Aerial imaging study of the mask-induced line-width...
Transcript of Aerial imaging study of the mask-induced line-width...
Aerial imaging study of the mask-induced line-width roughness of EUV lithography masks
Antoine Wojdyla1,*, Alexander Donoghue1, Markus P. Benk1, Patrick P. Naulleau1
and Kenneth A. Goldberg1
1Lawrence Berkeley National Laboratory, 1 Cyclotron Road, Berkeley, CA 94720 (USA)
ABSTRACT
EUV lithography uses reflective photomasks to print features on a wafer through the formation of an aerial image. The aerial image is influenced by the mask’s substrate and pattern roughness and by photon shot noise, which collectively affect the line-width on wafer prints, with an impact on local critical dimension uniformity (LCDU). We have used SHARP, an actinic mask-imaging microscope, to study line-width roughness (LWR) in aerial images at sub-nanometer resolution. We studied the impact of photon density and the illumination partial coherence on recorded images, and found that at low coherence settings, the line-width roughness is dominated by photon noise, while at high coherence setting, the effect of speckle becomes more prominent, dominating photon noise for exposure levels of 4 photons/nm2 at threshold on the mask size. Keywords: EUV lithography, LER, LWR, actinic inspection, LCDU, X-Ray Microscopy, photon noise, partial coherence, power-spectral density, speckle, metrology, real-space imaging
1. INTRODUCTION
As extreme ultraviolet (EUV) lithography pushes lithographic resolution closer to the atomic scale, the specifications on critical dimension uniformity (CDU) are becoming more stringent. That is especially true for local CD uniformity (LCDU), which is a direct manifestation of the resist line-edge roughness (LER).
Though often considered within the context of the well-known resolution, LER and sensitivity (RLS) trade-off for photoresists, the final LER in printed features (Figure 1, right) arises from chemical processes in the resist, and from imperfections of the mask, transferred to the aerial image. These include pattern roughness (Figure 1, left) and the influence of speckle (i.e. phase roughness). Phase effects come from mask and multilayer roughness, and can only be seen in the actinic imaging. (Figure 1, center).
Extreme Ultraviolet (EUV) Lithography VII, edited by Eric M. Panning, Proc. of SPIE Vol. 9776,97760H · © 2015 SPIE · CCC code: 0277-786X/15/$18 · doi: 10.1117/12.2219513
Proc. of SPIE Vol. 9776 97760H-1
Downloaded From: http://proceedings.spiedigitallibrary.org/ on 04/26/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx
Figure 1. Pattmask blank rousimilar yet diff Because of thtypically on treflective mu2.) In most ca
Figure 2. Aercoherent illumcaused by the Early studiessimulations [between LERprinting [5,6]
tern transfer inughness are amo
fferent critical dim
he reflective nathe order of 2-Åultilayer coatinases, decreasin
ial images of a tmination, the spec
off-axis imaging
s of aerial imag[2], predicting R on independe]: they found th
EUV lithograpong the parametemensions; they’r
ature of EUV pÅ-RMS [1]. Dg, causing pha
ng the illuminat
typical EUV phckle caused by mg conditions.
ge LER by Nauthe effects of rent wafer printhat speckle cou
phy, from maskers that affect thre for illustration
photomasks, aeepending on th
ase errors that ation coherence
hoto-mask blankmask roughness c
ulleau and collereplicated surfas [4]. Later autuld be replicate
k to wafer. Impee final line-edgen purpose only.)
erial images arehe deposition teappear as specke decreases the
k, under coherencan clearly be di
eagues have foace roughness othors made inded, but only for
erfections of the pe roughness on a)
e highly sensitiechnique, subskle under coherspeckle contra
nt (left) and incoistinguished. Bo
ocused on settinon aerial image
direct measuremr mask-roughn
pattern on the pha printed wafer (t
ive to substratestrate roughnesrent illuminatioast.
