PDF 原檔：報告_OCP2026_NVIDIA_Microsoft_800VDC安全設計_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）
| 檔名 | 頁 | 分類 | 親眼所見內容 |
|---|---|---|---|
報告_OCP2026_NVIDIA_Microsoft_800VDC安全設計_20261015_p3.png |
p3 | 真資料圖 | 簡化 800 VDC 資料中心單線圖：MVAC 饋線→MVAC/800Vdc 轉換器→800Vdc 主母線（掛雙向 DC/DC 與儲能）→800Vdc 開關櫃→機櫃母線→各機櫃 TOB，並列四項系統特性 |
報告_OCP2026_NVIDIA_Microsoft_800VDC安全設計_20261015_p5.png |
p5 | 真資料圖 | 三種接地系統比較示意圖：高阻中點接地（IMD+RCM）、浮接/不接地（IMD）、實接地（RCM），附各自優缺點條列 |
報告_OCP2026_NVIDIA_Microsoft_800VDC安全設計_20261015_p8.png |
p8 | 真資料圖 | DC 斷路裝置比較表：MCCB（數十 ms）、SSCB（約 10 μs）、Hybrid CB（約 ms）、Fuse（百 μs 至 ms）、E-fuse（<2–10 μs）、Pyro-fuse（<2 ms）的特性、用途與限制 |
報告_OCP2026_NVIDIA_Microsoft_800VDC安全設計_20261015_p9.png |
p9 | 真資料圖 | 保護協調風險示意：Case 1 TOB SSCB（約 10 μs）在 PSU 保險絲（100 μs–ms）之前先跳、Case 2 上下游兩顆 SSCB 速度相近可能同時跳脫 |
報告_OCP2026_NVIDIA_Microsoft_800VDC安全設計_20261015_p10.png |
p10 | 真資料圖 | 三種機櫃供電型態圖：傳統 AC 機櫃（數十 kW）、Diablo400 sidecar 支援機櫃（數百 kW）、原生 HVDC 機櫃（數 MW 級）的電力路徑 |
原始內容
p1
Xianyong Feng [Principal Electrical Engineer, Microsoft] Mike Tu, P.E. [Principal Electrical Engineer, NVIDIA] Sai Somayajula [Principal Electrical Engineer, NVIDIA] Designing Safe 800 VDC Systems: Protection, Fault Management, and Operations
p2
MW-Scale DC Distribution Is Already Proven Across Industry SOLAR PV Native DC generation EV CHARGING High-power DC delivery ENERGY STORAGE Bidirectional DC power RAIL & MARINE Electrified propulsion Transfer proven DC practices for AI Data Center application.
p3
Key considerations of 800 VDC data centers 1. Converter-dominated system — Power electronics govern power flow, dynamic response, and fault behavior. 2. Low-impedance system — Disturbances propagate quickly between connected loads. 3. Native 800 VDC rack feed — Direct busway connection reduces in-rack delivery current. 4. EDPP transient impact — Rapid rack load currents cause voltage transients and obscure series-arc signatures, complicating fault detection. Simplified 800 VDC Data Center SLD MVAC FEED MVAC / 800 Vdc CONVERTER 800 Vdc SWITCHBOARD 800 Vdc RACK BUSWAY 800 Vdc MAIN BUS BIDIRECTIONAL DC/DC Energy Storage TOB RACK 1 TOB RACK 2 TOB RACK n ··· ···
p4
Key protection requirements in 800 VDC data centers 1. Fault Identification — Detect and classify electrical faults quickly, accurately, and reliably. 2. Ground-fault location — Rapidly locate ground faults for various grounding system setup. 3. Fault Clearing time — Use DC-rated interrupting devices to quickly clear high-di/dt faults 4. Protection Coordination — Isolate only the affected section while maintaining power to healthy loads. 5. Arc and surge control — Detect DC arcs, reduce incident energy, and limit transient overvoltage. 800 VDC Data Center Protection System Protection must act fast, reliably, selectively, and across the full DC system 1 DETECT high di/dt fault, high impedance fault, series arc, ground fault, etc. 2 LOCATE faulted branch 3 ISOLATE selective trip 4 STABILIZE limit surge / fast recovery
p5
Grounding System Options
p6
DC Fault Detection: Two-Zone Architecture • Zone 2 — Source / Main DC Bus • Insulation Monitoring Device (IMD) • Continuously monitors insulation resistance to ground • Detects the first insulation fault in floating or HRG networks • Provides early warning before a second fault creates a hazardous path • Initiates fault-location logic across downstream branches • Zone 1 — Branch Circuit / Compute Rack • Residual Current Monitoring (RCM) • Hall-effect sensor measures current imbalance • Detects leakage current from the branch circuit to ground • Identifies the affected rack or branch • Supports targeted isolation without interrupting healthy loads DC Fault Detection
p7
BRANCH C Ground-Fault Detection and Localization Central IMD detects system-wide insulation degradation; zonal RCMs identify the affected branch. SIMPLIFIED GROUND-FAULT PROTECTION SLD MVAC / 800 VDC CONVERTER 800 VDC MAIN BUS CENTRAL IMD insulation resistance System-wide detection RCM 1 BRANCH A RCM 2 BRANCH B RCM 3 RCMs at protection-zone boundaries PROTECTION SEQUENCE 01 DETECT IMD flags insulation degradation →02 LOCALIZE RCMs compare branch leakage current →03 CONFIRM Correlate central and zonal indications →04 ISOLATE Clear only the affected branch Level 1: Alarm Level 2: Trip AFFECTED BRANCH 1. RCM Pickup 2. IMD Confirmed
