Daohui Wang

LithoMamba: High-fidelity lithography simulation with State Space Models

Xinyu He, Daohui Wang, Shujing Lyu, Pourya Shamsolmoali et autres

Lithography simulation is a critical technology in modern semiconductor manufacturing, yet existing deep learning models often fail to accurately model the complex, long-range optical physics due to the inherent locality of convolution. This limitation results in insufficient simulation fidelity and poses significant …

cn (code pays fourni par la source)

Q-Infer: Towards Efficient GPU-CPU Collaborative LLM Inference via Sparsity-Aware Dynamic Scheduling

Kai Lü, Qiang Wei, P. X. Liu, Jiguang Wan et autres

Large Language Models (LLMs) have sparked a new wave of exciting AI applications, yet their large model size imposes significant computational and storage costs during inference. Offloading parameters to the CPU and conducting GPU-CPU collaborative inference is a highly cost-effective strategy to …

cn (code pays fourni par la source)

Huawei Cloud Model-as-a-Service on the CloudMatrix384 SuperPod

Ao Xiao, Bangzheng He, Baoquan Zhang, Baoxing Huai et autres

Scaled-out MoE LLMs and scaled-up SuperPods create new systems challenges for production Model-as-a-Service (MaaS), requiring disaggregation, low-latency communication, and decentralized serving. This report presents xDeepServe, the production serving system behind Huawei Cloud's MaaS offering on CloudMatrix384, a 48-server SuperPod with 384 Ascend …

DFlush: DPU-Offloaded Flush for Disaggregated LSM-based Key-Value Stores

Chen Ding, Kai Lü, Ting Yao, Daohui Wang et autres

Rapid increase of storage and network bandwidth incurs higher CPU consumption in modern data systems. This phenomenon is particularly evident for log-structured merged key-value stores (LSM-KVS), which rely on resource-intensive background operations to flush and compact disk data. While extensive research has …

cn (code pays fourni par la source)

Photolithographic image prediction with conditional adversarial network and parameter encoding

Xinyu He, Daohui Wang, Wenzhan Zhou, Kan Zhou et autres

Photolithography is a pivotal stage in integrated circuit chip manufacturing, exerting a direct influence on both the performance and yield of the chips. Its efficacy hinges heavily on the meticulous control of parameters such as focus and exposure dose. Traditionally, the production …

cn (code pays fourni par la source)

Scythe: A Low-latency RDMA-enabled Distributed Transaction System for Disaggregated Memory

Kai Lü, Siqi Zhao, Haikang Shan, Qiang Wei et autres

Disaggregated memory separates compute and memory resources into independent pools connected by RDMA (Remote Direct Memory Access) networks, which can improve memory utilization, reduce cost, and enable elastic scaling of compute and memory resources. However, existing RDMA-based distributed transactions on disaggregated memory …

cn (code pays fourni par la source)

Rcmp: Reconstructing RDMA-Based Memory Disaggregation via CXL

Zhonghua Wang, Y. M. Guo, Kai Lü, Jiguang Wan et autres

Memory disaggregation is a promising architecture for modern datacenters that separates compute and memory resources into independent pools connected by ultra-fast networks, which can improve memory utilization, reduce cost, and enable elastic scaling of compute and memory resources. However, existing memory disaggregation …

cn (code pays fourni par la source)

SepHash: A Write-Optimized Hash Index On Disaggregated Memory via Separate Segment Structure

Xinhao Min, Kai Lü, Pengyu Liu, Jiguang Wan et autres

Disaggregated memory separates compute and memory resources into independent pools connected by fast RDMA (Remote Direct Memory Access) networks, which can improve memory utilization, reduce cost, and enable elastic scaling of compute and memory resources. Hash indexes provide high-performance single-point operations and …

cn (code pays fourni par la source)

DoW-KV: A DPU-offloaded and Write-optimized Key-Value Store on Disaggregated Persistent Memory

Yiwen Zhang, Guokuan Li, Jiguang Wan, Junyue Wang et autres

Disaggregated Persistent Memory (DPM) is a promising technology offering elasticity, high resource utilization, persistent data storage, and lower power consumption. While building KV stores on the DPM benefits from these merits, achieving efficient writes also faces two primary challenges: 1) limited scalability …

cn (code pays fourni par la source)

PetaKV: Building Efficient Key-Value Store for File System Metadata on Persistent Memory

Yiwen Zhang, Jian Zhou, Xinhao Min, Song Ge et autres

Previous works proposed building file systems and organizing the metadata with KV stores because KV stores handle entries of various sizes efficiently and have excellent scalability. The emergence of the byte-addressable persistent memory (PM) enables metadata service to be faster than before …

cn (code pays fourni par la source)

Application of Adaptive Sub-band Filters on Active Noise Control

Chengtao Du, Daohui Wang, Liang Cai, Min Fan

Active noise control is an effective method for active noise reduction for low frequency noise. In this paper, the forward control method is adopted for noise reduction special in semi-enclosed spaces. The primary(original) noise from noise sources is detected first, and then, …

cn (code pays fourni par la source)

Dual-Stimuli-Responsive Cross-Linked Graphene Oxide/Poly(vinyl alcohol) Membranes with Anisotropic Liquid Penetration Behaviors

Xianfeng Li, Caiyun Wang, Daohui Wang, Zhen Liu et autres

Stimuli-responsive membranes exhibit a flexible adjustment in response to environmental stimuli and have been established many applications. In this paper, dual-stimuli-responsive ultrafiltration membranes were designed by stacking graphene oxide (GO) nanosheets cross-linked with poly(vinyl alcohol) (PVA) onto the surface of a porous …

cn, ca (code pays fourni par la source)