NSF Award
2439922
Nontechnical Description:<br/>Optical metamaterials are a class of materials that leverage features smaller than the wavelength of light to engineer their optical transmission characteristics. Optical metamaterials have already enabled transformative applications, including flat lenses, optical cloaking, and super-resolution imaging. Due to these broad successes, researchers are now developing active optical metamaterials: metamaterials whose optical transmission characteristics can be changed on the fly, for example, at high speeds (gigahertz) enabled by electrical circuits used to control them. In this project, we aim to leverage a new class of active metamaterials to improve the energy efficiency of large-scale computing applications running in data centers, such as training large language models (LLMs) for artificial intelligence (AI),and performing scientific computing workloads. Specifically, we will leverage active metamaterials to improve the energy efficiency of optical data communication in datacenters, by using electrical control signals for high-speed reconfigurable communication between large networks of computing and memory systems. For many of today’s data centers, the overheads of data communication limit the overall power and performance of computation, often referred to as the “communication wall” or “memory wall”. Using active metamaterials to improve energy efficiency of communication, will directly translate into energy efficiency benefits for humanity’s largest computing applications. <br/><br/>Technical Description:<br/>We propose a disruptive technology for high-performance optical switches based on optical active metamaterials – optical metamaterials whose transmission characteristics can be electrically modulated – for improving the energy efficiency of data communication in large-scale data centers. We are targeting communication in data centers, since overall computing performance is often limited by data communication overheads within the large networks of sub-systems that comprise today’s data centers, including racks of general-purpose processors, application-specific hardware accelerators, and memories. Thus, improving the energy efficiency of communication directly improves the energy efficiency of computing.<br/>In this project, we will explore the benefits of active metamaterials to enable a new class of energy-efficient Active Metamaterial Optical Switches (AMOS), targeting high-performance communication and computation in data centers. We will explore three areas of focus: (1) Design and simulation of AMOS devices, considering three potential device structures (details in the full proposal), which we will compare based on their relative power, performance, and area. (2) Experimental fabrication of AMOS devices in our cleanroom (Harvard’s Center for Nanoscale Systems),and experimental measurements to calibrate our device models. (3) System-level projections to quantify the benefits of our AMOS devices for overall power consumption and execution time of computing applications in data centers. Importantly, our analysis will account for interactions between AMOS devices and the power/performance overheads of electrical circuits required to modulate them. This is essential for realistic performance projections of real-world applications. If successful, AMOS devices will enable strictly better trade-offs in switching time, latency, bandwidth, power consumption, and physical size of optical network switches. The resulting energy efficiency benefits of communication will translate directly into energy efficiency benefits of computation for large-scale data centers employing AMOS devices.<br/><br/>This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
