AI summary by TwinMind
What ASML Has Next After EUV
Asianometry
15m 19s
Current EUV Lithography Status and Challenges
- Current EUV machines cost $150 million each, using dozens to create nanometer-wide patterns
- Industry worked decades on this technology, but ASML already developing next generation
- High-NA EUV machines will cost $300+ million each, improving upon regular EUV systems
- Current commercial challenges center on throughput limitations
EUV Throughput and Economic Issues
- TSMC's N5 process uses 14-15 EUV layers
- EUV machines slower than older machines, requiring more units for same wafer output
- 2020: TSMC bought maximum EUV machines from ASML, capturing 50% of install base
- N5 wafer costs ~$17,000 (price of a car), representing 80% theoretical price jump
- Cost reduction requires higher throughput to achieve economies of scale
N3 Process Node Challenges
- TSMC's N3 node uses 20-25 EUV layers (vs N5's 14-15)
- Risk of further throughput decline
- Multiple patterning technique required for N3
- Uses additional exposures to improve feature density
- Doubles or triples processing time per layer
- Makes already slow EUV "decidedly uneconomical"
- ASML already selling EUV machines at maximum production capacity
High-NA EUV Technical Fundamentals
- NA = Numerical Aperture (dimensionless number measuring light collection/focus capability)
- Rayleigh formula: Resolution = K1 × wavelength / NA
- Resolution improvement options:
- Shrink K1 (at physical limit)
- Reduce wavelength (EUV already at 13.5nm, don't want to push further)
- Increase NA (chosen approach)
Current vs High-NA EUV Specifications
- Current ASML NXE-3600D: NA 0.33, 26nm pitch, 160 wafers/hour
- Meets TSMC N5 requirements (28nm pitch needed)
- High-NA EXE-5000: NA 0.55, 16nm pitch
- 67% improvement in resolution
- Preferred method for N3 process implementation
- Higher resolution = fewer defects = more good wafers per day
- Can replace multi-patterning layers with single exposures
High-NA EUV Engineering Challenges - Optics
- Starlight 5000 optics system redesign required
- Must allow light at more angles to hit wafer
- Old 0.33 NA had thin light cone at Mirror 6 (M6)
- 0.55 NA requires thicker light cone
- Challenge: M6 thicker cone interferes with Mirror 5 (M5)
- Multi-layer mirrors limited to ~13° reflection angles
- M5 positioned very close to M6 light cone
- Interference causes imaging errors
Mirror Engineering Solutions
- Super mirrors: 1 meter wide, picometer accuracy
- Solution: Drill oval-shaped holes ("obscurations") in mirrors
- Reduces angular spread, significantly increases transmission
- Zeiss developed this technology (mechanism not publicly detailed)
Mask/Reticle Design Constraints
- Mask redesign costs millions of dollars
- Design requirement: mask size cannot change
- High NA at wafer requires high NA at mask (mathematically linked)
- Current NXE-3300B: 0.33 NA, 4× magnification, 6° reflection
- Projects 132×104mm reticle area onto 33×26mm wafer area
High-NA Mask Challenges and Solutions
- Higher NA creates bigger, overlapping light cones (1° overlap)
- Cannot increase reflection angle (6° to 9°) due to mask shadowing effects
- Solution: Increase magnification from 4× to 8×
- Problem: 8× magnification shrinks wafer area to 16.5×13mm (quarter size)
- Economically unacceptable - would make machine 4× slower
Anamorphic Imaging Solution
- Cannot make masks larger (12-inch masks unfeasible)
- ASML/Zeiss developed anamorphic imaging (announced 2014)
- Borrowed from movie industry (widescreen projection technology)
- Different magnification factors for X and Y axes
- 8× magnification only needed on Y-axis
- Final projection: 16.5×26mm field (half size, barely acceptable)
High-NA EUV Engineering Challenges - Resist
- Photoresist applied before EUV exposure
- Higher NA requires thinner resist to prevent pattern collapse
- Risk: Lines too close together after exposure
- During washing/drying, uneven drying causes water surface tension distortion
- Lines may fall over or peel off substrate
- Target: Resist layer 50% thicker than pitch
- 16nm pitch requires 20nm thick resist
- ASML partnering with IMEC, PSI, CXRO for resist development
- Still working on viable candidates
Additional Engineering Challenges
- Bigger mirrors required
- Mask frame layout modifications
- Pellicle system updates
- Metrology system challenges
- Inspecting sub-20nm features
- Investigating e-beam inspection techniques
- Not yet finalized
Development Timeline and Specifications
- First EXE-5000 machines: Late 2022 delivery to customers
- Improved EXE-5200 version: 2024-2025
- EXE-5200 throughput: 220 wafers/hour
- Machine size: "Big as crouched T-Rex"
- Price: $300+ million each
- Can buy second most expensive painting in world
- Fund 8.8 Squid Games productions
- Pay gas fees for one NFT
Economic Justification Requirements
- Must justify $300+ million price tag
- Technology exists but needs to become shipping-competitive
- Challenge: Making it competitive against next best alternative