oherent illuminatth images have a
ng relevant tolees [3], and findments of the maness well above
hoto-mask and inthese samples ha
e roughness, ws is replicated on conditions (
tion (right.) Witha vertical focal g
erances througding correlationask-induced LEe current
ntrinsic ave
which is into the (figure
h gradient
h ns ER by
Proc. of SPIE Vol. 9776 97760H-2
Downloaded From: http://proceedings.spiedigitallibrary.org/ on 04/26/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx
exposun5 6 8
G = 0.3
5 25 40 5 6 8
II
J
10 12 15
I
G = 0.6
25 40
511 11111
0 12 15 2
II 'I II
lLf =0.8
manufacturin(SHARP) on
Mascomplex inteLWR. Actiniline-edge roumask manufa
We used the images of a pEUV lithogra13.5-nm-wavallows full coIn recent defe[9]. SHARP’ Theto form line-adimensions a
Imafrom the studnoise in the m(The partial c
Figure 3. SamPhoton noise dthe exposure ti The images wfocal gradien1% along theeach exposur
ng capabilities.n a mask blank sk substrate roueractions, it is iic aerial image ughness, isolatiacturers to imp
SHARP EUV patterned EUVaphy scanners.velength light (oherence contrfect-imaging stu’s real-space ime photomask usand-space test are expressed inages of the pattedy. We varied (measurement, acoherence facto
mple 160 nm (4xdecreases with exime for compari
were cropped tnt (arising frome length of the re time and illu
More recentlywith a roughneughness is hardmportant to demeasurements
ing the contribuprove over time
mask-imagingV photomask. S
It is a synchro(E/∆E = 1450) rol, providing cudies, SHARP
maging enablessed for this studpatterns, inclun mask units.) erned field wer(1) the exposurand (2) the partor is defined as
xCD) bright linexposure time, wson.)
o a 5x5µm2 arem the off-axis c
line, covering umination cond
y, the aerial imaess gradient [7d to improve anevelop a metrols can help us toutions of subste.
2. DATA
g microscope atHARP emulate
otron-based miusing diffracti
conventional an has been show
s it to perform ady is a standard
uding a field of
re acquired usire time from 5 tial coherences σ = NAillum/N
e from the collehile a decrease i
ea, where the ionfiguration) aabout 14 conti
dition, measure
age LER rough]. nd difficult to qlogy to performo study the drivtrate roughness
A COLLECT
t Lawrence Bees the imagingcroscope that rive Fresnel zonnd engineered wn to faithfullyaerial image md EUV reflectif 200-nm-CD li
ing a 0.33-4xNseconds to 40 of the illumina
NAlens, with σ =
ected aerial imain coherence blu
imaging perforand sample illuguous lines (in
ements were re
hness has been
quantify in pram the experimeving factors bes and photon sh
TION
erkeley Nationag properties of crecords real-spne plate lenses.illuminations.
y predict wafer measurements oive Mo/Si mulines with 1:1 li
NA lens at best seconds to pro
ation, σ, from 0= 0 being the fu
ages, with variouurs out speckle (i
rmance is diffraumination unifondependent andpeated ten time
n studied with a
actice. Since phental evaluationehind the formahot noise, and p
al Laboratory [current and futace images wit Its Fourier syn
printing at curof line-width, altilayer blank pine-to-space ra
focus. Figure 3obe the influen0.3 to 0.8 to vaully coherent il
us exposure timeintensity levels a
action limited, ormity cause Cd identically dies to build a st
an actinic micr
hase roughnessns of the aerialation of aerial iproviding feed
[8] to record aeture generationth monochromnthesis illumin
rrent EUV genand LER or LWpatterned with aatio (in this pap
3 shows imagece of photon sh
ary the image sllumination con
es and partial cohare linearly adjus
while the vertiCD variations leistributed (i.i.dtatistical ensem
roscope
s causes l image image
dback for
erial ns of
matic, nator
nerations. WR. absorber
per, all
e details hot slope. ndition.)
herence. sted with
ical ess than
d.)) For mble.
