Zhenguo Wu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Zhenguo Wu has a total of 22 co-authors. Here are the top co-authors by number of shared publications:
- Keren Bergman
Columbia University
5 shared publications - Asher Novick
Department of Electrical Engineering, Columbia University, New York, NY, USA
2 shared publications Liang Yuan Dai
Department of Electrical Engineering, Columbia Universityhttps://ror.org/00hj8s172, 500 W 120th St., New York, New York 10027, USA
2 shared publications- Yuyang Wang
Columbia University, New York, NY, USA
2 shared publications - Aneek James
University of Georgia, United States
1 shared publication - Anthony Rizzo
Department of Electrical Engineering, Columbia University, New York, NY, USA
1 shared publication
Latest Publications
- Open Access
- 5 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Micro
- Vol. PP (02721732) , 1-10
SiPAM: Silicon Photonic Accelerated Memory Pooling Architecture
Training and inference of large-scale machine learning models demand significant processing power, memory capacity, and memory bandwidth. Current compute units are equipped with a fixed amount of high-bandwidth memory (HBM) modules placed near the compute die which limits the system’s flexibility to adapt to different compute and memory demands. In this work, we present SiPAM, a silicon photonic accelerated memory pooling architecture that integrates dense wavelength-division multiplexing-based photonic I/Os directly along the compute die’s shoreline. This design creates a unified high-bandwidth communication domain that decouples memory and network scaling from packaging constraints and enables demand-driven dynamic resource allocation. We develop a quantitative optimization methodology to allocate compute and memory resources efficiently, with up to 3.5× faster iteration time and 4.5× reduction in energy consumption compared to state-of-the-art compute systems.
- 5 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Micro
- Vol. PP (02721732) , 1-10
Open AccessSiPAM: Silicon Photonic Accelerated Memory Pooling Architecture
Training and inference of large-scale machine learning models demand significant processing power, memory capacity, and memory bandwidth. Current compute units are equipped with a fixed amount of high-bandwidth memory (HBM) modules placed near the compute die which limits the system’s flexibility to adapt to different compute and memory demands. In this work, we present SiPAM, a silicon photonic accelerated memory pooling architecture that integrates dense wavelength-division multiplexing-based photonic I/Os directly along the compute die’s shoreline. This design creates a unified high-bandwidth communication domain that decouples memory and network scaling from packaging constraints and enables demand-driven dynamic resource allocation. We develop a quantitative optimization methodology to allocate compute and memory resources efficiently, with up to 3.5× faster iteration time and 4.5× reduction in energy consumption compared to state-of-the-art compute systems.
- Open Access
- 15 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 1211-1222
ACTINA: Adapting Circuit-Switching Techniques for AI Networking Architectures
- 15 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 1211-1222
Open AccessACTINA: Adapting Circuit-Switching Techniques for AI Networking Architectures
- 1 January 2025
- Published by Optica Publishing Group
- p. M4H.2
Flexible Photonic Memory Pooling Architecture For Efficient Compute Resource Allocation
We present a photonic memory pooling architecture and a co-designed optimization methodology for flexible allocation of compute, memory, and network resources. We demonstrate up to 5.4× to 7.5× improvements in DL training and inference time.
- 1 January 2025
- Published by Optica Publishing Group
- p. M4H.2
Flexible Photonic Memory Pooling Architecture For Efficient Compute Resource Allocation
We present a photonic memory pooling architecture and a co-designed optimization methodology for flexible allocation of compute, memory, and network resources. We demonstrate up to 5.4× to 7.5× improvements in DL training and inference time.
- 9 January 2024
- Published by Optica Publishing Group in Journal of Optical Communications and Networking
- Vol. 16 (2) , A157-A168
Flexible silicon photonic architecture for accelerating distributed deep learning
The increasing size and complexity of deep learning (DL) models have led to the wide adoption of distributed training methods in datacenters (DCs) and high-performance computing (HPC) systems. However, communication among distributed computing units (CUs) has emerged as a major bottleneck in the training process. In this study, we propose Flex-SiPAC, a flexible silicon photonic accelerated compute cluster designed to accelerate multi-tenant distributed DL training workloads. Flex-SiPAC takes a co-design approach that combines a silicon photonic hardware platform with a tailored collective algorithm, optimized to leverage the unique physical properties of the architecture. The hardware platform integrates a novel wavelength-reconfigurable transceiver design and a micro-resonator-based wavelength-reconfigurable switch, enabling the system to achieve flexible bandwidth steering in the wavelength domain. The collective algorithm is designed to support reconfigurable topologies, enabling efficient all-reduce communications that are commonly used in DL training. The feasibility of the Flex-SiPAC architecture is demonstrated through two testbed experiments. First, an optical testbed experiment demonstrates the flexible routing of wavelengths by shuffling an array of input wavelengths using a custom-designed spatial-wavelength selective switch. Second, a four-GPU testbed running two DL workloads shows a 23% improvement in job completion time compared to a similarly sized leaf-spine topology. We further evaluate Flex-SiPAC using large-scale simulations, which show that Flex-SiPAC is able to reduce the communication time by 26% to 29% compared to state-of-the-art compute clusters under representative collective operations.
