The footprint is small. The leverage is enormous.
足迹不大，杠杆极大Chipmaking is, on a planetary balance sheet, a rounding error — on the order of a few tenths of a percent of global electricity. But it is a rounding error with three unusual properties, and each of them is why this sector gets outsized attention.
从全球总量看，芯片制造只是零头——大约占全球用电的零点几个百分点。但它是一个具备三种特殊性质的零头，这正是它被高度关注的原因。
One — it is concentrated. Advanced capacity sits on a handful of grids (Taiwan, South Korea, Japan, and increasingly Arizona, Saxony, Kumamoto). A fab cannot decarbonize faster than the wires it is plugged into. Two — it is chemical. Scope 1 is dominated by fluorinated gases with GWPs in the thousands to tens of thousands, abatable today with known equipment. Three — it is multiplying. Each node adds process steps, EUV layers and energy per wafer, while the demand curve from AI compute bends steeply upward.
其一，高度集中——先进产能集中在少数几张电网上（台湾、韩国、日本，以及正在崛起的亚利桑那、萨克森、熊本）。晶圆厂的脱碳速度，不可能快过它所连接的那张电网。其二，本质是化学——范围一排放以含氟气体为主，GWP 高达数千至数万倍，但今天已有成熟的减排设备可用。其三，正在放大——每前进一个节点，工序、EUV 层数与单片晶圆能耗都在增加，而 AI 算力带来的需求曲线正陡峭上行。
2026 – 2035
用已知技术，把已知的排放清掉Global — the no-excuse decade / 全球：没有借口的十年
- Point-of-use abatement retrofitted across all etch and chamber-clean tools; destruction-and-removal efficiency reported per tool, not per site. 在全部刻蚀与腔体清洗设备上加装点位式（POU）尾气处理，并按设备而非按厂区披露破坏去除效率。
- NF₃ substitution and remote-plasma clean optimisation; SF₆ phase-down in legacy lines. NF₃ 替代与远程等离子清洗优化；成熟制程产线逐步淘汰 SF₆。
- Electricity contracting moves from unbundled certificates to additional, hourly-matched clean power. This is the single largest credibility test of the decade. 电力采购从"非捆绑绿证"转向"具备额外性、逐小时匹配"的清洁电力——这是这十年最大的可信度考验。
- Facility systems (chillers, make-up air units, exhaust, dry pumps in idle) treated as a capital programme, not a maintenance line item. Typically 40–50% of fab site energy. 厂务系统（冰机、外气空调箱、排气、干泵待机）应作为资本项目而非维修科目管理——通常占厂区能耗的 40–50%。
- Scope 3 gets a real inventory: wafers, ultrapure chemicals, specialty gases, and — the blind spot — the embodied carbon of the equipment itself. 范围三建立真实清单：晶圆、超纯化学品、特气，以及最大的盲区——设备本身的隐含碳。
2030 – 2035 — from efficiency to structure / 从效率走向结构
- Industrial heat pumps and fab waste-heat export to district or neighbouring process loads (Northern Europe leads; Asia largely untapped). 工业热泵与晶圆厂余热外供（区域供热或邻近工艺负荷）——北欧领先，亚洲基本尚未开发。
- Clean firm power procurement: geothermal, long-duration storage, and first SMR contracts around fab clusters. 清洁"可调度"电力采购：地热、长时储能，以及围绕晶圆厂集群的首批小型模块化反应堆合约。
- Carbon becomes a design variable — chiplets, advanced packaging and node-choice evaluated on carbon-per-function, not only cost-per-transistor. 碳成为设计变量——小芯片、先进封装与节点选择开始以"单位功能碳排"而非仅"单位晶体管成本"衡量。
- Rare-gas (neon, helium) recovery and tool refurbishment mature into a real circular market. 稀有气体（氖、氦）回收与设备翻新，发展为真正的循环市场。
2035 – 2050
剩下的都是难题：残余排放与供应链What is left when the easy 70% is gone / 当容易的 70% 做完之后
- Residual process emissions. Some chemistries have no drop-in substitute; abatement efficiency asymptotes below 100%. Expect a hard floor requiring either new chemistry or verified removals. 残余工艺排放：部分化学体系没有直接替代品，减排效率无法达到 100%。将出现一条硬底线，需要新化学或经核证的碳移除来处理。
