As more connected devices demand an increasing amount of bandwidth for tasks like teleworking and cloud computing, it will become challenging to manage the amount of wireless spectrum available for all users to share. Against this backdrop, MIT researchers have developed an AI hardware accelerator that is specifically designed for wireless signal processing. The researchers’ optical processor performs machine-learning computations at the speed of light, classifying wireless signals in a matter of nanoseconds. The current class of sophisticated digital AI accelerators for wireless signal processing convert the signal into an image and run it through a deep-learning model to classify it. While this approach is highly accurate, the computationally intensive nature of deep neural networks makes it infeasible for many time-sensitive applications. Optical systems can accelerate deep neural networks, by encoding and processing data using light, which is also less energy-intensive than digital computing. But researchers have struggled to maximize the performance of general-purpose optical neural networks when used for signal processing, while ensuring that the optical device is scalable. To combat this issue, the MIT researchers developed an optical neural network architecture specifically for signal processing, which they call a multiplicative analog frequency transform optical neural network (MAFT-ONN). The MAFT-ONN addresses the problem of scalability by encoding all signal data and performing all machine-learning operations within the frequency domain, before the wireless signals are digitized. Artist’s interpretation of an optical processor for an edge device, developed by MIT researchers, that performs machine learning computations at the speed of light, classifying wireless signals in a matter of nanoseconds. Courtesy of MIT. The researchers designed their optical neural network to perform all linear and nonlinear operations in-line. Both types of operations are required for deep learning. Only one MAFT-ONN device is needed per layer for the entire optical neural network, as opposed to other methods, which require one device for each computational unit, or “neuron.” According to the researchers, they can fit 10,000 neurons onto a single device and compute the necessary multiplications in a single shot. The researchers accomplished this feat using a technique called photoelectric multiplication, which dramatically boosts efficiency. It also allowed them to create an optical neural network that can be readily scaled up with additional layers without requiring extra overhead. MAFT-ONN takes a wireless signal as input, processes the signal data, and passes the information along for later operations that the edge device performs. For example, by classifying a signal’s modulation, MAFT-ONN would enable a device to automatically infer the type of signal to extract the data it carries. One of the biggest challenges the researchers faced when designing MAFT-ONN was determining how to map the machine-learning computations to the optical hardware. When they tested their architecture on signal classification in simulations, the optical neural network achieved 85% accuracy in a single shot, which can quickly converge to >99% accuracy using multiple measurements. MAFT-ONN only requires ~120 ns to perform the entire process. Currently, according to the researchers, the photonic chip is approximately 100x times faster than the best digital alternative, while converging to about 95% percent accuracy in signal classification. Moving forward, the researchers want to use multiplexing schemes, to perform more computations and scale up the MAFT-ONN. They also want to extend their work into more complex deep learning architectures that could run transformer or large-language models. Already, the researchers have shown that the device could be useful in future 6G wireless applications, such as cognitive radios that optimize data rates by adapting wireless modulation formats to the changing wireless environment. Since the hardware accelerator is scalable and flexible, so it could be used for a variety of high-performance computing applications, as well as in/for autonomous vehicles and smart pacemakers, for example. Importantly, it is smaller, lighter, cheaper, and more energy-efficient than digital AI hardware accelerators. “What we’ve presented in our paper could open up many possibilities for real-time and reliable AI inference. This work is the beginning of something that could be quite impactful,” said Dirk Englund, a professor in the MIT Department of Electrical Engineering and Computer Science, principal investigator in the Quantum Photonics and Artificial Intelligence Group and the Research Laboratory of Electronics (RLE), and senior author of the paper describing the work. The work was funded, in part, by the U.S. Army Research Laboratory, the U.S. Air Force, MIT Lincoln Laboratory, Nippon Telegraph and Telephone, and the National Science Foundation. The research is published in Science Advances (www.doi.10.1126/sciadv.adt3558).