Today there are three actively competing approaches to generating extreme ultraviolet (EUV) light for chip lithography. And there's a fourth one — soft X-ray — that's too early to call a competitor, but worth watching.
ASML's LPP dominates — it's what powers every 7nm-and-below fab from TSMC, Samsung, and Intel. But it costs $200M per machine and took $20B+ in R&D.
China's LDP route trades peak performance for accessibility — simpler hardware, lower cost, but stuck on electrode erosion.
Russia's gas cluster approach targets a fundamentally different wavelength (6.7nm vs 13.5nm) and produces no debris because its "target" is a gas.
And then there's soft X-ray / B-EUV — 6.5–6.7nm, 185–190 eV photon energy. Still in the lab. No production light source exists. No mirrors exist for this wavelength. But in 2025, Johns Hopkins published a resist breakthrough in Nature, and a startup called Substrate raised $100M at a $1B valuation with a X-ray lithography story.
Let's break down all four.
The Four Routes at a Glance
| Route | Representative | Wavelength | Source | Target | Power |
|---|---|---|---|---|---|
| LPP | ASML (Netherlands) | 13.5nm | CO₂ laser→tin droplet plasma | Tin (Sn) droplets | 250-600W (production) |
| LDP | Multiple Chinese institutes | 13.5nm | Pulsed discharge→tin vapor plasma | Tin electrodes/vapor | ~100W (lab) |
| Gas Cluster | Russian Academy of Sciences | 6.7nm (target) | Yb femtosecond laser→gas cluster plasma | Li-Xe nanoclusters | 150-200W (theoretical) |
| Soft X-ray / B-EUV | Johns Hopkins / Substrate (lab) | 6.5–6.7nm | Gd LPP (experimental) | Zinc-based resists | No production source |
How Each Works
LPP (Laser Produced Plasma): A CO₂ laser (10.6μm) hits a stream of 30μm tin droplets at 50,000 times per second. Each impact vaporizes the droplet into 500,000°C plasma that emits 13.5nm light. ASML has spent $20B+ perfecting this.
The three fundamental problems:
- Tin debris — fragments coat the collector mirror. ASML uses hydrogen buffers + self-cleaning coatings. It works, but adds enormous complexity.
- CO₂ laser efficiency — CO₂→EUV conversion is ~5-6%.
- Plasma absorption — tin plasma reabsorbs some 13.5nm radiation, limiting usable plasma size.
LDP (Laser Discharge Plasma): A pulsed electrical discharge runs through tin vapor to create plasma. Simpler and cheaper.
The single bottleneck: electrode erosion. Every discharge eats the electrodes. Lifetimes range from thousands to millions of shots. In a 24/7 fab, that's uptime. Research directions include rotating electrodes (like X-ray tubes), liquid tin streams, and cryogenic plasma maintenance.
Gas Cluster (Russian Academy of Sciences, May 2026): Lithium vapor mixed with xenon gas forms Li-Xe nanoclusters. A Yb femtosecond laser (1030nm) excites these clusters to emit 6.7nm EUV light.
The crucial difference: everything is a gas. When the cluster explodes, the fragments are also gas — pumped out by vacuum pumps. No debris, no mirror contamination.
Three bottlenecks:
- Multilayer mirrors for 6.7nm — Mo/Be or Mo/Y coatings need Å-level precision. Thinner, tighter, harder than 13.5nm Mo/Si stacks.
- Gas jet stability — laser-to-cluster coupling depends on cluster size distribution. Complex fluid dynamics.
- Ion damage — 6.7nm photons carry more energy. Different damage mechanisms, zero long-term data.
Timeline Reality Check
| Milestone | ASML LPP | China LDP | Russia Gas Cluster | Soft X-ray / B-EUV |
|---|---|---|---|---|
| Proof of concept | 2000s | 2010s | 2010s-present | ~2025 (JHU resist) |
| Lab prototype | 2012 | ~2023-2025 | ~2026 (announced) | None (needs source+optics) |
| Fab prototype | 2016 | ~2027-2029 | 350nm node is real | No roadmap |
| Mass production | 2018 (hundreds delivered) | ~2030+ | 6.7nm EUV: 10-15 years | 15+ years, if ever |
The Fourth Route: Soft X-ray / Beyond EUV
Soft X-ray (B-EUV) is not competing with the three routes above. It's on a different wavelength track entirely.