- Upstream materials. Low-carbon polysilicon, green-hydrogen-based chemicals, electrified ultrapure water and gas plants. This is heavy industry's problem arriving inside the chip's ledger. 上游材料：低碳多晶硅、以绿氢为基础的化学品、电气化的超纯水与气体工厂——重工业的难题最终落到芯片的账本上。
- Equipment embodied carbon. A modern fab is thousands of tonnes of steel, aluminium and rare materials. Green steel and refurbishment become procurement criteria. 设备隐含碳：一座现代晶圆厂由数千吨钢、铝与稀有材料构成。绿色钢材与设备翻新将成为采购标准。
- Grid, not fab. By 2040 the marginal tonne is decided by transmission build-out, permitting speed and firm clean capacity in Taiwan, Korea, Japan and the US Southwest — not by anything a process engineer can do. 决定因素是电网而非厂房：到 2040 年，边际减排量取决于台湾、韩国、日本与美国西南部的输电建设、审批速度与可调度清洁电力，而不是制程工程师还能做什么。
- Verification. Product-level carbon footprints, audited and comparable between suppliers, become a commercial gate — the way yield and defect density are today. 核验：可审计、可跨供应商比较的产品级碳足迹，将像今天的良率与缺陷密度一样，成为商业准入门槛。
The same fab, four different carbon numbers.
同一座晶圆厂，四个不同的碳数字Grid carbon intensity is the dominant variable in a fab's Scope 2. Identical process, identical yield, sited on different grids, produces wafers with materially different footprints. Values below are approximate 2023–24 order-of-magnitude figures for orientation — check the live source before citing.
电网碳强度是晶圆厂范围二排放的主导变量。相同制程、相同良率，坐落于不同电网，产出晶圆的碳足迹显著不同。下表为 2023–24 年的量级参考值，引用前请核对原始数据源。
| Region 区域 | Grid intensity 电网碳强度 ≈ gCO₂/kWh | Structural constraint 结构性约束 | What would move it 关键突破口 |
|---|---|---|---|
| Taiwan 台湾 | ≈ 490–500 | Island grid, limited interconnection, gas-heavy transition 孤岛电网、互联受限、以天然气为主的过渡 | Offshore wind build-out, corporate PPA market depth, storage 离岸风电、企业购电市场深度、储能 |
| South Korea 韩国 | ≈ 430–450 | Single-buyer power market, slow direct-PPA liquidity 单一购电体制，直接购电流动性不足 | Direct PPA reform, nuclear + renewables mix 直接购电改革、核电与可再生能源组合 |
| Japan 日本 | ≈ 470–490 | Post-2011 fossil dependence, regional grid fragmentation 2011 年后化石依赖、区域电网割裂 | GX policy execution, interregional transmission 绿色转型政策落地、跨区输电 |
| United States 美国 | ≈ 350–380 | Interconnection queues, transformer & transmission lead times 并网排队、变压器与输电工期 | Permitting reform, firm clean power, CHIPS conditionality 审批改革、可调度清洁电力、法案附加条件 |
| European Union 欧盟 | ≈ 230–260 | Highest power price, industrial competitiveness pressure 电价最高、产业竞争力压力 | Cheapest carbon per wafer already; waste-heat reuse 单位晶圆碳排已最低；重点在余热利用 |
| Mainland China 中国大陆 | ≈ 530–560 | Coal-weighted baseload, rapid but uneven renewable build 以煤为主的基荷、可再生装机迅猛但不均衡 | Green power trading, provincial siting, mature-node efficiency 绿电交易、省域选址、成熟制程能效 |
| SE Asia / India 东南亚与印度 | ≈ 400–700 | New packaging & test capacity on carbon-heavy grids 新增封测产能落在高碳电网上 | Design-in clean power from day one, not retrofitted 从第一天就设计清洁电力，而非事后补救 |
Eight areas, ranked by how fast they pay back.
八个方向，按回收速度排序Payback ranges below are indicative, synthesised from public sustainability disclosures, equipment-vendor case studies and industry consortium material. They vary enormously by site, power price and utilisation. Treat them as a screening order, never as a business case — and note the confidence flag on each.
