報告_OCP2026_Dell_容錯量子電腦傳統基礎設施_20261015

PDF 原檔：報告_OCP2026_Dell_容錯量子電腦傳統基礎設施_20261015_original.pdf

講者公開簡報（OCP 官方 Recorded Content 頁提供下載）。文字以 PyMuPDF 逐頁抽取，### p<n> 為頁碼；圖表內容見原檔。

圖片清單（已驗證 2026-10-19）

原始內容

p1

The Classical Infrastructure of Fault-tolerant Quantum Computing Smriti Bajaj Applied Researcher Office of the CTO, Dell Technologies

p2

C Real-Time QEC Needs Tight Integration ​ Quantum Accelerator Pulse Processing Real-time Classical Processing Quantum Processing Unit Real-time classical workloads • Hybrid Applications: Connecting to and leveraging the QPU as an accelerator for subroutines in larger workflows (ms) • Error Correction: Low latency (~us) access to dedicated FPGA decoders with GPU-equipped classical servers • Error Mitigation: Post-processing plus classical optimization of analog control signals • Calibration: Improvement to fidelity and uptime via extraction of QPU parameters​ (s to min) Software solutions to ease quantum workload deployment Deploying and supporting quantum infrastructure at customer sites Quantum Hybrid Workload Accelerated HPC systems for classical workloads and quantum simulation1 1 3 4 Infrastructure to integrate quantum accelerators within classical systems 2

p3

C Real-Time QEC Needs Tight Integration ​ Quantum Accelerator Pulse Processing (Qblox) Real-time Classical Processing Quantum Processing Unit Real-time classical workloads • Hybrid Applications: Connecting to and leveraging the QPU as an accelerator for subroutines in larger workflows (ms) • Error Correction: Low latency (~us) access to dedicated FPGA decoders with GPU-equipped classical servers • Error Mitigation: Post-processing plus classical optimization of analog control signals • Calibration: Improvement to fidelity and uptime via extraction of QPU parameters​ (s to min) Software solutions to ease quantum workload deployment Deploying and supporting quantum infrastructure at customer sites Quantum Hybrid Workload Accelerated HPC systems for classical workloads and quantum simulation1 1 3 4 Infrastructure to integrate quantum accelerators within classical systems 2

p4

1 Is it worth it and where does it run? Route across heterogeneous compute; estimate the resources each workload needs. 2 How does it connect? Close the quantum–classical loop fast enough for real-time work including decoding. 3 Can I trust the answer? Get usable results off noisy hardware today; build toward correction tomorrow. The three fundamental questions ​

p5

QPUs QPU A CryoCMOS Silicon Spin Qubits QPU B Superconducting Qubits QPU N Neutral Atoms Qubits Control Infrastructure Control Box Instruments Quantum Intelligent Orchestrator (QIO) Problem submission Client API Circuit cutting / knitting Resource optimization ML model Runtime prediction ML model Error mitigation QIO Master Orchestration Engine HPC Layer Control Layer Job Manager Quantum Tasks Classical Tasks QIO Worker Slurm Daemon Classical Job Execution QPU A Driver QPU Emulation Transpilation Backends QPU N Driver QPU Access API Transpilation Backends Application time (ms) Real time / Deterministic Time (𝜇s) Physical time (𝑛s) 𝒕 Quantum Layer Control Layer Partners

p6

Dell HQCC: Built for the Real-Time Loop <4 usec roundtrip latency TCP/IP, RoCEv2 NVQLink (RoCEv2) Analog NVQLink Framework

p7

Comprehensive tool incorporating AI and broader range of variables to deliver reliable results​ Predict quantum hardware costs with respect to classical approach for a given problem type and size Quantum Advantage Calculator Paredes Quinones, et al., "An ML-Based Quantum Advantage Calculator," in 2025 IEEE International Conference on Quantum Computing and Engineering (QCE), Albuquerque, NM, USA, 2025, pp. 376-377, doi: 10.1109/QCE65121.2025.10353.

p8

• Dell Developed the first solution (to our knowledge) for QEM using GenAI.​ • The solution is based on Diffusion Models, a SOTA technique for image generation that has also been applied to other modalities (e.g., audio and time series).​ • Diffusion models are implemented as a denoising mechanism, which makes a direct conceptual connection with QEM.​ • Testbed: Dell IQM’s Garnet AI for Quantum Error Mitigation (QEM) Quantum error mitigation (QEM) is a critical stepping-stone toward achieving real-world value or "quantum utility" from today's quantum computers. In this whitepaper (© Dell Technologies, 2025), Dell and quantum hardware partner IQM demonstrate how AI- driven techniques can improve the fidelity of quantum computations and accelerate progress in hybrid quantum-classical computing (HQCC) systems.

p9

1 Is it worth it and where does it run? Route across heterogeneous compute; estimate the resources each workload needs. Quantum Intelligent Orchestrator Resource estimation intelligence 2 How does it connect? Close the quantum-classical loop fast enough for real-time work including decoding. Dell HQCC architecture NVQLink interconnect 3 Can I trust the answer? Get usable results off noisy hardware today; build toward correction tomorrow. QPERT, Quantum XAI IQM + Dell whitepaper on QEM The three fundamental questions ​

p10

Dell’s journey: agnostic HQCC system innovator/integrator • Dell + EY applied QUBO quantum annealing to address portfolio optimization 2023 • SC25: HQCC video demo Dell R760xa + QIO + QuEra • Joined Diraq for DARPA QBI 2025 • Dell Runtime for IBM Qiskit created 2021 • Dell-internal Supply Chain QUBO work deployed on R750xa 2024 • DQCS: Dell R740xd + IonQ • QIO (Quantum Intelligent Orchestrator) 2022 • OCTO QC Research kick-off 2016 • DTW: Live HQCC demo (Dell R770 + QIO + Equal1) • IEEE QW: HQCC real-time QEC demo (QIO + Qblox) 2026 Success Stories

p11

• We aim to lower the barrier to entry for adopting hybrid quantum-classical stacks. • Join in our working group on Runtime & Scheduling. • How to get involved with this community? Join the OCP Future Technologies Initiative community (Jonathan Burnett & Boniface Yogendran) Get Quantum ready Dell.com/quantum-computing Quantum.assist@dell.com Smriti.Bajaj@dell.com

p12

（本頁無可抽取文字，內容在圖中）