Optical neural network identifies objects at speed of light

Engineers at UCLA have 3D printed a device that uses neural networks and machine learning to identify objects purely through light.  

The network, composed of a series of polymer layers, works using light that travels through it (Credit: UCLA Samueli / Ozcan Research Group)

Unlike typical machine vision systems that use cameras and imaging software, the new device does not convert light into data for processing. Called a “diffractive deep neural network,” it uses the light bouncing from the object itself to identify it in the same time it would take a computer to simply “see” the object. Essentially operating at the speed of light, it has many potential applications in areas such as instantaneous vision systems for autonomous vehicles and medical diagnostics. The research is published in Science.  

“This work opens up fundamentally new opportunities to use an artificial intelligence-based passive device to instantaneously analyse data, images and classify objects,” said Aydogan Ozcan, the study’s principal investigator and the UCLA Chancellor’s Professor of Electrical and Computer Engineering. “This optical artificial neural network device is intuitively modelled on how the brain processes information. It could be scaled up to enable new camera designs and unique optical components that work passively in medical technologies, robotics, security or any application where image and video data are essential.”

To create the device, the UCLA team used a 3D printer to create very thin, eight centimetre-square polymer wafers. Each wafer has uneven surfaces, which help diffract light coming from the object in different directions. While opaque to the eye, submillimetre-wavelength terahertz frequencies of light can travel through the wafers. Each layer is composed of tiny pixels that the light penetrates, essentially acting like many thousands of artificial neurons.

The series of pixelated layers functions as an “optical network” that shapes how incoming light from the object travels through them. The network identifies an object because the light is mostly diffracted towards a single pixel that is assigned to that type of object. The researchers then used a computer to train the network to identify the objects in front of it by learning the pattern of diffracted light from each object.

“This is intuitively like a very complex maze of glass and mirrors,” Ozcan said. “The light enters a diffractive network and bounces around the maze until it exits. The system determines what the object is by where most of the light ends up exiting.”

Costing less than $50 to create, the device was able to accurately identify handwritten numbers and items of clothing, both of which are commonly used tests in artificial intelligence and machine vision studies. According to Ozcan, it would also be possible to create neural networks that use visible, infrared or other frequencies of light.