下列回收期为指示性数值，综合自公开可持续发展披露、设备厂商案例与产业联盟资料，并会因厂区、电价与稼动率而巨大差异。请将其作为筛选顺序，而非商业论证依据，并留意每项的置信度标记。
Facility & HVAC optimisation
Chiller sequencing, make-up air setpoints, exhaust rebalancing, variable-speed dry pumps. The cheapest tonne in any fab, and the one most often deferred.
SEMI SCC ↗Point-of-use F-GHG abatement
Highest tonnes-per-dollar in Scope 1 because of GWP multipliers. Capex-heavy, operationally proven, and increasingly a regulatory expectation rather than a choice.
IPCC AR6 GWP ↗Waste-heat recovery & heat pumps
Economics depend entirely on a heat off-taker nearby. Excellent where district heating exists; stranded where it does not. Policy, not physics, is the constraint.
IEA Industry ↗Water reclaim & near-ZLD
Leading fabs report reclaim rates above 85–90%. The return is partly carbon, mostly resilience: drought is now a production risk, not an ESG topic.
TSMC ESG ↗Hourly-matched clean PPAs
Often cost-neutral to positive over a 10–15 year term in liquid markets; a premium in single-buyer markets. The ROI is risk transfer, not a payback number.
RE100 ↗Clean firm power (geothermal, SMR, LDES)
Fabs need 24/7 stability that intermittent supply alone cannot give. Early contracts are being signed by hyperscalers; fabs are watching, not yet leading.
IRENA ↗Design-for-carbon & LCA tooling
imec's sustainable-semiconductor modelling lets a design team see the carbon consequence of a node or packaging choice before tape-out. No payback formula — it changes the denominator.
imec SSTS ↗Circularity: rare gas, tools, solvents
Neon and helium recovery began as a geopolitical hedge and turned out to be a carbon lever. Refurbished tools cut embodied carbon far more than any efficiency tweak.
SEMI ↗Who is paying, who is adopting, who is watching.
谁在投入，谁在采用，谁在紧盯Three different roles are often confused. Capital is deployed by the fabs and equipment makers themselves — this is a self-funded transition, not a venture-funded one. Adoption is led by whoever has the most electricity at stake. Attention comes from downstream customers who inherit the footprint as Scope 3.
三种角色常被混为一谈。资本由晶圆厂与设备商自行投入——这是一场自筹资金的转型，而非风投驱动。采用由用电量最大的一方带头。关注来自下游客户——他们把这份足迹继承为自己的范围三。
TSMC
The largest single industrial electricity buyer in the sector. Net zero by 2050, RE100 target brought forward to 2040, sector-leading water reclaim. When TSMC moves the Taiwanese renewable market, the whole supply chain follows.
esg.tsmc.com ↗Samsung Electronics & SK hynix
Memory is energy-intensive and Korea's grid is carbon-heavy — a hard combination. Both carry 2050 net-zero and RE100 commitments; the binding constraint is national power-market reform, not corporate intent.
Samsung Semiconductor ↗Intel
Among the earliest with a chemical-abatement and renewable-electricity programme at scale; targets net-zero Scope 1&2 by 2040 and has pushed supplier engagement as a formal programme rather than a survey.
Intel CR ↗ASML · Applied Materials · Lam · TEL
Equipment makers decide the energy floor of every future fab. Idle-mode power, vacuum efficiency and abatement integration designed into the tool beat anything retrofitted later. ASML targets net zero across its value chain by 2040.
ASML ↗ · Applied ↗imec & the research base
Neutral ground. imec's Sustainable Semiconductor Technologies and Systems programme produced the shared modelling language the industry now argues in — the closest thing to an agreed method.
imec SSTS ↗Hyperscalers & device OEMs
Apple, Microsoft, Google, Amazon, Meta. They do not run fabs; they inherit fab emissions in their Scope 3. Their procurement requirements have moved the industry faster than any regulation to date.
CDP disclosure ↗Policy: CHIPS Act, EU Chips Act, GX
Public capital is now the largest new entrant to fab financing — and increasingly attaches environmental conditions to it. Where conditionality is real, it is the fastest lever in this entire document.
