Study station · AI PC & AI datacenter supply chain · v1研究站 · AI PC 与 AI 数据中心供应链 · v1
Etch and chamber clean in wafer fabs. Dry etch and PECVD cleans in display panel lines. SF₆ switchgear, chillers, immersion fluids and clean-agent suppression in the datacenter. Same gases, same logic: measure at the point, control at the point, prove it with evidence.
晶圆厂的刻蚀与腔体清洗；显示面板线的干法刻蚀与 PECVD 清洗；数据中心的 SF₆ 开关柜、冷水机、浸没式冷却液与洁净气体消防。气体相同，逻辑相同：在点位测量、在点位控制、用证据证明。
Once it leaves the stack the decision is permanent. A kilogram of CF₄ vented this shift is still up there in 50,000 years. 一旦排出烟囱，这个决定就是永久的。本班次排放的 1 公斤 CF₄，五万年后仍在大气中。
Log scale. Lifetimes: IPCC AR6 WG1 Ch.7 (see 12 Sources). This is why point-of-use control beats any later offset: there is no “later” for CF₄.
对数刻度。寿命数据来自 IPCC AR6 WG1 第 7 章（见 12 来源）。这就是为什么点位控制优于任何事后抵消：对 CF₄ 而言没有“事后”。
Start from measurement从测量开始
Mass alone tells you nothing. Mass × GWP tells you the climate number; lifetime tells you whether the mistake is reversible. Open a row to see how it changes the way you work.
只看质量毫无意义。质量 × GWP 给出气候数值；寿命告诉你这个错误是否可逆。展开条目，看它如何改变你的工作方式。
Click “+” to open anything; panels nest, so you can go deeper. Click a process node in 04 to jump to its control point. Tick boxes in 07 and 10 — they score live. Drag the molecule in 08.
点击“+”展开任意条目；面板可层层嵌套，越点越深。在 04 点击工艺节点直接跳到该控制点。在 07 与 10 勾选方框，会实时计分。在 08 拖动分子旋转。
On the numbers: GWP values and lifetimes are quoted from IPCC AR5/AR6; the calculation equations are from the IPCC 2019 Refinement, Vol.3 Ch.6 (Electronics). Parameter values in the worked examples and the four-band benchmark ladder are this study's illustrative constructs — use them to learn the shape of the problem, then replace them with your own tool data, supplier test reports and the official default tables.
关于数据：GWP 与大气寿命引自 IPCC AR5/AR6；计算公式引自 IPCC 2019 精细化指南第 3 卷第 6 章（电子行业）。示例中的参数取值与四级基准梯度为本研究的示意性设定：先用它理解问题的形状，再用你自己的设备数据、供应商测试报告与官方默认表替换。
| Gas气体 | GWP-100 AR6 | GWP-100 AR5 | Lifetime大气寿命 | Where you meet it出现在哪里 |
|---|---|---|---|---|
| CF₄ (PFC-14) | 7,380 | 6,630 | ~50,000 y | Etch, chamber clean, and a by-product of other PFCs刻蚀、腔体清洗，且是其他 PFC 的副产物 |
| C₂F₆ (PFC-116) | 12,400 | 11,100 | ~10,000 y | In-situ chamber clean, dielectric etch原位腔体清洗、介质层刻蚀 |
| C₃F₈ | 9,290 | 8,900 | ~2,600 y | Chamber clean腔体清洗 |
| c-C₄F₈ | 10,200 | 9,540 | ~3,200 y | Dielectric / high-aspect-ratio etch介质层与高深宽比刻蚀 |
| CHF₃ (HFC-23) | 14,600 | 12,400 | 228 y | Oxide etch; also an HCFC-22 production by-product氧化物刻蚀；也是 HCFC-22 生产副产物 |
| NF₃ | 17,400 | 16,100 | 569 y | Remote-plasma chamber clean (fab & display)远程等离子腔体清洗（晶圆与面板） |
| SF₆ | 24,300 | 23,500 | ~3,200 y | Display dry etch, deep-Si etch, MV switchgear面板干法刻蚀、深硅刻蚀、中压开关柜 |
| N₂O | 273 | 265 | 109 y | CVD/oxidation — not an F-gas, but reported alongsideCVD/氧化 —— 不属 F-gas，但需一并报告 |
| HFC-134a | 1,530 | 1,300 | 14 y | Chillers, DX cooling, some cleaning uses冷水机、直膨式空调、部分清洗用途 |
| R-410A (blend) | ~2,256 | ~1,924 | – | CRAC / DX units in server rooms机房 CRAC / 直膨机组 |
| HFC-227ea | 3,600 | 3,350 | 36 y | Clean-agent fire suppression洁净气体消防系统 |
| FK-5-1-12 | ~1 | <1 | ~14 days | Suppression, two-phase immersion (check PFAS policy)消防、两相浸没冷却（须核对 PFAS 政策） |
| HFO-1234ze | <1 | <1 | ~18 days | Low-GWP chiller refrigerant低 GWP 冷水机制冷剂 |
Values as published in IPCC AR5 / AR6 and EPA's GWP summary. Report with one consistent GWP set and state which — mixing AR5 and AR6 is the most common audit finding.数值引自 IPCC AR5 / AR6 及 EPA 的 GWP 汇总。报告时须统一使用同一套 GWP 并明确说明 —— 混用 AR5 与 AR6 是最常见的审计问题。
CF₄ is a small, perfectly symmetric, fully fluorinated molecule with the strongest single bond in organic chemistry (C–F, roughly 480–545 kJ/mol). Nothing in the troposphere attacks it: no OH reaction, no photolysis, no rain-out. Thermal abaters therefore need high temperature, oxygen/hydrogen chemistry and residence time to break it, and CF₄ is always the last gas to reach a high destruction rate.
