MIT Develops Photonic Processor for Ultra-Fast 6G Wireless Signal Processing

As the number of connected devices continues to surge, the demand for wireless bandwidth is skyrocketing. Managing the finite wireless spectrum effectively to support tasks like teleworking and cloud computing has become increasingly challenging. Engineers are turning to artificial intelligence (AI) to dynamically handle this spectrum, aiming to reduce latency and boost overall performance. However, traditional AI methods are often power-intensive and cannot operate in real-time. To address these issues, researchers at MIT have developed a novel photonic processor that performs machine-learning computations at lightning-fast speeds, specifically designed for wireless signal processing. The photonic chip, which operates entirely in the frequency domain before the signals are digitized, is approximately 100 times faster than top-tier digital alternatives and achieves about 95% accuracy in signal classification. This breakthrough could have significant implications for future 6G wireless technologies, particularly in cognitive radios that adapt modulation formats to optimize data rates based on the changing wireless environment. The core of this innovation lies in the multiplicative analog frequency transform optical neural network (MAFT-ONN). Unlike conventional digital processors, MAFT-ONN encodes and processes data using light, making it lighter, cheaper, and more energy-efficient. The MAFT-ONN's unique architecture performs both linear and nonlinear operations in-line, reducing the need for multiple devices per layer. This design allows the researchers to fit 10,000 neurons onto a single device, significantly enhancing computational efficiency. One of the primary challenges in designing MAFT-ONN was mapping machine-learning computations to the optical hardware. The team had to customize existing frameworks and leverage the physics of light to achieve their goals. Despite these hurdles, their simulations showed promising results, with the optical neural network achieving 85% accuracy in a single shot and converging to over 99% accuracy with multiple measurements. Additionally, the entire process took only about 120 nanoseconds, far surpassing the performance of current digital devices, which typically require microseconds. Ronald Davis III, the lead author and recent MIT PhD graduate, emphasizes the potential of MAFT-ONN to revolutionize edge computing. By enabling real-time deep-learning computations, MAFT-ONN could enhance the capabilities of various applications, including autonomous vehicles, which need to react instantly to changing environments, and smart pacemakers, which must continually monitor patient health. "The ability of MAFT-ONN to perform inferences in nanoseconds without significant loss of accuracy opens up a plethora of possibilities for real-time and reliable AI inference," Davis says. "This could truly transform the way we handle complex tasks in edge devices." Dirk Englund, a professor at MIT and senior author of the paper, highlights the broader impact of this research. The scalable and flexible nature of the photonic processor means it can be adapted for a wide range of high-performance computing applications. The team also plans to explore more advanced deep learning architectures, such as transformer models and large language models (LLMs). "This work marks the beginning of a new era in AI hardware accelerators, where optics play a crucial role in achieving unprecedented speed and efficiency," Englund states. "With further advancements, MAFT-ONN could pave the way for next-generation wireless communication systems and beyond." Industry insiders are enthusiastic about the potential of photonic processors in addressing the growing demands of wireless communication. According to Ryan Hamerly, a visiting scientist at MIT and senior scientist at NTT Research, the development represents a significant step forward in integrating AI into wireless infrastructure. The chip's performance, combined with its low cost and energy efficiency, could make it a game-changer for 6G and beyond. The MIT Quantum Photonics and Artificial Intelligence Group, led by Englund, has been at the forefront of this research. Their interdisciplinary approach, combining expertise in quantum photonics and AI, has yielded innovative solutions like MAFT-ONN, positioning them as key players in the development of advanced computing technologies. Funding for this research came from a variety of sources, including the U.S. Army Research Laboratory, the U.S. Air Force, MIT Lincoln Laboratory, Nippon Telegraph and Telephone, and the National Science Foundation. These collaborations underscore the broad interest and potential applications of photonic processors in both academic and industrial sectors.