Proc. of SPIE Vol. 9776 97760H-3
Downloaded From: http://proceedings.spiedigitallibrary.org/ on 04/26/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx
threshold
X-100
nominal CD
0
position [nm]
100
I
60 nm1-1
Aerial imageis different freffective pixea consistent awith a nominvarious partialine shape anlinear interpo
Figure 4. Dete40 sec exposur15x15 nm2. Ciimage and the Assuming tha{n, m} (with rising edge (f
and similarlyinterpolation which is anotLeft and righfor 160-nm liintensity for σ = 0.8. Finally, we enoise, and an
e measurementsrom the way inel width corresanalysis of the nal 160-nm lineal coherence se
nd intensity varolation of the im
ermination of thre time, a peak pircled dots are daretrieved left an
at the line is ve(x, y) = (n*dx
figure 4), the le
y for the right e(on pixel posi
ther approach tht edges are detine width correσ= 0.8; the n
extracted (1) thnalyzed (2) fou
s are intrinsican which edges asponds to 15-nmdata throughou
e-width in the bettings. The thrries. To assess mages, and det
he line edges thphoton density oata points, and cnd right line edge
ertical (y-directx, m*dy)), It is teft edge positio
edge (substitutition) avoids thto the row-by-rtermined indivesponds to appnormalized ima
e same line frourteen adjacent
ally continuousare extracted frm in mask unitut this study, wbright region. Treshold level isnm-scale edgetermine the thr
hrough thresholf 40 photons/nmrosses are interpes for many diffe
tion), that I(x,ythe threshold aon for row m is
,ing nt+ for the f
he problem of drow width calcidually, and th
proximately oneage log-slope (N
om ten indepenlines from the
, and edge extrrom mask or wts, well below
we set an intensThis width wass adjusted for ee-position inforreshold-crossin
ld and interpolam2, and a partial cpolated edge posferent positions a
y) = I[n, m] is tand nt+ the pixes approximatel, 1falling edge.) Tdiscretization oculation. he local width ie quarter of theNILS) values a
ndent images resame image, t
raction occurs wafer-print SEM
the resolution sity threshold ls chosen to alloeach dataset to rmation, it is neng positions (fig
ation. (left) One coherence of 0.3sitions at the threalong the line.
the intensity ofel index at the ly:
1, ∗ This method ofonto a finite gri
s calculated froe peak intensityare 5.0 for σ=
ecorded in sequto build the line
by interpolatinM micrographsof the imaginglevel to yield anow for a compamaintain the n
ecessary to pergure 4.)
cross-section of3; the effective peshold. (right) O
f the pixel at (inleft-side of the
1
f continuous edid before the th
om their differy for σ= 0.3 a 0.3, 6.6 for σ
uence to assesse statistics (figu
ng the gray leves. In SHARP, thg element. To pn average line-arison between
nominal width rform a row-by
f an image acquipixel size is (dx*dverlay of the aer
nteger) coordinthreshold on t
dge-position hreshold is app
ence. The threand half of the p
= 0.6 and 3.0
s the effects of ure 5.)
els. This he
perform -width n the as the
y-row
ired with dy) = rial
nate the
lied,
shold peak for
f
shot
3. LINNE-EDGE REETRIEVAL PROCEDURRE
Proc. of SPIE Vol. 9776 97760H-4
Downloaded From: http://proceedings.spiedigitallibrary.org/ on 04/26/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx
E-c-16E.
:5 16C
o
ixp : 40s, a=(
1
position ram
D.3
211
E£165t 160
,T
\
exp
1
PC
: 40s, a=0.3
2)sition 1-,u,m1
3
Figure 5. Linemeasurement (offset for clari
4. INFLUE
The noise in LWR, by addAssuming the
Line-width rotimes, measufeatures of th
Figure 6. Influof the same dethe lines, whileunderlying (rep The noise conpixel [10]. Thapproximatel
e-width retrieva(left) and variouty.
ENCE OF TH
the acquired imding a contributey’re statistica
oughness is deurements yield he intrinsic rou
uence of photonense line pattern e the solid line sproducible) line
ntribution to thhe measured Lly 64 photons/p
al from aerial imus lines extracted
HE PHOTON
mages impacts tion LWRphoton
ally independen
LW
fined as three tnoisy data (figghness, compa
n shot noise on with (left) 5 sec
shows the row-av-width contour b
he measured LWLWR decreasespixel/s at thresh
mages using thrd from the same
N NOISE ON
the line-edge rto the intrinsic
nt, the measure
WR = times (“3σ”) th
gure 6, left), buarable to measu
the line-width ac and (right) 40 sverage. The stanbecomes apparen
WR should scas as expected whold for σ = 0.
reshold and inteimage (right). Th
N MEASURE
retrieval [10] tc roughness LWed LWR can be
he measured linut averaging muurements made
and LWR meassec exposure timndard deviation ant for longer exp
ale as the of thewith exposure ti
3) toward an a
erpolation. Samhe lines have the
ED LINE-WI
thus the overallWRmask caused b
expressed as :
2
ne-width standultiple exposure with longer ex
surement. Supermes, and σ = 0.3. at each position iposure time (soli
e inverse squarime (in this exp
asymptote (figu
me line extractede same average n
IDTH ROUG
l measured lineby the mask str
dard deviation. res allows us toxposure times
rposition of ten sDots indicate this indicated by thid line).
re root of the nperiment, the dure 7). At the h
d from various renominal width b
GHNESS
e-width roughnructure itself.
