Optical I/O for Next-Generation High-Performance and AI Computing
The exponential growth of high-performance computing (HPC) ,artificial intelligence (AI) and machine learning (ML) has shifted the system bottleneck from compute to communication. Modern accelerators and GPUs deliver unprecedented computational density, yet their performance and energy efficiency are increasingly constrained by the electrical interconnects that link processors, memory, and storage. Optical input/output (I/O) has emerged as the most promising technology to overcome these barriers, offering orders-of-magnitude improvements in bandwidth density, latency, and energy per bit. This special issue will serve as a platform to consolidate and advance the state of knowledge on optical I/O technologies for HPC, AI, and ML. By bringing together contributions from both academia and industry, it will highlight innovations that enable scalable, energy-efficient, and disaggregated computing architectures. The issue will also showcase cross-disciplinary approaches that span device physics, integrated photonics, photonic-electronic integration, packaging, system architecture, and HPC/AI/ML applications
Topics covered include, but are not limited to:
- Fundamental Devices and Components 1. Devices and components: Comb laser sources and integrations, Wavelength-division multiplexing (WDM), high-speed modulators, detectors, and optical amplifiers, etc
- Photonic-Electronic Integration 1. Photonic–electronic co-designed integrated circuits targeting low-latency, bandwidth-intensive AI accelerators
- Photonic Interconnects & Packaging 1. Co-packaged optics to alleviate chip-to-chip and GPU interconnect bottlenecks in AI clusters and supercomputers 2. Disaggregated architectures enabled by in-package optical I/O for composable infrastructure in HPC and AI
- Architectural Innovations 1. Disaggregated and composable computing architectures enabled by optical I/O 2. Photonic interconnects for GPU/TPU clusters and memory fabrics
- Optical Switching & Processing 1. Optical switching & processing units for routing infrastructure in HPC and AI
- System-Level Performance and Applications 1. Energy efficiency, scalability, and latency benchmarks of optical I/O in HPC and AI 2. Deployment perspectives from data centers and emerging supercomputing platforms
- Industrial Perspectives and Future Outlook 1. Insights from leading companies developing photonic interconnects and optical processors (Lightmatter, Nvidia, Ayalab, Broadcom, Ranovus, Intel, etc) 2. Roadmaps toward standardization, manufacturability, and ecosystem adoption(from Industry or Academia)
Guest Editors
Songtao Liu (Intel, USA)
Yuan Yuan (Northeastern Univeristy, USA)
Yating Wan (KAUST, KSA)