Collaborative Research: ASCENT: Optically-Accelerated Heterogeneous AI Computing Chiplet (OCTANT)

Regents Of The University Of Michigan - Ann Arbor, Ann Arbor MI

Investigators

Abstract

Nontechnical Description The rapid rise of generative artificial intelligence (AI) has ushered society into a new era of supercomputing-driven data exploration. This tipping point is intensifying the gap between the exploding size of AI models and the limited computing throughput available today. A major bottleneck lies in data movement, specifically, the limitations of current interconnect technologies, further constrained by the aging von Neumann architecture. To address this, this project will employ a co-designed approach spanning architecture, packaging, and device innovation to meet the demands of next-generation interconnects, including high bandwidth, low energy use, low latency, scalability, and reliability. Supported by the ASCENT program, this project introduces a novel 3.5D integrated photonic interconnect solution that combines breakthroughs in network architecture, photonic devices, and advanced packaging. By leveraging the complementary expertise of academic and industry collaborators across several ECCS clusters, this effort drives interdisciplinary innovation, trains the next generation of engineers, and enables more powerful, efficient, and scalable computing systems that benefit society. Technical Description This project advances the field of integrated photonics by introducing a co-designed solution across architecture, packaging, and devices to meet the demands of next-generation computing. It proposes a transformative photonic interconnect-switching architecture based on a novel wavelength-mode division multiplexing scheme, enabled by athermal, energy-efficient, high-speed modulation and advanced hybrid Cu-Cu bonding techniques in 2.5D/3.5D integration. The goal is to achieve terabit-per-second data transmission with dynamic AI workload optimization. Key research tasks include the design of a reconfigurable and resource-aware photonic interposer network, exploration of 2.5D/3.5D network architectures with AI workload analysis, fabrication of heterogeneous athermal capacitive modulators and switches, development of microring-based transceiver and switch testbeds, and advancement of hybrid Cu-Cu bonding technology. This tightly integrated, interdisciplinary effort will drive innovation across multiple fronts, directly aligning with the core mission and priorities of the ASCENT program. 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.