AI summary by TwinMind
Why 157nm Lithography Failed
Asianometry
17m 45s
Historical Context of Moore's Law and Lithography
- Semiconductor industry's primary economic engine: Moore's Law
- Doubling devices on integrated chip every 18 months
- Three engines drove advances: semiconductor design, increasing wafer sizes, lithography improvements
- Optical lithography techniques = true driving force behind faster chips
- By new millennium: lithographic progress slowing to halt
Evolution of Lithography Wavelengths (1980-1999)
- 1980: G-line mercury arc lamps, 435nm wavelength
- Leading process node: 1.5 micrometer (1,500nm half pitch)
- 1,500nm vs 435nm = like drawing thick lines with sharp pencil
- Second half 1980s: I-line at 365nm
- Leading edge process node: 800nm
- Late 1990s: Switch from lamps to excimer lasers
- 1999: 248nm KrF lasers for 180nm process node
- 193nm ArF lithography scheduled for 2001
The Critical Transition Point (1999)
- Moore's Law acceleration: 1987-1999, process nodes went 800nm → 180nm
- Early 1980s to 1999: minimum feature size shrank ~100x
- 1999: First time leading edge process node < most advanced lithographic wavelength
- Manufacturing became much harder from this point
The Gap Problem and 157nm Solution
- 193nm ArF lasers launching 2001
- Analysts predicted 193nm unsuitable for 100-70nm process nodes (needed ~2003)
- EUV = agreed long-term future
- EUV problems: completely new technology, mirrors instead of lenses, underpowered light source, mirror flatness issues, defect problems
- EUV predicted ready 2010+ (actually much longer)
- 157nm conceived as bridge from 193nm to EUV for 100-70nm nodes
- 157nm vs 193nm: only 36nm difference but 20% reduction
- 157nm = fluorine excimer laser, similar to 248nm/193nm processes
- Timeline: 157nm equipment hit clean rooms 2003 for 100nm process node
Optical Lithography Technology Overview
- Three system parts: exposure tool, mask, resist
- Process: coat wafer in photoresist → send light through mask onto wafer → resist reacts to light → pattern printed → used as stencil for etching
- Three differentiating components:
- Uses photons
- Transmissive photo masks (transparent)
- Projection optics to shrink image (mask pattern 4-5x larger than chip)
Development and Early Progress
- Development began 1997: microstepper prototype at MIT Lincoln Laboratory
- Capable of patterning 80nm wide lines (sufficient for 100nm node)
- All three elements needed extensive re-engineering: exposure tool, mask, resist
Exposure Tool Challenges
- Air absorption problem: 157nm light easily absorbed by ordinary air
- Solution: entire light pathway purged of air, replaced with inert nitrogen
- Added substantial cost to system
- Laser power: initially worried about output power vs 248nm/193nm
- Laser suppliers quickly produced 10-20 watts power (ahead of expectations)
- Major laser problem: line width too wide
- Normal correction with different glass materials impossible
- 157nm light so easily absorbed, only one lens material option: crystalline calcium fluoride
Optical System Material Challenges
- Only viable lens material: crystalline calcium fluoride
- Specialty glass companies (Corning, Schott) worked on high-purity crystal production
- Challenge: minimize stress-induced birefringence
- Birefringence = light splits into two when entering crystal
- Would scatter 157nm light, reduce contrast, cause image aberrations
- 2001 discovery: crystal inherently produced birefringence regardless of production quality
- Intrinsic birefringence: 11.2nm per centimeter of crystal
Birefringence Solutions
- Considered second lens material: barium fluoride crystals (frankdixonite)
- Similar intrinsic birefringence but opposite direction
- Supply chain not prepared for new material, idea dropped
- Final solution: crystal orientation pairs canceling each other's distortions
- Example: 111 orientation crystal paired with two 100 orientation crystals
- Second 100 crystal tilted 45 degrees relative to first
- Required system-wide changes: optical lens coatings, embedded software rewrites
Mask Technology Limitations
- 157nm = last feasible optical lithography wavelength due to mask constraints
- Industry standard: fused silica (pure silicon dioxide) masks
- Fused silica preferred for temperature stability: 1°C change = 0.00005% structure shift
- 50mm wide mask features: 1°C variation causes only 25nm distortion
- 157nm = last wavelength for which fused silica remains transparent
- Normal fused silica absorbs 157nm light
- Corning developed dehydrated fused silica: 85% UV light transmission
- Alternative materials (crystalline calcium fluoride) not viable due to warping/distortion
Resist Challenges
- Conventional resist: modified hydrocarbons for 248nm/193nm
- Hydrocarbons at conventional thickness absorbed too much 157nm light
- Solutions: thinner resist layer (60-90nm) with accuracy consequences, or find alternatives
- New materials very exotic, synthesized from known carcinogens
- Unlikely to formulate/commercialize proper resist alternative in time for 90-100nm nodes
- Final approach: make resist thinner
The Competitive Landscape and Downfall
- Early 2000s: 193nm immersion (193i) emerged as legitimate next-generation candidate
- Immersion lithography: UV rays through immersion fluid to improve printing resolution
- Three competing technologies: 193i, 157nm, EUV
- Sematech organized "shark tank style bake off" between three technologies
- 157nm emerged as weakest link:
- Conservative but not as conservative as 193i
- 193i required less new technology
- No commercially suitable 157nm immersion fluid found (193i used simple ultra-pure water)
- Not ready for 100nm process node rollout
- Crystal calcium fluoride yields: 1% range
- Cost factor: 157nm machines estimated 2x cost of 193nm machines
- Lithography machines = >50% of total fab capital costs
Industry Abandonment
- May 2003: Intel (original champion) dropped 157nm from roadmap
- Cited technical factors, not economic ones
- Industry continued briefly:
- Fourth annual international symposium on 157nm lithography, Yokohama, August 2003
- ASML shipped prototype Microscan 7 to IMEC, September 2003
- Within one year of Intel announcement: other semiconductor manufacturers halted 157nm development
- Suppliers collectively invested hundreds of millions of dollars, now lost:
- Lithography tools, optical materials, photo resists, anti-reflection coatings
- Pellicles, masks, mask blanks, metrology tools, inspection tools
Industry Pivot
- Short term: pivot to 193i
- Long term: focus on EUV
- ASML released first immersion lithography TwinScan: 2006
- Optical lithography's golden years ended
- Lithography industry focused solely on bringing EUV to market
- Foundries would "tread water" until EUV arrived
