Aug. 30, 2021 — Optical technology is used for transmitting, storing, displaying and identifying data. It provides the processing speed that data centers need by offering efficient means for communication and analysis operations. The technology comes at a time when the massive scale of today’s datasets is straining the capacity of digital and electronic computers to compile them and extract key information. The research community has a strong interest in optical-based information processing for performing the high-speed calculations necessary in machine learning tasks.
Source: Alain Herzog, EPFL
“Light transmits information without any physical interference from cables. That’s the core advantage of optical technology when it comes to transferring data,” says Demetri Psaltis, head of EPFL’s Optics Laboratory within the School of Engineering. “To take artificial intelligence as an example, many AI programs require accelerators to carry out rapid calculations using minimal power. For now, while optical technology could theoretically meet that need, it has not yet reached the applied stage – despite a half-century of research. That’s because optical computing and decision-making do not yet save either time or energy.”
Inspired by Neural Networks
Designing optical computing devices remains a challenge. Although the computations are performed rapidly, the obstacle comes in transferring the results to memory at that same speed and in an energy-efficient manner. This obstacle is what the engineers at Psaltis’ lab, along with colleges at Christophe Moser’s Laboratory of Applied Photonic Devices, also within the School of Engineering, decided to address. They developed a machine learning method named SOLO, for Scalable Optical Learning Operator, that can recognize and classify information formatted as two-dimensional images. Their findings have recently been published in Nature Computational Science.
Computer scientists have designed electronic computers with inspiration from the brain’s neural networks. These machines work by using neuron-like processors and the connections between those neurons. The networks are constructed in layers, and it is these layers that create the processing power. The more layers, the more sophisticated the computer’s decision-making ability. In 1990, these networks were one layer deep representing 1 million neural connections. Today, the most powerful networks contain tens of layers and billions of connections. This is a technological triumph, but the large number of connections consume a large amount of energy.
“The goal of our research is to reduce the energy requirement by using other processing methods, in particular photonics,” says Moser. His team, therefore, looked at using optical fibers to perform certain calculations. “The calculations are executed automatically by the propagation of light pulses inside the fiber. This simplifies the computer’s architecture, retaining only a single neuronal layer, making it a hybrid system,” adds Ugur Tegin, the lead co-author of the work.
Cutting the Power Requirement by a Factor of 100
To test their technology, the team used a dataset consisting of X-ray images of lungs affected by various diseases, including COVID-19. They then ran the data through SOLO to identify the organs affected by the coronavirus. For the purposes of comparison, they also ran the data through a conventional artificial intelligence system with 25 layers of neurons. What did they find? “Both systems classified the X-rays equally well. However, our system consumed 100 times less energy,” says Moser. That marked the first time engineers were able to demonstrate quantified power savings. SOLO’s greater energy efficiency could also open the door to new opportunities in other areas of ultra-fast optical computing.
Hybrid optical computing systems are emerging as a promising new technology. “They combine the bandwidth and speed of optical processing with the flexibility of electronic computing. When coupled with artificial intelligence programs in robotics, microscopy and other visual computing tasks, these hybrid systems could achieve some of the transformative capabilities that were, for a long time, imagined as the sole purview of optical computers,” says Psaltis.
Read the article published in Nature Computational Science here.