New Light-Powered Computer Chip Could Make AI Smarter and Smaller

Researchers have developed an electronic chip that mimics the way the human brain processes visual information, combining AI algorithms with the hardware necessary for capturing images.

An international team of researchers from the United States, China, and Australia have collaborated on a new electronic chip designed to enhance artificial intelligence by combining sophisticated software and hardware in a tiny device powered by light. The research was led by RMIT University and recently published in the journal Advanced Materials.

The prototype device created by the research team is on the nanoscale, and it integrates AI software with computer imaging hardware thanks to the use of metamaterials that alter how the chip responds to light. With further refinement, the technology used to create this could serve as the underpinning for even smaller and more smart devices, as well as drones and robots.

According to RMIT Associate Professor Sumeer Walia, the new chip prototype enables brain-like functionality by combining modular components into a complex system.

“Our new technology radically boosts efficiency and accuracy by bringing multiple components and functionalities into a single platform,” explained Walia via RMIT news. “It’s getting us closer to an all-in-one AI device inspired by nature’s greatest computing innovation – the human brain.”

According to Walia, the goal of the research team is to emulate one of the primary ways the brain learns – the encoding of visual information as memories. While there is still a substantial amount of work left to do, the prototype created by the research team represents a big step towards improved human-machine interaction, scalable bionic systems, and neurobiotics.

Most commercial level AI applications rely on off-site software and data processing, leveraging cloud computing. In order to make on-site applications more powerful and reliable, the prototype chip integrates intelligence and hardware together in an example of edge AI. Devices like autonomous vehicles and drones must be able to process a large amount of data onsite, making them ideal use cases for technology like the new chip prototype. Walia explained that a dash-cam in a car, loaded with the neuro-inspired hardware the researchers developed, could recognize lights, other vehicles, signs, pedestrians, plants, and more. According to Walia, it’s possible that the chip can deliver “unprecedented levels of efficiency and speed in autonomous and AI-driven decision-making.”

The technology that the prototype employs is based on earlier chips developed by RMIT researchers. These earlier prototypes made use of light to build and modify “memories”. The new features created by the research team means that the chip can automatically captures images, manipulate images, and train machine learning models that recognize objects with over 90% accuracy.

The prototype chip’s design was influenced by optogenetic technology. Optogenetics refers to emerging biotechnology tools that enable scientists to manipulate neurons with prevision using light. The AI chip developed by the RMIT team makes use of black phosphorous, a semiconducting material. Black phosphorous is extremely thin and it changes its electrical resistance as wavelengths of light change. As different wavelengths of light are shined on the material, the material changes its properties, becoming useful for different functions like memory storage and imaging. As the lead author on the study, Dr. Taimor Ahmed of RMIT, explained, light-based computing systems are less energy-intensive, more accurate, and faster than traditional computing methods.

According to Ahmed, the benefit of combining modular systems into one nanoscale device is that AI systems and machine learning algorithms can be used in smaller devices. As an example, Ahmed explained that scientists could miniaturize the technology they developed to enhance artificial retinas and improve the accuracy of bionic eyes.

“Our prototype is a significant advance towards the ultimate in electronics: a brain-on-a-chip that can learn from its environment just like we do,” said Ahmed.

The prototype chip has been designed with easy integration with other technologies and existing electronics in mind.