CF₄ 是一个小而完全对称、全氟化的分子，其 C–F 键是有机化学中最强的单键（约 480–545 kJ/mol）。对流层中没有任何机制能攻击它：不与 OH 反应、不光解、不被雨水冲洗。因此热式尾气处理需要高温、氧/氢化学与足够停留时间才能破坏它，CF₄ 永远是最后才达到高去除率的气体。
The operational consequence: a site can honestly claim ">95% DRE" on NF₃ and still emit most of its CO₂e as CF₄. Always ask for DRE by gas, and always ask for CF₄ separately — including the CF₄ formed inside the chamber from C₂F₆, CHF₃ and c-C₄F₈.
运营含义：一个厂区可以真实地声称 NF₃ 的 DRE ">95%"，却依然以 CF₄ 的形式排掉大部分 CO₂e。永远要求按气体提供 DRE，并单独索取 CF₄ 数据 —— 包括腔体内由 C₂F₆、CHF₃、c-C₄F₈ 生成的 CF₄。
NF₃ has a GWP-100 of 17,400 — higher than C₂F₆. Yet moving a chamber clean from in-situ C₂F₆/C₃F₈ to a remote plasma NF₃ clean typically reduces the CO₂e of that step by an order of magnitude, because remote plasma dissociates almost all of the NF₃ (utilisation typically well above 0.9, versus roughly 0.3–0.6 for in-situ perfluorocarbon cleans) and generates far less CF₄ by-product.
NF₃ 的 GWP-100 为 17,400，高于 C₂F₆。但把腔体清洗从原位 C₂F₆/C₃F₈ 改为远程等离子 NF₃ 清洗，该工序的 CO₂e 通常可下降一个数量级：远程等离子几乎将 NF₃ 全部解离（利用率通常远高于 0.9，而原位全氟碳清洗约为 0.3–0.6），且生成的 CF₄ 副产物远少。
How to think: the number you emit is mass not consumed × GWP, not mass purchased × GWP. High GWP with high utilisation and good abatement beats moderate GWP that walks straight out of the pump.
思考方式：你排放的是未被消耗的质量 × GWP，而不是采购质量 × GWP。"高 GWP + 高利用率 + 良好尾气处理" 优于 "中等 GWP 却直接从泵口跑掉"。
In a display panel line SF₆ appears in dry etch and some chamber cleans; at GWP 24,300 a modest kilogram figure becomes a large CO₂e figure. In an AI datacenter the same molecule sits inside medium-voltage gas-insulated switchgear (GIS) and ring main units as an insulating and arc-quenching medium — Scope 1, on your own site, often with no measurement at all beyond a nameplate charge and a pressure gauge.
在显示面板线上，SF₆ 出现于干法刻蚀与部分腔体清洗；GWP 高达 24,300，因此不大的公斤数会变成很大的 CO₂e。在 AI 数据中心，同一分子存在于中压 GIS 与环网柜中，作为绝缘与灭弧介质 —— 属于范围一、就在你自己的场地内，且往往除了铭牌充装量和压力表之外毫无测量。
Two very different control points, one gas: upstream it is a process parameter (recipe, abatement); on your own site it is an asset register plus a leak rate. Both belong in the same F-gas inventory.
同一种气体，两个完全不同的控制点：上游是工艺参数（配方、尾气处理）；在自有场地则是资产台账加泄漏率。两者应纳入同一份 F-gas 清单。
Liquid-cooled AI racks change the fluid picture. Two-phase immersion has used fluoroketones and PFPEs; 3M announced it will exit PFAS manufacturing by the end of 2025, which makes fluid choice a supply-security question as well as a climate one. Clean-agent suppression using HFC-227ea (GWP 3,600) sits in a cylinder for twenty years and then either stays there or becomes an emission. Chiller and CRAC charges are usually the largest single F-gas mass on a datacenter site.
液冷 AI 机柜改变了流体图景。两相浸没冷却曾使用氟代酮与 PFPE；3M 已宣布在 2025 年底前退出 PFAS 生产，这使冷却液选择同时成为供应安全与气候问题。使用 HFC-227ea（GWP 3,600）的洁净气体消防系统在钢瓶中静置二十年，之后要么继续留在瓶中，要么变成一次排放。冷水机与 CRAC 的充装量通常是数据中心场地中单项质量最大的 F-gas。
First move: build a charge register (unit, fluid, kg, GWP, t CO₂e, install date, last leak check). That register is both your inventory and your audit evidence.
第一步：建立充装量台账（设备、介质、公斤数、GWP、t CO₂e、安装日期、上次检漏）。这份台账既是你的清单，也是你的审计证据。
Order of operations行动顺序
Most programmes start at rung 4 (buy an abater) and never do rung 1. The cheapest tonne is the gas that never enters the chamber; the most defensible tonne is the one you measured.
多数项目从第 4 级（买一台尾气处理）开始，却从不做第 1 级。最便宜的一吨是从未进入腔体的气体；最经得起质疑的一吨是你测量过的那一吨。
Delete the step. Fewer chamber cleans per wafer through longer clean intervals, in-line endpoint detection, or a process that does not need a fluorinated etch at all. Zero use needs no destruction and no evidence.
删掉这一步。通过延长清洗间隔、在线终点检测，或改用无需氟化刻蚀的工艺，减少每片晶圆的清洗次数。零使用即无需去除，也无需举证。
Same step, less gas: recipe optimisation, flow trimming, endpoint control instead of fixed over-clean time, no idle-flow. This is the fastest reduction available without a capital project.
