EUV Scanner

1. Theory

• Definition: An EUV scanner images a reflective mask onto resist with 13.5 $\mathrm{nm}$ light from a tin plasma, using multilayer mirrors throughout because no material transmits EUV (Fomenkov et al.; halbleiter.org; Farrar et al.).
• Resolution: At NA 0.33 production tools print 13 $\mathrm{nm}$, and the 0.55 NA EXE platform prints 8 $\mathrm{nm}$ (ASML).
• Mirrors: Mo/Si multilayers reflect about 70 per cent each, so a train of six to ten reflections transmits only a few per cent (Farrar et al.; Fomenkov et al.).
• Source: Tin droplets heated by a CO2 laser convert about 5 per cent of laser energy into in-band EUV (Fomenkov et al.; halbleiter.org).
• Stochastics: EUV photons carry 92 $\mathrm{eV}$, so a given dose has about 14 times fewer photons than at 193 $\mathrm{nm}$ and shot noise grows (LaPedus; Brainard et al.).

2. Working Principle

• Source: Droplets of tin are hit by a pre-pulse and a main CO2 pulse at about 50 $\mathrm{kHz}$; a collector mirror focuses the plasma emission to the intermediate focus (halbleiter.org; Fomenkov et al.).
• Illuminator and mask: Mirrors shape the illumination onto a reflective mask, illuminated at an angle (halbleiter.org).
• Projection: Six to eight aspheric mirrors reduce the image 4x onto the wafer in vacuum (halbleiter.org); the high-NA optics are anamorphic with a half field (ASML).
• Dose control: Pulse bursts expose a 26 x 33 $\mathrm{mm}$ field with dose error below 0.5 per cent (Fomenkov et al.).

3. Applications

• Leading-edge logic and DRAM: 7 $\mathrm{nm}$ to 2 $\mathrm{nm}$ logic nodes and DRAM critical layers (ASML).
• Single exposure in place of multiple patterning: Pitches that ArF needs multiple patterning for (ASML; Resolution Enhancement and Multiple Patterning page).
• Inspection demand: Finding rare stochastic defects across billions of contacts (LaPedus; E-Beam Inspection and Defect Review page).

Imaging

Name	Formula	Related Formulae	Variables/Constants
Resolution at 13.5 nm	$HP=k_1\frac{\lambda }{NA}$	$NA=0.33\to 13\mathrm{nm},NA=0.55\to 8\mathrm{nm}$	$\begin{array}{r}ASML:NXEat0.33and13\mathrm{nm},EXEat0.55\\ and8\mathrm{nm},about1.7timessmaller\\ features;RayleighformfromtheHigh-NA\\ DUVImmersionScannerpage\end{array}$ $\begin{array}{r}(computed:k_1=0.318forNXEand0.326for\\ EXE;theNAratio1.67matchesthe13/8=\\ 1.63resolutionratio)\end{array}$
Depth of focus	$DOF=k_2\frac{\lambda }{N{A}^2}$	$\frac{DO{F}_{0.33}}{DO{F}_{0.55}}={\left(\frac{0.55}{0.33}\right)}^2$	$\begin{array}{r}paraxialDOFformfromtheHigh-NADUV\\ ImmersionScannerpageappliedatEUV(my\\ reading)\end{array}$ $\begin{array}{r}(computed:124k_2\mathrm{nm}atNA0.33and44.6k_2\\ \mathrm{nm}at0.55,2.8timesless)\end{array}$
Anamorphic field	$M_x\ne M_y,A_{field,EXE}=\frac{1}{2}{A}_{field,NXE}$	$26\times 33\mathrm{mm}\to 26\times 16.5\mathrm{mm}$	$\begin{array}{r}ASML:anamorphicmirrorsneedingafield\\ halftheNXEsizeandfasterstages;\\ Fomenkovetal.:26x33\mathrm{mm}field\end{array}$ $\begin{array}{r}(computed:thehalvedfieldis26x16.5\\ \mathrm{mm};the4xand8xmagnificationsusually\\ quotedarenotstatedintheopened\\ sources)\end{array}$

