A team of researchers from MIT and elsewhere has developed a new photonic chip that can perform all key computations of a deep neural network optically on the chip. The device was able to complete machine-learning classification tasks in less than half a nanosecond while achieving over 92 percent accuracy, comparable to traditional hardware.
The photonic chip is composed of interconnected modules that form an optical neural network and is fabricated using commercial foundry processes, which could enable scaling and integration into electronics. This technology has the potential to lead to faster and more energy-efficient deep learning for applications like lidar, scientific research in astronomy and particle physics, or high-speed telecommunications.
The researchers built an optical deep neural network on a photonic chip using three layers of devices that perform linear and nonlinear operations. The system encodes parameters of a deep neural network into light and uses programmable beamsplitters to perform matrix multiplication, followed by nonlinear optical function units (NOFUs) that implement nonlinear functions.
The device achieves ultra-low latency by staying in the optical domain until the end when reading out the answer. This enables efficient training of deep neural networks on the chip, a process known as in situ training that typically consumes significant energy in digital hardware.
Future work will focus on scaling up the device and integrating it with real-world electronics like cameras or telecommunications systems. The researchers also aim to explore algorithms that can leverage the advantages of optics to train systems faster and more energy-efficiently.
Source: https://news.mit.edu/2024/photonic-processor-could-enable-ultrafast-ai-computations-1202