Impact — F-Gas Dossier
F-gases are not an emissions issue. They are an innovation challenge.
The hardest emissions to reduce often matter the most — a field dossier on F-gas inside semiconductor manufacturing, layer by layer.
Impact — F-Gas Dossier
The hardest emissions to reduce often matter the most — a field dossier on F-gas inside semiconductor manufacturing, layer by layer.
The interrogation
Before any data, the honest exchange this dossier is built on.
"We already do good. If not — what will be?"
02 — Definitions
Fluorinated gases — PFCs, HFCs, SF₆ and NF₃ — are man-made compounds prized for one property: extreme chemical stability. That same stability is why they etch silicon with atomic precision, and why, once released, they persist in the atmosphere for centuries to millennia.
GWP100 (100-year global warming potential, relative to CO₂ = 1) per IPCC AR5 (2013), the values most commonly used in industry sustainability reporting. AR6 revises several of these figures slightly; treat as order-of-magnitude, not certificate-grade.
03 — Origin
Not from packaging, not from shipping — from the two steps every fab repeats hundreds of times per wafer: carving pattern into film, then cleaning the tool before the next wafer arrives.
Thin films of oxide, nitride or metal are layered onto the wafer.
Plasma fed by SF₆, CHF₃ or C₄F₈ carves the pattern into the film.
NF₃ or C₂F₆ plasma strips residue from tool walls between wafers.
Unreacted gas and byproducts (often incl. CF₄) leave via the vacuum line.
Where installed: thermal, plasma or catalytic units destroy most of the gas.
Whatever survives abatement — or every gas molecule, where none exists.
04 — Scale
F-gases broadly (mostly refrigerant HFCs, not fab gases) sit around 2–3% of global GHG emissions today. The semiconductor slice is a sliver of that — but each molecule in the slice can outweigh a tonne of CO₂ by four orders of magnitude.
Sources cited in §08. Where a precise, verified global figure for semiconductor-specific F-gas emissions does not exist in public form, this dossier says so rather than inventing precision — see the modeled projection in §07.
05 — Disclosure
Ranked by wafer output / revenue scale. The blank spaces in the "treatment disclosed" column are the finding, not a gap in this dossier.
| Company | Segment | F-gas / PFC disclosure | Treatment % disclosed |
|---|---|---|---|
| TSMC | Foundry | Reports Scope 1 process-gas total & PFC intensity target | Local abatement widely reported, coverage % not itemized per tool |
| Samsung Electronics | IDM / Foundry | Reports F-gas / PFC emissions in sustainability report | Reduction target disclosed, abatement % not broken out |
| Intel | IDM | Long-standing PFC reduction commitments (WSC member) | Historic intensity data published, current coverage % not itemized |
| SK hynix | Memory | Reports greenhouse gas total incl. process gases | Not disclosed at gas level |
| Micron Technology | Memory | Reports Scope 1 & F-gas reduction goals | Not disclosed at gas level |
| Kioxia | Memory | Reports GHG total, references PFC reduction activity | Not disclosed at gas level |
| GlobalFoundries | Foundry | Reports Scope 1 total | Not disclosed at gas level |
| UMC | Foundry | WSC member, reports PFC reduction progress | Intensity trend published |
| Texas Instruments | IDM | Reports Scope 1 total | Not disclosed at gas level |
| STMicroelectronics | IDM | Reports process-gas emissions within Scope 1 | Not disclosed at gas level |
Best-available public disclosure as of each company's most recent sustainability report at time of writing. This table should be re-verified against primary sources before citing further — its patchiness is itself part of the argument in Layer 03.
06 — The reframe
Each layer asks one real question, then answers it three ways: why it exists, what it looks like on the ground, and how it moves.
Why can't the industry simply stop using F-gas?
Plasma etch and chamber clean need chemistries that attack silicon, oxide and nitride with atomic precision while leaving the mask untouched — F-gases remain the only molecules proven at production scale to do this.
SF₆, NF₃, CF₄, C₂F₆, CHF₃ and c-C₄F₈ run through nearly every fab on earth, at every node from mature 28nm to leading-edge 2nm and advanced packaging.
Low-GWP alternatives (fluoroketones, iodofluorocarbons, F-free recipes) are early-stage R&D; nearer-term gains come from swapping high-GWP in-situ clean gas for remote-plasma NF₃ with near-total destruction.
If treatment exists, why isn't all F-gas destroyed before release?
Point-of-use abatement can remove well over 90% of exhaust gas, but every unit adds capital cost, energy draw and maintenance per chamber, per tool, per fab.
Abatement is standard on new leading-edge lines; legacy tools and many mature-node lines — which make most of the world's chips by volume — run partial or no abatement.
Make retrofit a line item on every node roadmap, and let regulation (EU F-gas Regulation, US EPA Subpart I) require coverage disclosure, not just an intensity target.
Why can we name every chip inside a laptop, but not the gas used to make it?
Process-gas emissions are Scope 1, folded into one aggregated CO₂e line, rarely broken out by gas or fab, and rarely audited with the rigor of a financial statement.
Most major chipmakers publish a Scope 1 or PFC total; very few publish gas-by-gas volumes, abatement coverage, or an emissions figure per wafer or per SKU.
A standard "F-gas nutrition label" per fab — gas mix, tonnage, GWP-weighted CO₂e, abatement % — would let a buyer finally ask the question this dossier opens with.
Can efficiency gains outrun AI-driven demand growth?
Generative AI, data-center buildout and device refresh are pushing wafer demand — especially advanced logic and HBM memory — up faster than at almost any point in the industry's history.
A fab that halves emissions per wafer still emits more in absolute terms if wafer output triples. The industry's own 2010-era voluntary targets were normalized, not absolute.
Track absolute F-gas CO₂e, not only intensity — and pair every AI hardware roadmap with an abatement or alt-chemistry roadmap funded at the same urgency as the compute buildout.
If humans introduced F-gas into the process, who owns what happens next?
PFCs and SF₆ entered fabs in the 1980s–90s because they solved an etch/clean problem better than anything before — a deliberate industrial choice, not an accident.
A life-cycle view spans gas sourcing, in-fab process design, point-of-use destruction and end-of-life accounting — split today across suppliers, tool makers, fab operators and regulators with no single owner.
Treat F-gas the way the industry already treats yield or defect density: a named, owned, continuously improved number on every fab's dashboard.
07 — Projection
No verified public total exists for global fab F-gas emissions. Below is a transparent, order-of-magnitude model — same AI-driven wafer growth, two different innovation choices.
Baseline 2024: ~360M 200mm-equivalent wafers/year (≈30M/month, SEMI order-of-magnitude) × assumed 15 kg CO₂e raw F-gas per wafer (mid-point, cross-node average, pre-abatement) × 35% net after ~65% global average abatement coverage ≈ 1.89 Mt CO₂e.
Scenario — stalled: abatement coverage frozen at 65%; wafer volume grows with AI/HBM/logic demand (~8%/yr, reaching ~35% cumulative growth by 2030) → net factor unchanged, absolute emissions rise to ~2.55 Mt CO₂e by 2030.
Scenario — scaling: same wafer growth, but abatement coverage climbs from 65% → 85% by 2030 via retrofit mandates and remote-plasma NF₃ adoption → net factor falls, absolute emissions fall to ~1.09 Mt CO₂e by 2030.
Every input above is a stated assumption, not a measured fact. Treat the spread between the two bars — not either single number — as the finding: the same demand growth can point two different directions depending on innovation choices made now.
08 — Bibliography
Primary sources worth verifying directly for anyone taking this further. No links are fabricated here — search each title at its issuing organization.