- 9 January 2024
- Published by Optica Publishing Group in Journal of Optical Communications and Networking
- Vol. 16 (2) , A157-A168
Flexible silicon photonic architecture for accelerating distributed deep learning
The increasing size and complexity of deep learning (DL) models have led to the wide adoption of distributed training methods in datacenters (DCs) and high-performance computing (HPC) systems. However, communication among distributed computing units (CUs) has emerged as a major bottleneck in the training process. In this study, we propose Flex-SiPAC, a flexible silicon photonic accelerated compute cluster designed to accelerate multi-tenant distributed DL training workloads. Flex-SiPAC takes a co-design approach that combines a silicon photonic hardware platform with a tailored collective algorithm, optimized to leverage the unique physical properties of the architecture. The hardware platform integrates a novel wavelength-reconfigurable transceiver design and a micro-resonator-based wavelength-reconfigurable switch, enabling the system to achieve flexible bandwidth steering in the wavelength domain. The collective algorithm is designed to support reconfigurable topologies, enabling efficient all-reduce communications that are commonly used in DL training. The feasibility of the Flex-SiPAC architecture is demonstrated through two testbed experiments. First, an optical testbed experiment demonstrates the flexible routing of wavelengths by shuffling an array of input wavelengths using a custom-designed spatial-wavelength selective switch. Second, a four-GPU testbed running two DL workloads shows a 23% improvement in job completion time compared to a similarly sized leaf-spine topology. We further evaluate Flex-SiPAC using large-scale simulations, which show that Flex-SiPAC is able to reduce the communication time by 26% to 29% compared to state-of-the-art compute clusters under representative collective operations.
- Open Access
- 15 November 2023
- Published by AIP Publishing in Applied Physics Reviews
- Vol. 10 (4)
High-bandwidth density silicon photonic resonators for energy-efficient optical interconnects
The growth of artificial intelligence applications demands ever larger and more complex deep learning models, dominating today's—and tomorrow's—data center and high-performance computing systems. While traditional electronics are failing to keep pace with application demands, silicon photonic (SiPh) interconnects have emerged as a necessary technology to support these systems. SiPh-driven wavelength-division multiplexing (WDM) offers a particularly promising path toward supporting incredibly high-aggregate link bandwidth in a compact and efficient form factor. One of the basic building blocks of these integrated WDM interconnects is the SiPh resonator. Their inherent wavelength selectivity and compact footprint allow for efficient data transmission multiplexed across dozens of carrier wavelengths. Used as add-drop (AD) filters, SiPh resonators are critical to constructing integrated tunable wavelength-selective optical circuit switches as well as for demultiplexing the different carrier wavelengths toward independent wavelength-insensitive photodiodes in a dense wavelength-division multiplexing receiver. Resonators in the all-pass (AP) configuration are widespread as well, allowing for wavelength-selective modulation to drive aggregate link bandwidths far beyond the individual channel data rate. Unlike SiPh Mach–Zehnder modulators (MZM), resonant modulators can be driven using low, complementary metal-oxide-semiconductor drive voltages, allowing for tight co-integration between photonic integrated circuits, fabricated with larger process node technologies, and electronic integrated circuits, designed to exploit the advantages of the latest node. To push toward practical peta-scale interconnects, a comprehensive review of SiPh resonators is required, addressing bottlenecks and design constraints at both the architecture and device levels. We first describe the predominant integrated link architectures and identify their limits. We then discuss the device-level design considerations that can be made for both AD and AP configuration resonators to overcome the system level limits with novel resonator device designs. Analytical models and numerical simulation of resonators are validated by experimental measurement of devices fabricated in a commercial 300-mm foundry, showing a clear path toward volume manufacturing. The demonstrated resonant modulators and filters support the feasibility of increasing the aggregate bandwidth of resonator-driven SiPh interconnects into the peta-scale regime.