同一步骤、更少气体：优化配方、削减流量、以终点控制取代固定超时清洗、消除待机流量。这是无需资本项目即可实现的最快减排。
Same function, better molecule or better delivery: remote NF₃ instead of in-situ C₂F₆; SF₆-free switchgear; HFO/low-GWP refrigerant; inert or FK-5-1-12 suppression. Judge on emitted CO₂e, not on GWP alone.
功能不变，换更好的分子或更好的供给方式：远程 NF₃ 取代原位 C₂F₆；无 SF₆ 开关柜；HFO/低 GWP 制冷剂；惰性气体或 FK-5-1-12 消防。以实际排放的 CO₂e 判断，而非单看 GWP。
Point-of-use abatement on every tool exhaust that carries PFC / NF₃ / SF₆, with interlocks so a tool cannot process while its abater is down. Coverage × DRE × uptime — all three, or the number is fiction.
对所有含 PFC / NF₃ / SF₆ 的设备尾气加装点位处理装置，并设联锁：处理装置故障时设备不得运行。覆盖率 × DRE × 可用率 —— 三者缺一，数字即虚构。
Mass balance from purchase to heel return, FTIR/QCL at abater inlet and outlet, calibration certificates, uptime logs, a signed method note. Unproven reduction is not reduction; it is a claim.
从采购到残气回收的物料平衡、处理装置进出口的 FTIR/QCL 测量、校准证书、可用率日志、经签署的方法说明。未经证明的减排不是减排，只是主张。
↻ Then re-baseline annually: measure → find hotspot → control → verify → report → target → measure↻ 之后每年重新基线：测量 → 找热点 → 控制 → 验证 → 报告 → 定目标 → 再测量
Supply chain mapping供应链映射
For a brand, almost all of it is Scope 3 category 1 sitting three tiers away. For an operator, a surprising amount is Scope 1 inside your own fence. Open each domain to see what is in scope, what is deliberately out, and roughly what share to expect.
对品牌而言，绝大部分位于三级之外的范围三类别 1；对运营方而言，相当一部分就在自家围墙内的范围一。展开每个领域，查看纳入什么、有意排除什么、以及大致的占比预期。
Domain A领域 A
Wafer
fab
Etch + CVD chamber clean. Usually the single largest F-gas CO₂e block behind an AI PC or an AI accelerator.
刻蚀 + CVD 腔体清洗。通常是 AI PC 或 AI 加速器背后最大的一块 F-gas CO₂e。
Domain B领域 B
Display
panel
Large-area PECVD cleans and SF₆ dry etch. High volumes, historically lower abatement coverage.
大面积 PECVD 清洗与 SF₆ 干法刻蚀。用量大，历史上尾气处理覆盖率较低。
Domain C领域 C
Datacenter
& own sites
SF₆ switchgear, refrigerant charges, immersion fluid, clean-agent suppression. Your Scope 1.
SF₆ 开关柜、制冷剂充装、浸没冷却液、洁净气体消防。属于你的范围一。
• Fab electricity — Scope 2, separate account. Do not net it against F-gas.
• PFAS in photoresists, surfactants and coolants — a chemical-management and product-compliance topic, not a GHG line. Track it, but in the substance programme.
• Heel returned to the gas supplier — not your emission if the certificate proves it was reclaimed or destroyed; keep the certificate.
• HF / SiF₄ / COF₂ acid gases — air-quality and wet-scrubber scope; they are not GWP-relevant.
• Upstream gas manufacture — the gas maker's Scope 1; it reaches you as Scope 3 cat. 1 embodied in the chemical, do not double count with your own use.
• 厂区用电 —— 属范围二，单独核算，不得与 F-gas 相抵。
• 光刻胶、表面活性剂与冷却液中的 PFAS —— 属化学品管理与产品合规议题，不是温室气体条目。要跟踪，但归入物质管理项目。
• 返还气体供应商的残气 —— 若有证书证明已回收或销毁，则不计为你的排放；请保留证书。
• HF / SiF₄ / COF₂ 等酸性气体 —— 属大气污染与湿式洗涤范围，与 GWP 无关。
• 上游气体制造 —— 属气体厂商的范围一；以化学品内含形式进入你的范围三类别 1，不得与自身使用重复计算。
1 · Medium-voltage switchgear. SF₆ GIS, ring main units, some breakers. Nameplate charge, annual leak rate, and end-of-life recovery. Specify SF₆-free alternatives (vacuum + dry air, or fluoronitrile mixtures) in every new build.
1 · 中压开关设备。SF₆ GIS、环网柜、部分断路器。记录铭牌充装量、年泄漏率与报废回收。所有新建项目应指定无 SF₆ 方案（真空 + 干燥空气，或氟腈混合气）。
2 · Cooling. Chillers, CRAH/CRAC DX units, heat-pump loops, CDUs. Refrigerant type, charge in kg, GWP, and a real leak rate — not the design assumption.
2 · 冷却系统。冷水机、CRAH/CRAC 直膨机组、热泵环路、CDU。记录制冷剂类型、充装公斤数、GWP，以及真实泄漏率 —— 而非设计假设值。
3 · Liquid & immersion cooling. Single-phase dielectric oils carry no F-gas issue. Two-phase fluoroketone/PFPE fluids do: fluid loss during maintenance is the emission, and PFAS phase-out is a supply risk.
3 · 液冷与浸没冷却。单相介电油无 F-gas 问题；两相氟代酮/PFPE 冷却液则有：维护期间的液体损失即为排放，且 PFAS 退出带来供应风险。
4 · Clean-agent fire suppression. HFC-227ea and HFC-125 cylinders. Emissions come from discharges, tests and decommissioning — never test-discharge a clean agent.