Shorter exposuo capture the fi(figure 6, right
sequential measuhe measured widhe filled area be
number of photdetector sees highest exposur
epeated but are
ness
ure ine t.)
urements ths along low. The
tons per
re time,
Proc. of SPIE Vol. 9776 97760H-5
Downloaded From: http://proceedings.spiedigitallibrary.org/ on 04/26/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx
10
8
bM
6Ec
photon noiseWR (estimate)
B 10 1:
IId [ph/nm2]
-u = 0.3Q = 0.6Q = 0.8
10
8
6
ii
SÌI
2 4
photon count
CL
,-
mulations
6 8
at threshold [
.= 0.6
to =,0.8
2losition [pm]
10 12
:ph /nm2]
1
0 2
photon cot
o-150 -100
-.-LWR with- - -- intrinsic 1_1
4 6
unt at thresho
the shot noiseaerial image LWR to be 5
Figure 7. Influusing correspoincreased, the In the conditiimage LWR note that 1 phacquisition isphoton noise
Generally spespatial cohereincoherent illdecrease in sheventually trameasurement
e contribution LWR by fittin
5.8 nm for these
uence of photononding physical pinherent line-wi
ions of the expwhen the expohoton/nm2 at λ s limited by the will affect prin
eaking, cohereence caused bylumination tenharpness comeading intrinsic t or printing in
becomes smallg the data and e lines.
n shot noise on parameters (righidth roughness c
periment, the inosure is 15 seco
= 13.5 nm core physical procnting in a stoch
5. EFFence increases ty small phase vds to blur the s
es with a higheedge roughnes resist.
l compared to textracting the
the magnitude ht). Here, σ = 0.3aused by the pat
ntrinsic LWR aonds (photon drresponds to 1.cess of photon hastic fashion,
FECT OF ILLthe NILS, and variations fromspeckle, resultier sensitivity toss for stochasti
the intrinsic linasymptote. Fo
of the LWR, fro
3, and the nominttern and the mas
and the photon density in the or47 mJ/cm2 on counting (all eincreasing the
LUMINATIO
intensifies the m the reflective ing in apparento photon noise ic roughness (f
ne-width roughor example, wit
om measuremennal line width is 1sk blank roughn
noise have a crder of 4 photothe mask side.
external source e aerial image L
ON AND NIpresence of spmultilayer coa
tly smoother linfor comparable
figure 8.) that o
hness. We can th σ = 0.3, we e
nt (left) and as pr160 nm. As the e
ness becomes dom
comparable impons/nm2 at the ) Since the noiof noise being
LWR at low ex
LS peckle. This is ating or underlynes, in many cae level of incomonly affects aer
estimate the inestimate the int
redicted by simuexposure time isminant.
pact on the aernominal threshise in the imageg negligible), thxposure dose.
a manifestationying substrate.ases. Howeverming photon flrial image
ntrinsic trinsic
ulations s
rial hold; e he
n of the That is,
r, this lux,
Figure 8. Effect of partial coherence on line-width roughness. (left) overlaid cross-sections for a line from a single measurement acquired with a .33 4xNA lens, 40 s exposure time and partial coherence σ = 0.3, σ = 0.6 and σ = 0.8; (right) overlaid line-width measurements for many measurements of the same line, for partial coherence σ = 0.3, σ = 0.6 and σ = 0.8. Lower coherence decreased speckle but increased sensitivity to photon noise.
Proc. of SPIE Vol. 9776 97760H-6
Downloaded From: http://proceedings.spiedigitallibrary.org/ on 04/26/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx
/0-150 -100 -50 0 50 100 150
position [nm]
180
1 bua
150
;,.,..;. ä.
14001 2position [µm]
3
6=0.3, N=14 i.i.d
-r=5sT=15sT=40s
Gh
5 10 15 20frequency [cycle /µm]
25
'r=40sec, N=14 i.i.d
-o-=0.8-o-=0.6
o-=0.3
5 10 15 20frequency [cycle /µm]
25
There is however a slight advantage to incoherent illumination in that the roughness no longer has definite coherence length. In that respect, engineered partial coherence can be used [11] to reduce the sensitivity to some kind of defects such as bridging, that would be caused by mask roughness (attributable to a bright speck, akin to a slight phase defect.) Another way to influence the line width is to change the intensity threshold. Since the interpolation is somewhat aggressive, with measured line-width variations smaller than the pixel size, changing the threshold by a small amount does not necessarily involve different image pixels, resulting in a linear scaling. However, the image is oversampled by a factor of 2.7 relative to the Nyquist limit, and the effective pixel size in our experiment is small enough to allow using pair-wise distinct interpolation points (figure 9, left.) Our experiment shows that there is no appreciable change in the LWR or sensitivity to photon noise while changing the threshold within ten percent: CD and CDU do not affect LWR or LCDU measurement (figure 9, right.)