Source and Optics

Name	Formula	Related Formulae	Variables/Constants
Tin plasma source	$P_{IF}=CE\cdot P_{laser}\cdot {\eta }_{collect}$	$CE\approx 5\%,P_{IF}\approx 200\mathrm{W}$	$P_{IF}:poweratintermediatefocus$ $\begin{array}{r}CE:conversionefficiency;Fomenkovet\\ al.:CEabout5percent,200Wneeding\\ morethan20kWoflaserpower,30\mu \mathrm{m}\\ dropletsat50\mathrm{kHz};halbleiter.org:\\ pre-pulseandmainpulse,about5percent\\ usable;collectionproduct(myreading)\end{array}$ $\begin{array}{r}(computed:200Wfrom20kWis1percent\\ overall,socollectionandspectral\\ filteringkeepaboutonefifthofthe5\\ percentCE)\end{array}$
Multilayer optics transmission	$T=R^N$	$R\approx 0.67\text{to}0.70$	$R:peakreflectivitypermirror$ $\begin{array}{r}N:numberofreflections;Fomenkovetal.:\\ 10reflectionsat67percent,about2per\\ cent;Farraretal.:about70percent,\\ six-mirrorlensabout12percent,below1\\ percentwithilluminatorandmask;\\ halbleiter.org:sixtoeightmirrorsat4x\\ (seetheQ/A)\end{array}$ $\begin{array}{r}(computed:0.67^{10}=1.8percentand0.70^6=\\ 11.8percent)\end{array}$
Power at the wafer	$P_{wafer}\approx 550\mathrm{mW}(100\text{wph})$	$\frac{P_{wafer}}{P_{IF}}$	$\begin{array}{r}Fomenkovetal.:about550mWatthewafer\\ for100wphat60percentdutycycle,200\\ Watintermediatefocus\end{array}$ $\begin{array}{r}(computed:0.55/200=0.28percent,\\ withinFarra{r}^{\prime }sbelow-1percenttotal)\end{array}$

Photons and Stochastics

Name	Formula	Related Formulae	Variables/Constants
Photon counts	$N_{ph}=\frac{DA}{h\nu }$	$h{\nu }_{EUV}=92.5\mathrm{eV},h{\nu }_{ArF}=6.42\mathrm{eV}$	$D:dose$ $\begin{array}{r}A:area;Brainardetal.:about6.7\times 10^{13}\\ photonsperc{m}^{2}permJ/c{\mathrm{m}}^{2}atEUVand\\ 125\times 10^{13}listedat193\mathrm{nm};LaPedus:14\\ timesfewerEUVphotonsatequaldose(see\\ theQ/A)\end{array}$ $\begin{array}{r}(computed:6.75\times 10^{13}atEUVand9.72\times 10^{14}at\\ ArFperc{m}^{2}permJ/c{\mathrm{m}}^2,aratioof14.4)\end{array}$
Shot noise and roughness	$\frac{{\sigma }_N}{N}=\frac{1}{\sqrt{N}}$	$LER\propto D^{-1/2}$	$\begin{array}{r}Brainardetal.:Poissonstatistics,LER\\ fittingdos{e}^{-1/2}forEUVandDUV,41percent\\ absorptionin125\mathrm{nm},quantumefficiency\\ 2.08;LaPedus:2to4electronsper\\ photon,2\mathrm{nm}secondary-electronblur\end{array}$ $\begin{array}{r}(computed:at50mJ/c{\mathrm{m}}^{2}a10x10\mathrm{nm}pixel\\ receives3350photonsandabsorbsabout\\ 1370,a2.7percentcountfluctuation)\end{array}$
Stochastic failures	$P_{chipfail}=1-{\left(1-p\right)}^{N_{contact}}$	$N_{contact}\approx 10^9$	$\begin{array}{r}p:failureprobabilitypercontact;\\ LaPedus:achipfailsfromonebadcontact\\ amongabillionormore,missingand\\ kissingcontacts,slowe-beamsearch;\\ productofcontactyields(computed)\end{array}$ $\begin{array}{r}(computed:for99percentchipyieldfrom\\ contacts,pmustbebelow1\times 10^{-11})\end{array}$

Important Questions

Ques: Why does EUV print smaller features than ArF immersion, and what does high NA add?

Because the wavelength is 14 times shorter. ASML specifies 13 $\mathrm{nm}$ resolution for its 0.33 NA NXE scanners and 8 $\mathrm{nm}$ for the 0.55 NA EXE platform, features about 1.7 times smaller in a single exposure. Both correspond to $k_1$ near 0.32 (computed), comfortably above the 0.25 limit, whereas ArF immersion already runs near 0.27 (High-NA DUV Immersion Scanner page). The NA ratio of 1.67 matches the resolution ratio of 13/8 = 1.63 (computed). The cost of higher NA is depth of focus, which scales as $1/N{A}^2$ and falls 2.8 times (computed), and a field half the NXE size, because the anamorphic mirrors demagnify differently in the two directions, so ASML added faster wafer and reticle stages. Reference: ASML, ‘EUV lithography systems’ and ‘TWINSCAN EXE:5000’ product pages: NXE:3400C, 3600D and 3800E at NA 0.33 and 13 nm resolution, EXE:5000 and 5200B at NA 0.55 and 8 nm, features about 1.7 times smaller than NXE in a single exposure, anamorphic mirrors needing an exposure field half the NXE size, new faster wafer and reticle stages, dose 50 mJ/cm2 for the throughput figure (fetch-tool summary); High-NA DUV Immersion Scanner page of this vault.

