditions. They also used their system to run an image recognition model and assess its performance.

Results showed that these artificial neural cells are very promising, especially in implementing energy-efficient models for computer vision and image recognition tasks. Researchers plan to build new nature-inspired computing systems to apply this technology in various other tasks.

This step marks significant progress in developing energy-efficient artificial neural systems capable of learning and adapting to different lighting, paving the way for more advanced technologies in machine vision and pattern recognition.