4 · 洁净气体消防。HFC-227ea 与 HFC-125 钢瓶。排放来自喷放、测试与拆除 —— 切勿以喷放方式测试洁净气体系统。
Not included: grid SF₆ upstream of your point of delivery (the utility's inventory), refrigerant embodied in purchased equipment before commissioning (supplier's), and CO₂-based suppression (not an F-gas).
不纳入：供电接入点上游电网的 SF₆（属电力公司清单）、设备投运前内含的制冷剂（属供应商）、以及 CO₂ 消防系统（非 F-gas）。
A brand cannot measure a fab. What it can do is (a) rank suppliers by spend × process intensity to find where the CO₂e actually is, (b) ask a gas-level questionnaire — consumption by gas, utilisation source, abatement coverage, DRE by gas, method tier, and (c) allocate to a product by die area or panel area rather than by revenue, and state the allocation rule in the product footprint. Revenue allocation hides exactly the thing you are trying to manage.
品牌无法直接测量一座晶圆厂。可行的做法是：(a) 按采购额 × 工艺强度对供应商排序，找出 CO₂e 的真实所在；(b) 发放气体级问卷 —— 分气体耗量、利用率来源、尾气处理覆盖率、分气体 DRE、方法层级；(c) 按裸片面积或面板面积（而非营收）分摊至产品，并在产品足迹中说明分摊规则。按营收分摊恰好会掩盖你想管理的东西。
Process schematic · click a node工艺示意图 · 点击节点
F-gas is only visible when you draw the path. Switch between three flows, then click any node — red nodes are hotspots. Each panel tells you what enters, what leaves, the control point, how to measure it, the benchmark, and the document an auditor will ask for.
只有画出路径，F-gas 才会显形。切换三条流程，点击任一节点 —— 红色节点为热点。每个面板说明：进什么、出什么、控制点在哪、如何测量、基准是多少、以及审计员会索取哪份文件。
Feeding C₂F₆ or CHF₃ into a plasma creates CF₄ that was never purchased. If you calculate emissions only from purchases of each gas, that CF₄ is invisible — and it is the gas your abater destroys least well. The IPCC Tier 2b method handles it with a separate by-product term: BPE = (1 − h) · B · FC · (1 − a·d), where B is kg of by-product per kg of gas fed. Ask your equipment or abatement supplier for measured B values on your recipes.
向等离子体中通入 C₂F₆ 或 CHF₃ 会生成从未采购过的 CF₄。如果只按各气体采购量计算排放，这部分 CF₄ 就是隐形的 —— 而它恰是尾气处理装置去除最差的气体。IPCC Tier 2b 方法用独立的副产物项处理：BPE = (1 − h) · B · FC · (1 − a·d)，其中 B 为每公斤投入气体生成的副产物公斤数。请向设备或尾气处理供应商索取针对你配方的实测 B 值。
Illustrative parameters. The point is structural, not the decimals: utilisation moves the answer far more than GWP does.参数为示意值。要点在于结构而非小数：利用率对结果的影响远大于 GWP。
An abater at 99% DRE that is bypassed 5% of production hours behaves like an abater at roughly 94% DRE. Effective removal ≈ coverage × availability × DRE. Report all three, or report none.
一台 DRE 99% 但在 5% 生产时间被旁通的处理装置，实际相当于约 94% 的 DRE。有效去除率 ≈ 覆盖率 × 可用率 × DRE。三者一并报告，否则不要报告。
Publish per-gas mass, per-gas CO₂e, the GWP set used, the method tier, abatement coverage and weighted DRE, plus the intensity metric (CO₂e per cm² of wafer out, or per m² of glass). Then move one tier up in method quality every year and restate the baseline when the method changes. Year-on-year progress on a fixed method is the only progress an auditor can see.
公布：分气体质量、分气体 CO₂e、所用 GWP 版本、方法层级、尾气处理覆盖率与加权 DRE，以及强度指标（每 cm² 出货晶圆或每 m² 玻璃的 CO₂e）。之后每年将方法质量提升一个层级，并在方法变更时重述基线。只有在固定方法下的逐年进展，才是审计员能看见的进展。
Mostly O₂/N₂ plasma with no GWP relevance, but some strip recipes add CF₄ or use fluorinated residue removers. Screen the chemical list once, document that it is immaterial, and move on — that documented screen is exactly what "not included" should look like.
大多为 O₂/N₂ 等离子，与 GWP 无关；但部分去胶配方会添加 CF₄ 或使用含氟残留清除剂。对化学品清单做一次筛查，记录其不具重要性即可 —— 这份留痕的筛查，正是"不纳入"应有的样子。
kg CO₂e (F-gas only) per m² of glass out. Normalise before you compare: a Gen 10.5 line and a Gen 6 line are not comparable in absolute tonnes. Put this metric in the supplier scorecard with a year-on-year improvement rate, not just a level — a high-intensity supplier improving 15% a year may be a better partner than a flat low-intensity one.
每 m² 出货玻璃的 kg CO₂e（仅 F-gas）。比较前先归一化：10.5 代线与 6 代线的绝对吨数不可比。将该指标纳入供应商记分卡，并同时看逐年改善率而非仅看水平值 —— 强度较高但每年改善 15% 的供应商，可能优于强度低却停滞不前的供应商。
Direct-to-chip and single-phase immersion using water/glycol or dielectric oil raise no F-gas question — say so in writing and close the item. Two-phase immersion with fluoroketones or PFPEs does: the emission is fluid loss during top-up, maintenance and rack extraction, and the substance is in scope of PFAS restrictions. With 3M exiting PFAS manufacturing by the end of 2025, fluid choice is now simultaneously a climate, compliance and supply-continuity decision.