Ques: How much light reaches the wafer through an EUV mirror train?

Only a few per cent, and the sources quote different figures. Fomenkov et al. take 10 reflections at 67 per cent and get about 2 per cent, consistent with $0.67^{10}$ = 1.8 per cent (computed). Farrar et al. quote about 12 per cent for a six-mirror projection lens alone, matching $0.70^6$ = 11.8 per cent (computed), and below 1 per cent once the illuminator and mask are included. Fomenkov’s 550 mW at the wafer from 200 W at intermediate focus is 0.28 per cent (computed). The figures differ because they count different parts of the train; the full path from source to wafer, below 1 per cent, is the one that sets throughput (my reading). Farrar et al. note that 1 $\mathrm{nm}$ of tin on a mirror loses over 10 per cent reflectivity. Reference: I. Fomenkov et al. (Cymer/ASML), ‘Light sources for high-volume manufacturing EUV lithography: technology, performance, and power scaling’, Adv. Opt. Technol. 6, 173 (2017): Sn LPP with CO2 laser, conversion efficiency about 5 per cent (6 per cent in the lab), about 200 W at intermediate focus needing more than 20 kW of laser power, 210 W dose-controlled for 1 h, 30 um droplets at 50 kHz, dose error below 0.5 per cent, 13.5 nm with 2 per cent bandwidth, mirrors reflecting less than 70 per cent, 10 reflections at 67 per cent giving about 2 per cent transmission, about 550 mW at the wafer for 100 wph at 60 per cent duty cycle, 26 x 33 mm field (fetch-tool summary); N. Farrar, D. Brandt and N. Boewering (Cymer), ‘Optics for scanning: multilayer mirrors enable next-generation EUV lithography’, Laser Focus World (26 February 2009): Si/Mo multilayers with about 70 per cent maximum reflectivity at near-normal incidence, collector above 55 per cent, spacing graded about 5 per cent for 5 to 17 degree incidence, barrier layers against interdiffusion, substrate roughness below 0.2 nm, six-mirror projection lens about 12 per cent transmission and below 1 per cent including illuminator and mask, 1 nm tin causing over 10 per cent reflectivity loss (fetch-tool summary).

Ques: How is EUV light generated?

By laser-heating tin droplets into a plasma. Halbleiter.org describes tin droplets about 20 $\mu \mathrm{m}$ across, about 50000 per second, struck by a CO2 pre-pulse and then a main pulse of about 20 kW average power, with only about 5 per cent of the laser energy becoming usable EUV and hydrogen protecting the collector. Fomenkov et al. give droplets of 30 $\mu \mathrm{m}$ at 50 $\mathrm{kHz}$, conversion efficiency about 5 per cent, 6 per cent in the laboratory, and about 200 W at intermediate focus from more than 20 kW of laser, with 210 W demonstrated for an hour and dose error below 0.5 per cent. 200 W from 20 kW is 1 per cent overall (computed), so collection and filtering keep about a fifth of the converted light (my reading). Reference: halbleiter.org, ‘EUV tool technology’: tin droplets about 20 um hit by a CO2 pre-pulse and a main pulse of about 20 kW, about 50000 droplets per second, about 5 per cent of laser energy usable, hydrogen keeping the collector clean, six to eight aspheric mirrors each reflecting about 70 per cent at 4x reduction, surface errors of a few picometres, reflective mask illuminated at an angle (fetch-tool summary); I. Fomenkov et al. (Cymer/ASML), ‘Light sources for high-volume manufacturing EUV lithography: technology, performance, and power scaling’, Adv. Opt. Technol. 6, 173 (2017): Sn LPP with CO2 laser, conversion efficiency about 5 per cent (6 per cent in the lab), about 200 W at intermediate focus needing more than 20 kW of laser power, 210 W dose-controlled for 1 h, 30 um droplets at 50 kHz, dose error below 0.5 per cent, 13.5 nm with 2 per cent bandwidth, mirrors reflecting less than 70 per cent, 10 reflections at 67 per cent giving about 2 per cent transmission, about 550 mW at the wafer for 100 wph at 60 per cent duty cycle, 26 x 33 mm field (fetch-tool summary).

Ques: Why are stochastic effects worse in EUV, and are the photon counts consistent?