The physics argument is compelling. Halve the wavelength, double the resolution — even with moderate NA (0.3-0.5) you could reach <5nm features, comparable to what Hyper-NA EUV (0.75+ NA, estimated $1B per tool) promises at 13.5nm. The catch: everything needed to make it work doesn't exist yet.
Current Progress
Johns Hopkins University (Nature Chemical Engineering, 2025): Solved one of the four critical bottlenecks — resist chemistry. Zinc absorbs 6.5-6.7nm soft X-rays efficiently and emits electrons that trigger chemical reactions in imidazoles. They developed Chemical Liquid Deposition (CLD) to apply zinc-imidazolate frameworks (aZIF) at 1nm/sec onto silicon wafers and successfully created fine patterns.
The team's own assessment: "Years from building even an experimental prototype."
Substrate (California startup, October 2025): $100M funding, $1B valuation. Claims X-ray lithography (XRL) using synchrotron sources can reach sub-2nm nodes. Published 12-13nm CD images of random logic contact arrays. Has disclosed zero details on source type, optics, mask design, or 300mm wafer throughput.
Technology Readiness Comparison
| Dimension | LPP | LDP | Gas Cluster | Soft X-ray |
|---|---|---|---|---|
| Light source | ✅ Production | 🔧 Lab | 🔧 Proof of concept | ❌ None available |
| Optics | ✅ Production | ⚠️ Basic | 🔧 Multilayer needed | ❌ From zero |
| Resist | ✅ Mature | ✅ Reusable | ✅ Reusable | 🔧 JHU initial breakthrough |
| Supply chain | ✅ Complete | ⚠️ Partial gaps | ❌ Nearly zero | ❌ Entirely non-existent |
| First tool cost | ~$200M | ~$10M+ | ~$10-30M (est.) | Unknown (needs accelerator) |
Critical Obstacles
- Light source — No production-capable 6.5-6.7nm source exists. Gadolinium LPP is experimental, far below required power.
- Optics — 6.5nm light is absorbed by almost every material. Entirely new multilayer mirror designs needed. None exist.
- Zero supply chain — Masks, pellicles, metrology, source — everything must be invented from scratch. No company has the incentive to invest ASML-level capital at this stage.
- Substrate's XRL approach — Proximity printing (1:1 mask-to-wafer, no reduction). Synchrotrons are 30m to >1km in circumference, costing $100M+.
The Most Realistic Role
Soft X-ray / B-EUV is not an "EUV replacement." Its most plausible role: a wavelength-based alternative path when Hyper-NA EUV hits its cost ceiling (estimated $1B/tool) — assuming 6.7nm source and optics technology mature by then.
What This Actually Means
1. Four routes, four time windows, not zero-sum.
- 13.5nm (LPP + LDP) → the present. Covers advanced and mature nodes.
- 6.7nm gas cluster → next generation (10-15 years). Advantage in efficiency and wavelength differentiation.
- 6.5-6.7nm soft X-ray → long-term (15+ years). A wavelength hedge against Hyper-NA cost escalation.
2. EUV is a cost variable, not a survival threshold.
Chiplet architectures + DUV multi-patterning cover enormous real-world demand. Advanced packaging reduces dependence on any single lithography node.
3. The gas cluster route's real contribution.
Russia's scheme won't replace ASML. What it does is demonstrate that a third path might exist — breaking the "only LPP can do EUV" assumption.
4. Soft X-ray is the longest bet.
Johns Hopkins cleared one obstacle (resist). Dozens remain. B-EUV's timeline is measured in decades, not years — but if Hyper-NA EUV really hits $1B/tool, the industry will be glad someone started working on alternatives.
Sources: ASML annual reports, published Chinese LDP research papers, Russian Academy of Sciences report (May 2026), Johns Hopkins / Nature Chemical Engineering (2025), Substrate public disclosures, and public industry data.
Top comments (0)