- 15 November 2023
- Published by AIP Publishing in Applied Physics Reviews
- Vol. 10 (4)
Open AccessHigh-bandwidth density silicon photonic resonators for energy-efficient optical interconnects
The growth of artificial intelligence applications demands ever larger and more complex deep learning models, dominating today's—and tomorrow's—data center and high-performance computing systems. While traditional electronics are failing to keep pace with application demands, silicon photonic (SiPh) interconnects have emerged as a necessary technology to support these systems. SiPh-driven wavelength-division multiplexing (WDM) offers a particularly promising path toward supporting incredibly high-aggregate link bandwidth in a compact and efficient form factor. One of the basic building blocks of these integrated WDM interconnects is the SiPh resonator. Their inherent wavelength selectivity and compact footprint allow for efficient data transmission multiplexed across dozens of carrier wavelengths. Used as add-drop (AD) filters, SiPh resonators are critical to constructing integrated tunable wavelength-selective optical circuit switches as well as for demultiplexing the different carrier wavelengths toward independent wavelength-insensitive photodiodes in a dense wavelength-division multiplexing receiver. Resonators in the all-pass (AP) configuration are widespread as well, allowing for wavelength-selective modulation to drive aggregate link bandwidths far beyond the individual channel data rate. Unlike SiPh Mach–Zehnder modulators (MZM), resonant modulators can be driven using low, complementary metal-oxide-semiconductor drive voltages, allowing for tight co-integration between photonic integrated circuits, fabricated with larger process node technologies, and electronic integrated circuits, designed to exploit the advantages of the latest node. To push toward practical peta-scale interconnects, a comprehensive review of SiPh resonators is required, addressing bottlenecks and design constraints at both the architecture and device levels. We first describe the predominant integrated link architectures and identify their limits. We then discuss the device-level design considerations that can be made for both AD and AP configuration resonators to overcome the system level limits with novel resonator device designs. Analytical models and numerical simulation of resonators are validated by experimental measurement of devices fabricated in a commercial 300-mm foundry, showing a clear path toward volume manufacturing. The demonstrated resonant modulators and filters support the feasibility of increasing the aggregate bandwidth of resonator-driven SiPh interconnects into the peta-scale regime.
- Open Access
- 16 May 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in Journal of Lightwave Technology
- Vol. 41 (12) , 3737-3749
Peta-Scale Embedded Photonics Architecture for Distributed Deep Learning Applications
As Deep Learning (DL) models grow larger and more complex, training jobs are increasingly distributed across multiple Computing Units (CU) such as GPUs and TPUs. Each CU processes a sub-part of the model and synchronizes results with others. Communication among these CUs has emerged as a key bottleneck in the training process. In this work, we present SiPAC, a Silicon Photonic Accelerated Compute cluster. SiPAC accelerates distributed DL training by means of two co-designed components: a photonic physical layer and a novel collective algorithm. The physical layer exploits embedded photonics to bring peta-scale I/O directly to the CUs of a DL optimized cluster and uses resonator-based optical wavelength selectivity to realize hardware multi-casting. The collective algorithm builds on the hardware multi-casting primitive. This combination expedites a variety of collective communications commonly employed in DL training and has the potential to drastically ease the communication bottlenecks. We demonstrate the feasibility of realizing the SiPAC architecture through 1) an optical testbed experiment where an array of comb laser wavelengths are shuffled by a cascaded ring switch, with each ring selecting and forwarding multiple wavelengths to increase the effective communication bandwidth and hence demonstrating the hardware multicasting primitive, and 2) a four-GPU testbed running a realistic DL workload that achieves 22% system-level performance improvement relative to a similarly-sized leaf-spine topology. Large scale simulations show that SiPAC achieves a 1.4× to 5.9× communication time reduction compared to state-of-the-art compute clusters for representative collective communications.
- 16 May 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in Journal of Lightwave Technology
- Vol. 41 (12) , 3737-3749
Open AccessPeta-Scale Embedded Photonics Architecture for Distributed Deep Learning Applications
As Deep Learning (DL) models grow larger and more complex, training jobs are increasingly distributed across multiple Computing Units (CU) such as GPUs and TPUs. Each CU processes a sub-part of the model and synchronizes results with others. Communication among these CUs has emerged as a key bottleneck in the training process. In this work, we present SiPAC, a Silicon Photonic Accelerated Compute cluster. SiPAC accelerates distributed DL training by means of two co-designed components: a photonic physical layer and a novel collective algorithm. The physical layer exploits embedded photonics to bring peta-scale I/O directly to the CUs of a DL optimized cluster and uses resonator-based optical wavelength selectivity to realize hardware multi-casting. The collective algorithm builds on the hardware multi-casting primitive. This combination expedites a variety of collective communications commonly employed in DL training and has the potential to drastically ease the communication bottlenecks. We demonstrate the feasibility of realizing the SiPAC architecture through 1) an optical testbed experiment where an array of comb laser wavelengths are shuffled by a cascaded ring switch, with each ring selecting and forwarding multiple wavelengths to increase the effective communication bandwidth and hence demonstrating the hardware multicasting primitive, and 2) a four-GPU testbed running a realistic DL workload that achieves 22% system-level performance improvement relative to a similarly-sized leaf-spine topology. Large scale simulations show that SiPAC achieves a 1.4× to 5.9× communication time reduction compared to state-of-the-art compute clusters for representative collective communications.