使用水/乙二醇或介电油的芯片直冷与单相浸没冷却不涉及 F-gas 问题 —— 请以书面说明并结案。使用氟代酮或 PFPE 的两相浸没冷却则涉及：排放来自补液、维护与机柜取出时的液体损失，且该物质属 PFAS 限制范围。随着 3M 在 2025 年底前退出 PFAS 生产，冷却液选择已同时成为气候、合规与供应连续性的决策。
HFC-227ea (GWP 3,600) and HFC-125 cylinders emit only when discharged, tested by discharge, or decommissioned without recovery. Two rules cover almost all risk: never functional-test by discharging agent, and recover to a certified reclaimer at end of life. For new rooms, specify FK-5-1-12 or an inert gas system (IG-541 / N₂) and the F-gas line for that room becomes zero.
HFC-227ea（GWP 3,600）与 HFC-125 钢瓶仅在喷放、以喷放方式测试或未回收拆除时产生排放。两条规则可覆盖几乎全部风险：绝不以喷放药剂做功能测试；报废时交由持证再生商回收。新建机房应指定 FK-5-1-12 或惰性气体系统（IG-541 / N₂），该机房的 F-gas 条目即归零。
Equipment does not usually leak steadily; it leaks when it is opened. So the control point is a procedure, not a device: certified technicians, recovery cylinders weighed before and after, no venting to atmosphere, a signed logbook entry per intervention, and a reclaim certificate at decommissioning. If you can show the weighed-in / weighed-out record, you can defend the number.
设备通常不是持续泄漏，而是在被打开时泄漏。因此控制点是一套程序而非一台设备：持证技术人员、回收钢瓶作业前后称重、禁止向大气排放、每次作业签署记录簿、报废时取得再生证书。只要能出示"称进/称出"记录，数字就站得住。
Report F-gas as a named line inside Scope 1, in t CO₂e by gas, with the GWP set stated. For customer-facing numbers, allocate by IT load (per kW or per rack) rather than by floor area, and disclose the rule. An AI datacenter that grew 3× in load and held F-gas flat has a real story — but only if the intensity denominator is stated.
在范围一中以独立命名条目报告 F-gas，按气体列示 t CO₂e，并注明所用 GWP 版本。面向客户的数据应按 IT 负载（每 kW 或每机柜）而非按建筑面积分摊，并披露规则。负载增长 3 倍而 F-gas 持平的 AI 数据中心确有故事可讲 —— 但前提是说明强度分母。
Step by step · click through分步计算 · 逐步点开
The equation is not the hard part; the parameters are. Open step 1 and press "Next step" to walk the whole chain, then move the sliders to see how much of your reported number is chemistry and how much is choice.
难的不是公式，而是参数。展开第 1 步并点击"下一步"走完整条链路，然后拖动滑块，看看你报告的数字里有多少来自化学、有多少来自选择。
Equation structure per IPCC 2006 Guidelines, 2019 Refinement, Vol. 3 Ch. 6 (Electronics Industry) — Tier 2b form. See 12 Sources for the default parameter tables.公式结构依据 IPCC 2006 指南 2019 精细化版第 3 卷第 6 章（电子行业）Tier 2b 形式。默认参数表见 12 来源。
Tier 1 purchases × a default emission factor. Fast, defensible only as a screen.
Tier 2a by gas, with default parameters.
Tier 2b by gas and by process type (etch vs chamber clean), with by-products — the practical target for a fab or panel line.
Tier 2c / Tier 3 site-specific measured utilisation and measured DRE, verified at the stack.
Tier 1 采购量 × 默认排放因子。快，但仅可作为筛查。
Tier 2a 按气体，使用默认参数。
Tier 2b 按气体并按工艺类型（刻蚀 vs 腔体清洗），含副产物 —— 晶圆厂或面板线的现实目标。
Tier 2c / Tier 3 现场实测利用率与实测 DRE，并在烟囱端验证。
Why it matters for a brand: two suppliers reporting the same tonnage on different tiers are not comparable. Always collect the tier alongside the number.
对品牌为何重要：两家供应商在不同层级下报出的相同吨数并不可比。收集数字时必须同时收集层级。
A December cylinder delivery that is still full on 31 December is not an emission. This single correction is the most common first-year restatement.
12 月到货、12 月 31 日仍满瓶的钢瓶不构成排放。这一项修正是第一年最常见的数据重述。
Heel is gas-, pressure- and container-specific. If it goes back to the supplier and is reclaimed, it is not your emission — but you need the certificate, not the assumption.
残气量取决于气体种类、压力与容器型式。若返还供应商并被回收，则不计为你的排放 —— 但你需要的是证书，而不是假设。
U differs by gas, by process type and by tool generation. Use the IPCC default table by process class until you can measure your own with an abater-inlet FTIR campaign — then say in the method note which tools are measured and which are defaulted.
U 因气体、工艺类型与设备世代而异。在能通过处理装置进口 FTIR 实测之前，先按工艺类别使用 IPCC 默认表 —— 之后在方法说明中写明哪些设备为实测、哪些为默认。
Small in mass, awkward in audit: it is the term most often missing entirely. Include it even with a default B, and flag it as a default.
质量不大，但在审计中很棘手：它是最常被完全遗漏的一项。即使用默认 B 也要纳入，并标注为默认值。
Read that again: same fab, same wafers, same chemistry — an 86% difference in the reported number, produced entirely by coverage and destruction efficiency. This is the answer to "how far could it go".
