RMIT Researchers Develop All-in-One Neuromorphic Vision Chip for Bionic Eyes and Low-Power AI

Researchers at RMIT University have developed a prototype chip that integrates visual sensing, memory storage, and data processing into a single hardware architecture, laying potential groundwork for energy-efficient smart bionic eyes and artificial intelligence systems. As first reported by TechXplore, the neuromorphic vision technology mimics biological sight by handling computational workloads directly at the point of capture rather than offloading raw data to external servers or separate processor units.

The technology relies on molybdenum disulfide (MoS₂), a semiconductor material that is only an atom thick. The physical components responsible for sensing, computing, and retaining data are embedded on a chip measuring 2 centimeters by 2 centimeters (0.8 inches by 0.8 inches). This chip operates within a larger prototype casing measuring 15 centimeters by 14 centimeters by 3 centimeters (5.9 inches by 5.5 inches by 1.2 inches), which houses the requisite reading, processing, and communication circuitry.

Standard digital imaging systems record every video frame and transmit massive streams of uncompressed data to external memory banks and central processing units for analysis. By contrast, the RMIT prototype performs pattern recognition, motion tracking, and change detection locally on the sensor itself. In laboratory demonstrations, the system successfully recognized numbers and geometric shapes while identifying dynamic shifts in visual scenes.

The research initiative was spearheaded by Professor Sumeet Walia at RMIT's Centre for Opto-electronic Materials and Sensors (COMAS), alongside co-researcher Dr. Taimur Ahmed, an expert in neuromorphic vision devices. RMIT has formally submitted an international patent application for the innovation under the Patent Cooperation Treaty (PCT).

"Nature has already solved many of the challenges we're trying to address in electronics," Walia stated regarding the system's design philosophy. "The human eye and brain work together incredibly efficiently, processing vast amounts of information using remarkably little energy. Our research is helping lay the foundations for technologies that work in a more similar way."

Explaining the technical functionality, Ahmed emphasized that the integrated architecture eliminates the latency and power penalties associated with constant data movement. "This is not just a sensor that captures information, it's a sensor that can also process information," Ahmed said. "Rather than constantly moving data between separate memory and processing units, much of that work happens much closer to where the information is generated."

A core hurdle in deploying ultra-thin semiconductor devices has been the manufacturing process, which often introduces structural flaws during material transfer. In a study published in the journal ACS Applied Materials & Interfaces by lead author Jianfeng Mao and colleagues, the RMIT team detailed a novel lithography-free, water-based transfer technique utilizing polyvinyl alcohol (PVA). This cleaner fabrication method transfers atom-thin molybdenum disulfide and gold (Au) electrodes with substantially fewer defects, delivering higher electrical output and improved light sensitivity.

While commercial deployment of smart bionic vision remains years away, the researchers indicated that the architecture could find earlier application in industrial machine vision, autonomous vehicles, robotics, and distributed edge computing sensors. Furthermore, the technology offers a potential mitigation strategy for the escalating power consumption of cloud artificial intelligence infrastructure by drastically cutting data transmission requirements.

"This work combines advanced materials, engineering and artificial intelligence to address one of the defining challenges of our time: creating intelligent systems that are both powerful and sustainable," Walia said. He noted that if the RMIT technology can be scaled up, it could help reduce the volume of data that needs to be moved, stored, and processed, making future AI systems significantly more energy-efficient.