Hongzhong Zheng

Hongzhong Zheng

Auto-generated profile, based on publicly available information

Hongzhong Zheng

DAMO Academy, Alibaba Group, Hangzhou, China

Research Focus & Metrics

Academic Career Length

20Years (2007-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
18 June 2016 in ACM SIGARCH Computer Architecture News
Verified
Journal Article
Open access
PDF file
MGMingyu Gao
Mingyu Gao
CDChristina Delimitrou
Christina Delimitrou
DNDimin Niu
Dimin Niu
KMKrishna T. Malladi
Krishna T. Malladi
HZHongzhong Zheng
Hongzhong Zheng
CKChristos Kozyrakis
Christos Kozyrakis
FPGAs are a popular target for application-specific accelerators because they lead to a good balance between flexibility and energy efficiency. However, FPGA lookup tables introduce significant area and power overheads, making it difficult to use FPGA devices in environments with tight cost and power constraints. This is the case for datacenter servers, where a modestly-sized FPGA cannot accommodate the large number of diverse accelerators that datacenter applications need. This paper introduces DRAF, an architecture for bit-level reconfigurable logic that uses DRAM subarrays to implement dense lookup tables. DRAF overlaps DRAM operations like bitline precharge and charge restoration with routing within the reconfigurable routing fabric to minimize the impact of DRAM latency. It also supports multiple configuration contexts that can be used to quickly switch between different accelerators with minimal latency. Overall, DRAF trades off some of the performance of FPGAs for significant gains in area and power. DRAF improves area density by 10x over FPGAs and power consumption by more than 3x, enabling DRAF to satisfy demanding applications within strict power and cost constraints. While accelerators mapped to DRAF are 2-3x slower than those in FPGAs, they still deliver a 13x speedup and an 11x reduction in power consumption over a Xeon core for a wide range of datacenter tasks, including analytics and interactive services like speech recognition.

28 October 2023
Verified
Conference Paper
Open access
CBChen Bai
Chen Bai
JHJiayi Huang
Jiayi Huang
XWXuechao Wei
Xuechao Wei
YMYuzhe Ma
Yuzhe Ma
SLSicheng Li
Sicheng Li
YXYuan Xie
Yuan Xie
Design space exploration (DSE) for microarchitecture parameters is an essential stage in microprocessor design to explore the trade-offs among performance, power, and area (PPA). Prior work either employs excessive expert efforts to guide microarchitecture parameter tuning or demands high computing resources to prepare datasets and train black-box prediction models for DSE. In this work, we aim to circumvent the domain knowledge requirements through automated bottleneck analysis and propose ArchExplorer, which reveals microarchitecture bottlenecks to guide DSE with much fewer simulations. ArchExplorer consists of a new graph formulation of microexecution, an optimal critical path construction algorithm, and hardware resource reassignment strategies. Specifically, the critical path is constructed from the microexecution to uncover the performance-critical microarchitecture bottlenecks, which facilitates ArchExplorer to reclaim the hardware budgets of performance-insensitive structures that consume unnecessary power and area. These budgets are then reassigned to the microarchitecture bottlenecks for performance boost while maintaining the power and area constraints under the total budget envelope. Experiments show that ArchExplorer can find better PPA Pareto-optimal designs, achieving an average of higher Pareto hypervolume using at most fewer simulations compared to the state-of-the-art approaches.

4 December 2024 in IEEE Transactions on Circuits and Systems for Video Technology
Verified
Journal Article
Closed access
HZHao Zhang
Hao Zhang
SLSicheng Li
Sicheng Li
YGYupeng Gui
Yupeng Gui
ZLZhiyong Li
Zhiyong Li
SXShusong Xu
Shusong Xu
Yibo FanYibo Fan
Yibo Fan
To achieve high-quality and high-resolution image processing, this work presents a novel vision processor that facilitates deep learning-enhanced image processing pipelines. At the system level, by identifying that a divide-and-conquer approach is essential to synergize both classical image processing and image enhancement networks, we develop a tightly coupled system with strip-tile conversion dataflow to enable fine-grained low-latency data interactions between image signal processors (ISPs) and the deep learning accelerator (DLA). At the architecture level, we design a comprehensive set of 21 efficient image processing modules to construct classical ISP pipelines, a tile-based strip layer fusion DLA specifically optimized for networks, and a programmable pixel pool that seamlessly supports the data access patterns of the ISP and the DLA. At the software and hardware co-design level, we propose a comprehensive optimization framework to address the implementation overhead of networks while maintaining the image quality. Finally, evaluations of the AI-ISP vision processor demonstrate 53.95% external memory access reduction and 35.51% latency reduction, delivering superior image quality with minimal on-chip memory overhead. A throughput of up to 168.5 frames per second facilitates efficient processing of ultra-high definition (UHD) resolution images.

