EUV Presentation

Extreme Ultra-Violet Extreme Ultra-Violet LithographyLithography

• Why do we need EUV lithography?

• Brief overview of current technology

• What exactly is EUV?

• System diagram

• Challenges associated with EUV

• 13.5nm source

• Optics

• Masks

• Resists

Outline

Mask Maker’s Holiday:

“large” k1

Mask Maker’s Burden: “small” k1

Why EUV?

Minimum lithographic feature size =

k1: “Process complexity factor” – includes “tricks” like phase-shift masks

λ: Exposure wavelength

NA: Numerical aperture of the lens – maximum of 1 in air, a little higher in immersion lithography (Higher NA means smaller depth of focus, though)

k1*λ

NA

ftp://download.intel.com/research/silicon/EUV_Press_Foils_080204.pdf

There are only so many “tricks” to increase this gap, and they are very expensive … we MUST go to a shorter wavelength!

Why EUV? Why not the next excimer line?

• Hg (G line) @ 436nm Hg (H line) @ 405nm Hg (I line) @ 365nm

KrF Excimer @ 248nm ArF Excimer @ 193nm ???

• 157nm lithography based on the fluorine excimer laser has been largely shelved, which leaves 193nm with extensions for production

• Below that, no laser line has the required output power

• Excimer-based steppers expose 109 steps per 300mm wafer, and produce >100 wafers per hour – exposure times ~ 10-20ns

• Additionally, fused silica and atmospheric oxygen become absorptive by 157nm – so even incremental decreases in wavelength start to require a major system overhaul

Mask Maker’s Holiday:

“large” k1

Mask Maker’s Burden: “small” k1

ftp://download.intel.com/research/silicon/EUV_Press_Foils_080204.pdf

Why EUV? It’s all about the money.By decreasing λ by a factor of 14, we take pressure off k1 – this makes the masks less complicated and expensive because we can skip the “tricks”

For example: a 90nm node mask set:• Pixels:

• Number of pixels on 1 mask: 1012

• Defects:• Size that must be found and repaired: 100nm (25nm as projected on wafer)• Number of such defects allowed: 0

• Data:• Total file size needed for all 22-25 layers: 200GB

• Cost:• Cost for mask set (depreciation, labor, etc): ~$800k-1.3M

ftp://download.intel.com/technology/silicon/Chuck Gwyn Photomask Japan 0503.pdf

Current Lithographic Technology

193 nm Excimer Laser Source

Computer Console

Exposure Column(Lens)

Wafer

Reticle (Mask)

www.tnlc.ncsu.edu/information/ceremony/lithography.ppt

• Lenses are very effective and perfectly transparent for 193nm and above, so many are used

• A single “lens” may be up to 60 fused silica surfaces

• System maintained at atmospheric pressure

• Lens NA ~0.5-0.85

• Up to 1.1 for immersion

• Exposure field 26x32mm

• Steppers capable of >100300mm wafers per hourat >100 exposures per wafer

Basic Technology for EUV

All solids, liquids, and gasses absorb 13.5nm – so system is under vacuum

Mask must be reflective and exceptionally defect-free

13.5nm photons generated by plasma source

All-reflective optics

(all lens materials are opaque)

ftp://download.intel.com/technology/silicon/EUV_Press_Foils_080204.pdf (both images)

13.5nm Plasma Radiation Source

• The only viable source for 13.5nm photons is a plasma

• Powerful plasma required – temperature of up to 200,000oC, atoms ionized up to +20 state

• Plasma must be pulsed – pulse length in pico- to nanosecond range

Argon

Tin

http://www.sematech.org/resources/litho/meetings/euvl/20021014/16-Spectro.pdf

• Pre-ionized plasma excited by powerful IR laser or electric arc of up to 60,000 A to cause emission

Plasma Compositions for 13.5nm

Argon Tin

Argon

• 13.5nm photons only generated by one ion stage (Xe11+)

• Even this stage emits 10 times more at 10.8nm than 13.5

• Maximum population of this stage is 45%

• On the plus side, Argon is very clean and easy to work with

Argon is horribly inefficient: to produce 100W at 13.5nm, kilowatts of other wavelengths would have to be removed

Tin

• Optimum emission when tin is a low-percentage impurity

• All ion stages from Sn8+ to Sn13+ can contribute

• Tin tends to condense on optics

Tin is great as a 13.5nm source, if we can engineer a way to use it without destroying our optics

http://www.sematech.org/resources/litho/meetings/euvl/20021014/16-Spectro.pdf

Where Plasma and Optics Meet

- Ions in the source plasma have enough energy to sputter material off the lenses of the collector optics

- If the source uses tin, that will deposit on the lenses as well

At the power levels required for real exposures, collector optics have a lifetime of about a month

This is VERY bad for Cost of Ownership (CoO)

ftp://download.intel.com/technology/silicon/EUV_Press_Foils_080204.pdf

All-Reflective Optics

All solids, liquids, and gasses absorb 13.5nm photons

- So fused silica lenses are OUT …

- Indeed, all refracting lenses are OUT

http://www.zeiss.com/C1256A770030BCE0/WebViewAllE/D6279194C2955B2EC12570CF0044E537

Making EUV mirrors is no cakewalk, either …

• 50 or more alternating Mo/Si layers give the mirror its reflectivity

• Each layer is 6.7nm thick and requires atomic precision

• Since the angle of incidence changes across the mirror, so do the required Mo/Si layer thicknesses

• Acceptable surface roughness: 0.2nm RMS

• Aspheric

• Net reflectance: ~70%

Optics System - Exposure Field

Full field: ~109 exposures per 300mm wafer

Development-size field: > 500,000 exposures per 300mm wafer

-In July 2005, Carl Zeiss shipped the first 0.25NA full-field optics system to ASML for integration in an EUV systemPress release: http://www.zeiss.com/C1256A770030BCE0/WebViewAllE/D6279194C2955B2EC12570CF0044E537

ftp://download.intel.com/research/library/IR-TR-2003-39-ChuckGwynPhotomaskJapan0503.pdf

EUV Masks

ftp://download.intel.com/research/library/IR-TR-2003-39-ChuckGwynPhotomaskJapan0503.pdf

EUV Masks

NO defects are ever allowed in a completed mask• Extremely flat and defect-free substrate, perfected by smoothing layer

• All defects in multilayer reflecting stack must be completely repaired

• No defects allowed in absorber layer

• All defects in final absorber pattern must be completely repaired

(No wonder mask sets are so expensive!)

ftp://download.intel.com/research/library/IR-TR-2003-39-ChuckGwynPhotomaskJapan0503.pdf

EUV ResistsBest Positive Resist

2.3mJ/cm2 LER=7.2nm

Best Negative Resist

3.2mJ/cm2 LER=7.6nm

39nm 3:1 (space:line)

LER – Line Edge Roughness

ftp://download.intel.com/research/library/IR-TR-2003-39-ChuckGwynPhotomaskJapan0503.pdf

Conclusion

Will 193nm ever die?

• As recently as 2003, EUV was “the only viable solution” for the 45nm node

• Now Intel wants EUV for the 32nm node, but it may be pushed back more:

“In a nutshell, many believe that EUV will NOT be ready for the 32-nm node in 2009. Some say the technology will get pushed out at the 22- nm node in 2011. Some even speculate that EUV will never work.”

- EE Times, Jan 19, 2006

My opinion: never say “never” about this industry…

• A lot of work remains: increase output power of 13.5nm source, increase NA of reflective lenses, increase lifetime of collector optics (decrease cost of ownership)

• But the potential payoff is sufficient that we will make it work