p8
DC Interrupting Devices Device Clearing Key feature Typical use Advantages Limitations MCCB Tens of ms Compact molded case; thermal-magnetic or electronic trip Switchboards, branch and feeder protection Compact; economical; resettable; familiar coordination Slow; curve-dependent; arcing and wear SSCB ≈10 µs Semiconductor interruption; fast sensing and energy clamp Converters, batteries, and fast bus protection Fastest; no arc; resettable; programmable Conduction loss and cooling; higher cost; semiconductor failure rate Hybrid CB ~ms Low-loss mechanical path + semiconductor commutation High-power DC where speed and efficiency are both critical Fast interruption with low steady-state loss More components, controls and coordination; higher complexity/cost Fuse 100s µs - ms Melting element; current limiting; no control power Branch protection, module protection Low cost and loss; high interrupt rating Replacement/downtime; slower at low overcurrent levels E-fuse <2–10 µs MOSFET/SiC switch with current limit and reset Low/medium-power rails, EV auxiliaries, data centers Ultra-fast; resettable; configurable; no arc Semiconductor loss/thermal limits; bounded v and i range Pyro-fuse <2 ms Externally triggered with galvanic isolation EV/ESS battery crash or severe-fault isolation Low on-state loss; fast and decisive isolation One-shot; needs trigger logic; not general overload protection
p9
Coordinate protection as one integrated system • Speed mismatch — SSCBs act in microseconds; fuses and MCCBs act in milliseconds. • Failure mode — A fast upstream device can trip first and disconnect healthy loads. • Design rule — Set the fault window first, then align I–t curves, let-through energy, and delays. DC Protection Needs End-to-End Coordination Case 1 — Fast SSCB ahead of a slower PSU fuse 800 Vdc busway TOB SSCB ~10 µs PSU fuse 100 µs–ms Fault in PSU Risk — TOB SSCB clears first; one PSU fault drops the TOB. Case 2 — Two SSCBs with near-identical speed 800 Vdc switchboard Upstream SSCB ~10 µs TOB SSCB ~10 µs Fault in TOB feed Risk — Both SSCBs may trip without explicit discrimination.
p10
• Detection challenge — Fault current may remain below protective-device or power- supply OCP thresholds, delaying fault clearing. • Arc-flash risk — Longer clearing time increases arc-flash incident energy. • Enhanced sensing — Supplement current sensing with temperature or other advanced sensors for faster, reliable detection. AFIE Control for High Impedance Fault THREE RACK POWER-DELIVERY CONFIGURATIONS 1 Conventional AC rack Lower power • conversion in rack 2 Sidecar-supported rack High power • Diablo400 interface 3 Native HVDC rack Direct 800 Vdc distribution AC BUSWAY TOB LOWER-POWER IT RACK IN-RACK PSU SHELF AC BUSWAY TOB DIABLO400 SIDECAR AC/DC ISOLATION HIGH-POWER IT RACK MVAC / LVDC CONVERSION AC/DC ISOLATION 800 Vdc BUSWAY TOB NATIVE HVDC IT RACK Power Zone: tens of kW Power Zone: hundreds of kW Power Zone: a few MWs
p11
• Initiation — Loose or degraded contacts create an air gap that sustains a DC series arc. • Hidden hazard — Line current may not rise, allowing the arc to persist and create a fire risk. • Variable signature — Depends on current, materials, gap geometry, environment, circuit R-L-C, and load. • Detection challenge — Shall separate intermittent arc signatures from normal converter/load transients. DC Series Arc Fault Detection (1) constant speed electrode separation (0.1‘’/s) (2) fixed gap electrode (0.25‘’) *Ref: A DC Arc Model for Series Faults in Low Voltage Microgrids
p12
• Current Limiting Devices adoption & approval • Solid State Circuit Breaker development & certification • DC protective relays and fault sensing methods • 800 VDC System-Level and SST Spec Development • DC Arc Fault behavior, testing, code update Call to Actions
p13
Thank You!