28 October 2023
Verified
Conference Paper
Open access
ZQZheng Qu
Zheng Qu
DNDimin Niu
Dimin Niu
SLShuangchen Li
Shuangchen Li
HZHongzhong Zheng
Hongzhong Zheng
YXYuan Xie
Yuan Xie
Training Graph Neural Networks on large graphs is challenging due to the need to store graph data and move them along the memory hierarchy. In this work, we tackle this by effectively compressing graph embedding matrix such that the model training can be fully enabled with on-chip compute and memory resources. Specifically, we leverage the graph homophily property and consider using Tensor-train to represent the graph embedding. This allows nodes with similar neighborhoods to partially share the feature representation. While applying Tensor-train reduces the size of the graph embedding, it imposes several challenges to hardware design. On one hand, utilizing low-rank representation requires the features to be decompressed before being sent to GNN models, which introduces extra computation overhead. On the other hand, the decompressed features might still exceed on-chip memory capacity even with the minibatch setting, causing inefficient off-chip memory access. Thus, we propose the TT-GNN hardware accelerator with a specialized dataflow tailored for on-chip Tensor-train GNN learning. Based on the on-chip memory capacity and training configuration, TT-GNN adaptively breaks down a minibatch into smaller microbatches that can be fitted on-chip. The microbatch composition and scheduling order are designed to maximize data reuse and reduce redundant computations both across and within microbatches. To mitigate TT computation overhead, we further propose a unified algorithm to jointly handle TT decompression during forward propagation and TT gradient derivation during backward propagation. Evaluated on a series of benchmarks, the proposed software-hardware solution is able to outperform existing CPU-GPU training systems on both training performance (1.55 ∼ 4210 ×) and energy efficiency (2.83 ∼ 2254 ×). We believe TT-GNN introduces a new perspective to address large-scale GNN training and enables possibilities to train GNN models even under a significantly constrained resource budget.

1 April 2007
Verified
Conference Paper
Closed access
LJLin Jiang
Lin Jiang
HZHongzhong Zheng
Hongzhong Zheng
ZZZhichun Zhu
Zhichun Zhu
ZZZhao Zhang
Zhao Zhang
HDHoward David
Howard David
We have studied DRAM-level prefetching for the fully buffered DIMM (FB-DIMM) designed for multi-core processors. FB-DIMM has a unique two-level interconnect structure, with FB-DIMM channels at the first-level connecting the memory controller and advanced memory buffers (AMBs); and DDR2 buses at the second-level connecting the AMBs with DRAM chips. We propose an AMB prefetching method that prefetches memory blocks from DRAM chips to AMBs. It utilizes the redundant bandwidth between the DRAM chips and AMBs but does not consume the crucial channel bandwidth. The proposed method fetches K memory blocks of L2 cache block sizes around the demanded block, where K is a small value ranging from two to eight. The method may also reduce the DRAM power consumption by merging some DRAM precharges and activations. Our cycle-accurate simulation shows that the average performance improvement is 16% for single-core and multi-core workloads constructed from memory-intensive SPEC2000 programs with software cache prefetching enabled; and no workload has negative speedup. We have found that the performance gain comes from the reduction of idle memory latency and the improvement of channel bandwidth utilization. We have also found that there is only a small overlap between the performance gains from the AMB prefetching and the software cache prefetching. The average of estimated power saving is 15%

13 July 2023 in IEEE Transactions on Parallel and Distributed Systems
Verified
Journal Article
Closed access
YWYanhong Wang
Yanhong Wang
TGTianchan Guan
Tianchan Guan
DNDimin Niu
Dimin Niu
QZQiaosha Zou
Qiaosha Zou
HZHongzhong Zheng
Hongzhong Zheng
YXYuan Xie
Yuan Xie
Emerging graph neural network (GNN) has recently attracted much attention and has been used extensively in many real-world applications thanks to its powerful expression ability of unstructured data. The real-world graph datasets are very large-scale, which can contain up to billions of nodes and tens of billions of edges. It usually requires distributed system to train GNN on such huge datasets. As a result, the data communication overheads between machines become the bottleneck of GNN computation. Our profiling results show that getting attributes from remote machines during sampling phase in GNN occupies $> $ 75% of the time of the training process. To address this issue, in this article, we propose Coded Neighbor Sampling (CNS) framework, which introduces codes technique to reduce the communication overheads of GNN. In the proposed CNS framework, the codes technique is coupled with GNN sampling method to exploit the data excess among different machines caused by unstructured nature of graph data. An analytical performance model is built for the proposed CNS framework, whose results are corroborated by the simulation and validate the benefit of the proposed CNS framework over both conventional GNN training method and conventional codes technique. Performance metrics, such as communication overheads, runtime, and throughput, of the proposed CNS framework are evaluated on a distributed GNN training simulation system implemented on MPI4py platform. The results show that, on average, the proposed CNS framework can save communication overhead by 40.6%, 35.5%, and 16.5%, reduce the runtime by 12.1%, 17.0%, and 10.0%, and improve the throughput by 16.2%, 24.4%, and 11.2%, respectively, when training GNN models with Cora, PubMed, and Large Taobao.