请再看一遍：同一座厂、同样的晶圆、同样的化学 —— 报告数字相差 86%，完全由覆盖率与销毁效率决定。这就是"到底能做到多远"的答案。
Absolute tonnes drive the target; intensity makes it comparable year to year and supplier to supplier. Publish both, with the denominator defined. For scale: 305 t CO₂e is roughly the annual tailpipe CO₂ of 66 typical passenger cars, using EPA's ~4.6 t/vehicle/year equivalency.
绝对吨数决定目标；强度使其可跨年度、跨供应商比较。两者都要公布，并定义分母。作为量级参照：按 EPA 约 4.6 t/辆·年 的等效值，305 t CO₂e 约相当于 66 辆典型乘用车一年的尾气 CO₂。
305 t CO₂e reported F-gas emission from this one gas stream该单一气流的 F-gas 报告排放量
Single-gas, single-stream illustration excluding by-product formation. Use it to build intuition, not to file a report.单一气体、单一气流的示意计算，未含副产物生成。用于建立直觉，不可直接用于报告。
What is good, what is outstanding什么算好，什么算优秀
A supplier answer only becomes useful when you can place it. These four bands are this study's construct for doing that consistently; calibrate them against published fab and datacenter disclosures before you contract on them.
只有能定位，供应商的回答才有用。这四个等级是本研究为统一定位而设的框架；在写入合同前，请用公开的晶圆厂与数据中心披露数据加以校准。
| Metric指标 | Entry入门 | Managed受控 | Good良好 | Outstanding优秀 |
|---|---|---|---|---|
| 1 Abatement coverage (F-gas tool exhausts)尾气处理覆盖率 （含 F-gas 设备尾气） |
<70% | 70–90% | 90–99% | 100% + interlocks100% + 联锁 |
| 2 Weighted DRE, all F-gases全部 F-gas 加权 DRE | <85% | 85–92% | 92–97% | >97%, FTIR-verified>97%，FTIR 验证 |
| 3 CF₄-specific DRE (the acceptance test)CF₄ 专项 DRE （验收考题） |
<70% | 70–85% | 85–93% | >95% |
| 4 Chamber clean design腔体清洗方案 | In-situ PFC, timed原位 PFC、定时 | Mixed fleet混合配置 | Majority remote NF₃多数为远程 NF₃ | >90% remote NF₃ + endpoint control>90% 远程 NF₃ + 终点控制 |
| 5 Method & data quality方法与数据质量 | Spend or Tier 1按金额或 Tier 1 | Tier 2a | Tier 2b + by-productsTier 2b + 副产物 | Tier 2c/3, stack-reconciled ±10%Tier 2c/3，烟囱对账 ±10% |
| 6 Supplier data coverage (brand view, % of relevant spend)供应商数据覆盖 （品牌视角，占相关采购额） |
<20% | 20–50% | 50–80% | >80% gas-level + third-party assurance>80% 分气体 + 第三方核证 |
| 7 Datacenter refrigerant leak rate数据中心制冷剂泄漏率 | Unknown未知 | >5%/yr | 2–5%/yr | <2%/yr + low-GWP new plant<2%/年 + 新建采用低 GWP |
| 8 SF₆ switchgearSF₆ 开关设备 | No register无台账 | Register + monitors台账 + 监测 | New builds SF₆-free新建无 SF₆ | SF₆-free fleet + ≥95% recovery records全场无 SF₆ + ≥95% 回收记录 |
Each band boundary is placed where a different kind of work begins, not at a round number. Moving from Entry to Managed is an inventory and register exercise. Managed to Good is a capital and recipe exercise. Good to Outstanding is a measurement and assurance exercise — which is why the last step costs the least tonnes and the most credibility.
每条分界线都画在工作性质发生变化之处，而非某个整数。从"入门"到"受控"，是清单与台账的工作；从"受控"到"良好"，是资本与配方的工作；从"良好"到"优秀"，是测量与核证的工作 —— 这也是为什么最后一步减吨最少、却最能带来可信度。
To calibrate: pull the F-gas / PFC section of three public reports in your own segment — a leading logic foundry, a large panel maker, and a hyperscale operator — and locate their disclosed coverage, DRE and intensity on this ladder. Where a disclosure sits above your top band, raise the band. That is how a benchmark stays honest.
校准方法：取本行业三份公开报告中的 F-gas / PFC 章节 —— 一家领先逻辑代工厂、一家大型面板厂、一家超大规模运营商 —— 将其披露的覆盖率、DRE 与强度定位到本梯度上。若某项披露高于你的最高档，就上调该档。这样基准才能保持诚实。
Interactive screen · tick what is true交互筛查 · 勾选符合项
"So many items, so many works linked and mixed" — the way out is a screen you can defend. Tick the conditions that are true for your site or supplier and read the verdict. A "no" answer with a date and a signature is a legitimate result.
"条目太多、工作彼此交错" —— 出路是一套站得住的筛查。勾选对你的场地或供应商成立的条件，读取结论。带日期与签名的"否"同样是合法结果。
Verdict结论
Nothing ticked yet. If that is genuinely the case, the correct output is not silence: write a one-page screening note with these seven answers, the date and a signature, mark the topic "assessed, not material", and re-screen next year. That note is itself audit evidence.
尚未勾选任何项。若确实如此，正确的产出不是沉默：写一页筛查说明，记录这七项回答、日期与签名，将该议题标注为"已评估、不具重要性"，并在下一年重新筛查。这份说明本身就是审计证据。
Any process gas with GWP ≥ 150 that is consumed; any equipment charge over the local leak-check threshold; anything a regulator, customer or ecolabel explicitly asks for; every unabated F-gas tool exhaust; and by-product CF₄ wherever C₂F₆, CHF₃ or c-C₄F₈ is used.
