UCLA electrical and computer engineers have developed a physical artificial neural network that analyzes immense data and identifies objects at the speed of light. The device was created with a 3D printer at the UCLA Samueli School of Engineering.
The network, composed of a series of polymer layers, works using light that travels through it. Each layer is 8 centimeters square. (Source: UCLA Samueli / Ozcan Research Group)
There are many devices out there that use computerized cameras to identify objects. But this new device uses other imaging equipment, including optical sensors, cameras, processors and computing programs to process data while seeing and identifying objects.
The new device uses a diffractive deep neural network to operate. Light bounces from an object and back to the device to identify it as quickly as possible. The device doesn’t require advanced computing programs to actually process the images. It also doesn’t consume any energy when running — it only uses the diffraction of light.
New technologies that use this device could speed up the operation of data-intensive tasks that sort and identify objects. For example, this technology could be used in medical devices and autonomous cars.
"This work opens up fundamentally new opportunities to use an artificial intelligence (AI)-based passive device to instantaneously analyze 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 modeled 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."
Developing the device started with developing a computer simulated design that was then 3D printed. The design is a thin, 8 cm square of polymer wafers. These wafers have an uneven surface that diffracts light coming from the object and bounces it into different directions. The layers of the square may look opaque to the naked eye, but submillimeter-wavelength terahertz frequencies of light that can travel right through them. Each layer is made of over 10,000 artificial neurons.
The optical network in the device is made of a series of pixelated layers. These layers shape how the incoming light from the objects travels through the layers. The network then identifies an object based on how the light passes through. The network is trained with a computer and deep learning to identify objects that it is analyzing, using a learning pattern of diffracted light.
"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."
During testing, the device could accurately identify items of clothing and handwritten numbers. Images were placed in front of a terahertz light source and the device could see the images by using optical diffraction.
The AI device can be made even bigger with additional layers because it is easy to produce through 3D printing. Making the device bigger could allow it to identify more objects at the same time with a more complex data analysis. The components of the device are all inexpensive and the device can be produced for less than $50.
The paper on the new device was published in the journal Science.