NIST CHIPS ↗ · EU ↗Semiconductor Climate Consortium
Formed under SEMI, ~100+ members across the value chain, aligned on net zero by 2050 and — more usefully — on common reporting. The place to check whether a claim is industry-agreed or a press release.
SEMI SCC ↗Five levers decide the curve. Everything else is detail.
五个杠杆决定曲线，其余皆是细节If a market analysis of "value-advantaged sustainable semiconductors" reduces to anything, it reduces to these five. Ranked by the size of the tonne they control, not by how easy they are to talk about.
如果对"价值优势型可持续半导体"的市场分析必须浓缩为几点，那就是这五点。排序依据是它们所掌控的减排量级，而非讨论起来的难易程度。
Clean firm power at fab scale
Scope 2, ~60–75% of a leading fab's operational emissions. Hourly matching and additionality — not certificates — are what will survive audit in 2035.范围二，占先进晶圆厂运营排放约 60–75%。到 2035 年经得起审计的，是逐小时匹配与额外性，而非绿证。
Process chemistry & F-GHG
Scope 1. High GWP, high abatability, immediate. The one place where the industry can genuinely act alone, this decade, without waiting for anyone.范围一。高 GWP、高可减排性、可立即执行。这是产业唯一能在本十年内独立行动、无需等待他人的领域。
Facility systems & water
The unglamorous 40–50%: cleanroom air, chillers, exhaust, ultrapure water. Fastest payback, lowest board attention. That mismatch is itself a finding.不起眼的 40–50%：洁净室空气、冰机、排气、超纯水。回收最快，但董事会关注最少——这种错配本身就是一个发现。
Embodied carbon in the supply chain
Scope 3 upstream is usually the largest bar on the chart and the least trustworthy number on it. Materials, gases, chemicals — and the fab equipment itself.上游范围三通常是图表上最高的一根柱，也是最不可信的一个数字：材料、气体、化学品，以及晶圆设备本身。
Carbon productivity of the chip in use
The only lever with a negative sign. A power semiconductor that makes an EV inverter or a grid converter 3% more efficient can avoid, over its life, far more carbon than its manufacture emitted. But this must be measured honestly — "avoided emissions" is the industry's most abused metric, and until it is standardised and third-party verified it should be reported separately from the inventory, never netted against it.唯一带负号的杠杆。一颗让电动车逆变器或电网变流器效率提升 3% 的功率半导体，其生命周期内避免的碳排可远超制造排放。但必须诚实计量——"避免排放"是本产业被滥用最严重的指标；在其标准化并经第三方核证之前，应单独披露，绝不可与排放清单相抵扣。
On track / needs calling out
在轨道上 / 需要点名On track · 在轨道上
- F-GHG abatement deployment含氟气体减排设备部署Technically solved, widely deployed at leading fabs, and now a procurement expectation.
- Disclosure quality披露质量Semiconductor CDP reporting is among the better industrial sectors — comparable, annual, increasingly assured.
- Sector coordination产业协同The SEMI Semiconductor Climate Consortium gave the value chain a single table. Rare, and valuable.
- Water stewardship水资源管理Reclaim rates above 85–90% at leading fabs. Driven by drought risk, but the outcome is real.
- Shared measurement method共同的计量方法imec's modelling work turned "our number vs your number" into a debate with common units.
Needs calling out · 需要点名
- Unbundled certificates still count as clean非捆绑绿证仍被计为清洁电力Annual certificate matching flatters Scope 2 without changing a single grid dispatch decision. Hourly matching is the honest bar.
- Scope 3 is estimated, not measured范围三是估算而非实测The largest number in most inventories rests on industry-average factors and supplier surveys. Weakest data, biggest bar.
- Grid build-out is the true bottleneck电网建设才是真正瓶颈Taiwan, Korea and Japan cannot deliver corporate targets on current permitting and transmission timelines. This is not a fab problem, and fabs cannot fix it alone.
- Equipment embodied carbon is unowned设备隐含碳无人认领Fabs treat it as supplier scope; suppliers treat it as customer scope. It falls between the two and gets counted by neither with rigour.
- Advanced packaging & OSAT visibility先进封装与封测环节的透明度Growing fast, energy-intensive, often on carbon-heavy grids, and far less disclosed than front-end fabs.