任何被消耗的 GWP ≥ 150 工艺气体；任何超过当地检漏阈值的设备充装量；监管方、客户或生态标签明确要求的任何项；每一处未经处理的 F-gas 设备尾气；以及在使用 C₂F₆、CHF₃ 或 c-C₄F₈ 之处必然产生的副产物 CF₄。
Lab-scale and calibration-gas cylinders under ~100 t CO₂e in total; single-phase liquid cooling with no fluorinated fluid; CO₂ or inert suppression; acid-gas-only exhausts. Defer means "screened, quantified roughly, documented, dated, and re-screened annually" — it never means "not looked at".
合计低于约 100 t CO₂e 的实验室级与标准气钢瓶；不含氟化液的单相液冷；CO₂ 或惰性气体消防；仅含酸性气体的尾气。延后意味着"已筛查、已粗算、已留档、已注明日期、并按年重筛"，绝不意味着"没看过"。
Drag to rotate · structure explains behaviour拖动旋转 · 结构解释行为
Names like c-C₄F₈ are hard to hold in your head; shapes are not. Rotate each molecule and read what its geometry tells you about the abater you need.
c-C₄F₈ 这样的名字很难记住，形状却不难。旋转每个分子，读懂它的几何结构对"你需要什么样的尾气处理"意味着什么。
C–F is the strongest single bond in organic chemistry, roughly 480–545 kJ/mol depending on the molecule. Fully fluorinated, symmetric molecules therefore have no weak point for OH radicals, sunlight or rain — which is why lifetimes run to millennia and why destruction needs energy, not filtration. N–F in NF₃ is far weaker, which is why NF₃ both reacts well in the chamber and abates easily. Structure is not trivia here; it is the reason your DRE table looks the way it does.
C–F 是有机化学中最强的单键，视分子而定约为 480–545 kJ/mol。因此全氟化、对称的分子对 OH 自由基、阳光与雨水都没有弱点 —— 这正是其寿命长达千年、且销毁需要能量而非过滤的原因。NF₃ 中的 N–F 弱得多，所以 NF₃ 既能在腔体内充分反应，也易于被处理。这里的结构不是冷知识，而是你的 DRE 表长成那样的原因。
Year to year · a real sequence逐年推进 · 真实序列
You cannot target what you have not mapped, and you cannot verify what you have not targeted. Each year has one deliverable set and one evidence set — that is what makes progress visible year on year.
未映射就无法设目标，未设目标就无法验证。每一年只有一套交付物和一套证据 —— 这正是让进展逐年可见的原因。
Do: list every tier-1 and known tier-2 fab, panel line, packaging house and ODM. Rank by spend × process intensity. Issue one gas-level questionnaire: consumption by gas, process split (etch vs clean), utilisation source, abatement coverage by tool count, DRE by gas with test date, method tier, GWP set. Build the internal charge register for switchgear, chillers, immersion fluid and suppression on your own sites.
做：列出全部一级及已知二级晶圆厂、面板线、封装厂与 ODM，按采购额 × 工艺强度排序。发放一份分气体问卷：分气体耗量、工艺拆分（刻蚀 vs 清洗）、利用率来源、按设备台数的处理覆盖率、带测试日期的分气体 DRE、方法层级、GWP 版本。同时建立自有场地的开关柜、冷水机、浸没液与消防充装台账。
Evidence: supplier list with coverage %, signed questionnaires, the charge register, and a written method note naming every default you used.
证据：带覆盖率百分比的供应商清单、已签署问卷、充装台账，以及列明所有所用默认值的书面方法说明。
Do: verify the top ten suppliers by CO₂e — on site or by document review: abater register versus tool list, DRE test reports at production flow, alarm and bypass logs. Set the baseline year and an intensity target (CO₂e per cm², per m², per kW). Put F-gas clauses into contracts and a weighting into the procurement scorecard. Align the target with your science-based target boundary.
做：对 CO₂e 排名前十的供应商进行验证 —— 现场或文件审核：处理装置台账与设备清单核对、生产流量下的 DRE 测试报告、报警与旁通日志。设定基线年与强度目标（每 cm²、每 m²、每 kW 的 CO₂e）。将 F-gas 条款写入合同，并在采购记分卡中赋予权重。使目标与科学碳目标边界一致。
Evidence: verification reports with findings and dates, contract clause text, scorecard weighting, baseline restatement memo.
证据：含发现事项与日期的验证报告、合同条款文本、记分卡权重、基线重述备忘。
Do: convert chamber cleans to remote NF₃ where not yet done; retrofit abatement to reach full coverage with interlocks; upgrade CF₄ destruction specifically; specify SF₆-free switchgear and low-GWP refrigerant in every new datacenter build; decide the two-phase immersion fluid question ahead of the PFAS transition; move one method tier up and reconcile with a stack survey.
做：对尚未改造的腔体清洗切换为远程 NF₃；改造尾气处理以实现全覆盖并加装联锁；专项提升 CF₄ 销毁能力；在所有新建数据中心指定无 SF₆ 开关柜与低 GWP 制冷剂；在 PFAS 过渡之前决定两相浸没液方案；将方法提升一个层级并以烟囱普查对账。
Evidence: project register with kg and t CO₂e avoided per project, commissioning reports, updated specifications, restated inventory with the new tier declared.