Because each photon carries more energy. LaPedus notes 92 against 6.4 $\mathrm{eV}$ photons, 14 times fewer EUV photons at equal dose. Brainard et al. give about $6.7\times 10^{13}$ EUV photons per cm${}^2$ per $mJ/c{\mathrm{m}}^2$, matching $6.75\times 10^{13}$ from 92.5 $\mathrm{eV}$ (computed), but list $125\times 10^{13}$ for 193 $\mathrm{nm}$, whereas 6.42 $\mathrm{eV}$ gives $9.72\times 10^{14}$ (computed), a 14.4 ratio consistent with LaPedus. The computed value, which also matches the 9.7 figure in Brainard’s table, is more likely correct, and the 125 entry a transcription or layout error (my reading). At 50 $mJ/c{\mathrm{m}}^2$ a 10 nm pixel absorbs about 1370 photons, a 2.7 per cent fluctuation (computed), and LER scales as dose${}^{-1/2}$ (Brainard et al.). Reference: M. LaPedus, ‘Searching For EUV Defects’, Semiconductor Engineering (19 April 2018): missing and kissing contacts, a chip failing from one bad contact among a billion, 92 eV against 6.4 eV photons and 14 times fewer photons at equal dose, 2 to 4 electrons per absorbed photon, secondary-electron blur about 2 nm, low-contrast resist defects slow to find, e-beam inspection down to 1 nm but slow (fetch-tool summary); R. L. Brainard et al. (Rohm and Haas, Philips Research), ‘Shot Noise, LER and Quantum Efficiency of EUV Photoresists’, EUV Symposium 2004 slides: Poisson sigma_N = sqrt(N), LER proportional to dose^(-1/2), about 6.7e13 photons per cm2 per mJ/cm2 for EUV at 92.5 eV against 125e13 listed for 193 nm, EUV absorption 41 per cent in 125 nm film, quantum efficiency 2.08 acids per absorbed photon, 45 nm targets LER 2 nm and 2 to 15 mJ/cm2 (fetch-tool summary).

Ques: What stochastic defects matter for yield, and how are they found?

Rare missing and merged contacts. LaPedus reports that stochastic effects cause pattern roughness and missing or kissing contacts, that one bad contact among a billion or more can kill a chip, and that each absorbed photon releases 2 to 4 electrons with about 2 $\mathrm{nm}$ blur. If a chip has $10^9$ contacts, 99 per cent yield needs a per-contact failure probability below $10^{-11}$ (computed), so defects must be counted at very low rates. Such defects in resist have low contrast and can take days to find; e-beam inspection reaches 1 $\mathrm{nm}$ sensitivity but is slow (LaPedus; E-Beam Inspection and Defect Review page). Shot-noise roughness enters the same PSD framework as other resist roughness (Photoresist and Process Window page). Reference: M. LaPedus, ‘Searching For EUV Defects’, Semiconductor Engineering (19 April 2018): missing and kissing contacts, a chip failing from one bad contact among a billion, 92 eV against 6.4 eV photons and 14 times fewer photons at equal dose, 2 to 4 electrons per absorbed photon, secondary-electron blur about 2 nm, low-contrast resist defects slow to find, e-beam inspection down to 1 nm but slow (fetch-tool summary); E-Beam Inspection and Defect Review page of this vault; Photoresist and Process Window page of this vault.

References

1. ASML, “EUV lithography systems” and “TWINSCAN EXE:5000.” Available: https://www.asml.com/en/products/euv-lithography-systems and https://www.asml.com/products/euv-lithography-systems/twinscan-exe-5000
2. I. Fomenkov et al., “Light sources for high-volume manufacturing EUV lithography: technology, performance, and power scaling,” Advanced Optical Technologies, vol. 6, no. 3-4, pp. 173-186, 2017. Available: https://www.frontiersin.org/articles/10.1515/aot-2017-0029
3. halbleiter.org, “EUV tool technology,” Semiconductor Technology from A to Z. Available: https://www.halbleiter.org/en/photolithography/euv-tool-technology/
4. N. Farrar, D. Brandt and N. Boewering, “Optics for scanning: multilayer mirrors enable next-generation EUV lithography,” Laser Focus World, 26 February 2009. Available: https://www.laserfocusworld.com/optics/article/16566714/optics-for-scanning-multilayer-mirrors-enable-nextgeneration-euv-lithography
5. R. L. Brainard, P. Trefonas, J. H. Lammers, C. A. Cutler, J. F. Mackevich, A. Trefonas and S. A. Robertson, “Shot noise, LER and quantum efficiency of EUV photoresists,” EUV Symposium 2004. Available: https://euvlsymposium.lbl.gov/pdf/2004/presentations/day2/Re04_Brainard.pdf
6. M. LaPedus, “Searching for EUV defects,” Semiconductor Engineering, 19 April 2018. Available: https://semiengineering.com/searching-for-euv-defects
7. High-NA DUV Immersion Scanner page of this vault.
8. Photoresist and Process Window page of this vault.
9. E-Beam Inspection and Defect Review page of this vault.

Related pages in this vault: High-NA DUV Immersion Scanner, Photoresist and Process Window, Resolution Enhancement and Multiple Patterning, E-Beam Inspection and Defect Review, Defect Detection Fundamentals, Nanoimprint Lithography Tool.