- "Avoided emissions" inflation"避免排放"被夸大Claimed enthusiastically, verified rarely, and occasionally netted against actual emissions. This is the credibility risk of the next five years.
- No agreed sector decarbonization pathway缺乏公认的行业脱碳路径Unlike steel or cement, semiconductors lack a widely adopted sectoral 1.5°C pathway — so "aligned" means whatever each company decides it means.
More than the chip.
不止于芯片Here is the uncomfortable conclusion of this benchmark. The semiconductor industry is, on most measures, decarbonizing faster than the economy around it — and it will still miss, because the thing it cannot control is the thing that decides the answer: how quickly clean, firm electricity arrives on the specific grids where fabs sit.
这份对标得出的结论并不舒服：以多数指标衡量，半导体产业的脱碳速度快于它所处的整体经济——但它仍可能失手，因为决定结果的关键变量，恰恰是它无法控制的：清洁且可调度的电力，能多快抵达晶圆厂所在的那几张电网。
So the honest question is not "how do we make the fab greener?" It is: how does the chip make everything else decarbonize faster than the chip itself? Power electronics for grids and EVs. Sensors that make buildings and factories legible. Compute that shortens the design cycle of a battery or a catalyst. That is where a semiconductor company's carbon leverage exceeds its carbon footprint by orders of magnitude — provided it stops using that leverage as an excuse to move slowly on its own emissions.
因此，真正该问的不是"如何让晶圆厂更绿"，而是：芯片如何让其他一切的脱碳速度，快过芯片本身？用于电网与电动车的功率电子；让建筑与工厂变得"可读"的传感器；把电池或催化剂设计周期缩短的算力。在这些地方，一家半导体企业的碳杠杆比它的碳足迹大出数个量级——前提是，它不把这份杠杆当作自身减排缓慢的借口。
Next raw materials worth opening → R01 (IEA Energy and AI), R06 (imec SSTS method), R04 (SCC common reporting)
Open the raw material.
请打开原始资料Every figure on this page should be re-checked at source before it is used in a decision. Where a number is directional, it is flagged as such above.
本页所有数据在用于决策前都应回到原始来源核对；凡属方向性判断，上文已标注。
- IEA — Energy and AIData-centre electricity demand trajectory to 2030 国际能源署：能源与人工智能
- IEA — IndustryIndustrial energy & heat decarbonization 工业部门能源与热能脱碳
- IPCC AR6 Working Group IIIGlobal warming potentials (NF₃, SF₆, CF₄) 全球增温潜势
- SEMI — Semiconductor Climate ConsortiumSector net-zero alignment and common reporting 产业净零协同与统一披露
- TSMC ESGElectricity, water reclaim, RE100 2040, net zero 2050 台积电永续报告
- imec — Sustainable Semiconductor Technologies and SystemsNode-level carbon modelling method 节点级碳排建模方法
- EmberGrid carbon intensity by country 各国电网碳强度
- Samsung Semiconductor — SustainabilityMemory manufacturing targets 存储制造脱碳目标
- Intel — Corporate ResponsibilityScope 1&2 net zero 2040, supplier programme
- ASML — SustainabilityTool energy, value-chain net zero 设备能耗与价值链净零
- IRENARenewable capacity, cost and firm-power outlook 可再生能源展望
- Applied Materials — Corporate ResponsibilityEquipment energy & abatement 设备能效与减排
- Infineon — SustainabilityPower semiconductors & enabled savings 功率半导体与使能减排
- Science Based Targets initiativeTarget validation and sector guidance 科学碳目标
- RE100Corporate renewable electricity commitments 企业绿电承诺
- Greenhouse Gas ProtocolScope 1 / 2 / 3 accounting rules 温室气体核算准则
- CDPCompany-level disclosure and supply-chain data 企业披露与供应链数据
- US EPA — Greenhouse Gas Reporting ProgramFacility-level fluorinated gas data 设施级含氟气体数据
- NIST — CHIPS for AmericaUS public capital and conditions 美国公共资本与附加条件
- European Chips ActEU capacity build and sustainability framing 欧盟产能与永续框架
- Semiconductor Industry Association (SIA)Industry structure, capacity and outlook 产业结构与展望
- SEMIEquipment, materials, fab investment data 设备、材料与晶圆厂投资数据