证据：逐项列明避免公斤数与 t CO₂e 的项目台账、调试报告、更新后的规格书、声明新层级的重述清单。
Then repeat with a tighter boundary: tier-2 suppliers, packaging, substrates, and the fluids in your liquid-cooling fleet随后以更严的边界重复：二级供应商、封装、基板，以及液冷机群中的工作流体
Tick it · it scores live勾选 · 实时计分
Auditable means one thing: a stranger can reproduce your number from your paperwork. Tick what you actually hold today — the honest score is more useful than a full one.
"可审计"只有一个含义：陌生人能仅凭你的文件复现你的数字。勾选你今天真正拥有的项 —— 诚实的分数比满分更有用。
They pick one gas, one month and one tool. Then: invoice → weigh ticket → inventory sheet → workbook cell → parameter source → abater DRE report → alarm log for that month → reported tonne. If any link is missing, the whole figure drops to "estimated". Build the chain for one gas first and the rest becomes copy-paste.
他们会挑一种气体、一个月份、一台设备，然后顺链核查：发票 → 称重记录 → 库存表 → 工作簿单元格 → 参数来源 → 处理装置 DRE 报告 → 该月报警日志 → 报告吨数。任一环节缺失，整个数字都会被降级为"估算"。先为一种气体把链条打通，其余便是复制粘贴。
Regulatory & programme hooks法规与项目抓手
Requirements change; check the current text at each link before you commit to a claim. What follows is the shape of the ask, not legal advice.
要求会变化；在作出任何声明前，请以各链接的最新条文为准。以下呈现的是"要求的形状"，不构成法律意见。
EPEAT registration for computers and displays runs on criteria administered by the Global Electronics Council, including climate-related criteria addressing greenhouse gas reduction in the supply chain and, in the Climate+ track, verified manufacturing emissions. For an AI PC brand the practical translation is: you must be able to show, with evidence, that named suppliers report and reduce process F-gas emissions.
计算机与显示器的 EPEAT 注册依据由全球电子委员会（GEC）管理的条款，其中包含针对供应链温室气体减排的气候相关条款，Climate+ 路径还要求经核证的制造排放数据。对 AI PC 品牌而言，实际含义是：你必须能以证据证明，被指名的供应商在报告并减少工艺 F-gas 排放。
Prepare: supplier attestation letters, gas-level data, the allocation rule, and the assurance statement — the same pack as section 10, group F.
准备：供应商声明函、分气体数据、分摊规则、核证声明 —— 与第 10 节 F 组同一套资料。
epeat.net — registry & criteria— 注册库与条款 · globalelectronicscouncil.org
Covers HFCs, PFCs, SF₆, NF₃ and others: containment and leak-check duties by charge in CO₂e, record-keeping and logbooks, technician certification, recovery obligations, HFC quota phase-down, and dated bans including SF₆ in new switchgear. Directly relevant to any datacenter, office or fab facility in the EU — and a sensible global internal standard elsewhere.
涵盖 HFCs、PFCs、SF₆、NF₃ 等：按充装 CO₂e 规定的密闭与检漏义务、记录与记录簿、技术人员认证、回收义务、HFC 配额递减，以及包括新建开关设备禁用 SF₆ 在内的分期禁令。对欧盟境内的数据中心、办公与厂区直接适用 —— 在其他地区亦可作为合理的全球内部标准。
The most prescriptive public rulebook for fab F-gas accounting: which gases, which process types, how to derive utilisation and by-product factors, and how abatement credit must be substantiated. Even outside the US it is the best available template for what "defensible" looks like at tool level.
目前最具规定性的晶圆厂 F-gas 核算公开规则集：涵盖哪些气体、哪些工艺类型、如何推导利用率与副产物因子，以及尾气处理减排量必须如何举证。即便在美国以外，它也是"何为可辩护"的最佳设备级模板。
ESRS E1 requires gross Scope 1, 2 and 3 with methods and, in practice, disaggregation where a gas is material — F-gas in a fab or a datacenter qualifies. The GHG Protocol Scope 3 standard governs how you collect supplier data and how you may allocate. SBTi target boundaries include these emissions; CDP asks the question directly in its climate questionnaire.
ESRS E1 要求披露范围一、二、三的总量及方法，实践中对具有重要性的气体需拆分列示 —— 晶圆厂或数据中心的 F-gas 即属此类。GHG Protocol 范围三标准规范了如何收集供应商数据与如何分摊。SBTi 的目标边界涵盖此类排放；CDP 在气候问卷中直接提问。
ESRS delegated regulation授权法规 · GHG Protocol Scope 3 · SBTi Net-Zero · CDP
The World Semiconductor Council has run successive voluntary PFC emission-reduction commitments for the industry, expressed as normalised emission rates against a baseline year — a useful precedent for how to write a supply-chain target that survives growth. The Kigali Amendment phases down HFC production and consumption globally, which is what is quietly re-pricing every refrigerant and clean-agent decision you make in a datacenter.
世界半导体理事会（WSC）为行业先后推出多轮自愿性 PFC 减排承诺，以相对基线年的归一化排放率表述 —— 这是"如何写出能在增长中依然成立的供应链目标"的有用先例。《基加利修正案》在全球逐步削减 HFC 的生产与消费，正在悄然改变你在数据中心所做的每一项制冷剂与洁净气体决策的成本结构。
World Semiconductor Council · Kigali Amendment · CARB (SF₆ switchgear & refrigerant programmes)（SF₆ 开关柜与制冷剂项目）
Sources来源
Primary sources for the GWP values, lifetimes, calculation equations, regulatory thresholds and equivalency factors used on this page. Where this study has invented a band or a parameter, it says so in the text.
本页所用 GWP 数值、大气寿命、计算公式、法规阈值与等效因子的一手来源。凡本研究自设的等级或参数，均已在正文中注明。