Figure 9. Aerial image line-width for various thresholds, with σ = 0.3 and 40 s exposure time. (left) cross-section of the line and thresholds; (right) corresponding lines-width along the line. The local roughness is not affected by more than the photon noise.
6. POWER SPECTRAL DENSITY Comparing the line-width roughness of various lines in the same field allows us to estimate the power spectral density of the lines and their dependence on the imaging conditions (figure 10.)
Figure 10. Power spectral density of the LWR (averaged over 14 contiguous lines), for various exposure times at σ = 0.3 (left) and for three different σ values at a long exposure time (right.) All lines are statistically independent and identically distributed (i.i.d.)
Proc. of SPIE Vol. 9776 97760H-7
Downloaded From: http://proceedings.spiedigitallibrary.org/ on 04/26/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx
For longer exposure times and higher partial coherence settings (σ = 0.3), where the photon-noise contribution is small compared with the intrinsic LWR, it is possible to discern a sharp roll-off in the PSD at approximately 6 cycles/µm. We expect the limited numerical aperture (NA) of the lens to cause a spatial frequency cutoff near asin(4xNA/4)/ λ = 6.11 cycles/µm. Partial coherence (high σ) has the effect of the decreasing the steepness of the roll-off at higher spatial frequencies since there is less optical filtering, while the line width becomes more sensitive to photon-noise (which is not band-limited) at higher frequency.
7. CONCLUSION
Though resist line-edge roughness is generally governed by chemical processes (chemical shot noise, acidic diffusion), the roughness embedded in the aerial image, either as a replication of the mask pattern roughness, replicated surface roughness or photon noise, will continue to play a more significant role as pattern dimensions shrink. We have shown that it is possible to study the intrinsic aerial image LWR from actinic photo-mask imaging with sub-nanometer precision. In the examples presented, the conjunction of the illumination coherence and the intrinsic roughness of the mask creates speckle that cause LWR on the order of 5.8 nm on the mask side when studying a nominal CD of 160 nm, which can be measured using photon densities in excess of 4 photons/nm2, at the threshold intensity level. In addition, we found that although incoherent illumination (i.e. higher σ values) reduces speckle amplitude, a smaller NILS makes measurements more sensitive to photon shot noise. The (repeatable) effect of the reproduced mask roughness is replaced by the (statistical) influence of the photon noise, and results seems to indicate that increased exposure time does not affect the minimum LWR set by speckle formation, reducing the impact of speckle mitigation [11]. We recognize that pattern size, threshold level, illumination and NA play a role in the final measured LWR, and require further investigation.
ACKNOWLEDGEMENT
We would like to thank Intel Corporation for the manufacturing of the mask used in these experiments, Suchit Bhattarai and Valeriy Yashchuk for fruitful discussions, and David Johnson for his participation in the data collection. The LBNL EUV program is supported by Eureka, and initial funding for SHARP was from SEMATECH. The Advanced Light Source is supported by the Director, Office of Science, Office of Basic Energy Sciences, of the U.S. Department of Energy under Contract No. DE-AC02-05CH11231.
REFERENCES
[1] George, S. A., Naulleau, P. P., Mochi, I., Salmassi, F., Gullikson, E. M., Goldberg K. A., Anderson, E. H., “Extreme ultraviolet mask substrate surface roughness effects on lithographic patterning,” J. Vac. Sci. Technol., B 28(6), 23-30 (2010). [2] Naulleau, P. P., “Relevance of mask-roughness-induced printed line-edge roughness in recent and future extreme-ultraviolet lithography tests,” Appl. Opt. 43(20), 4025-4032 (2004). [3] Naulleau, P. P. and Gallatin, G., “Spatial scaling metrics of mask-induced line-edge roughness,” J. Vac. Sci. Technol., B 26(6), 1903-1910 (2008). [4] Naulleau, P. P., George, S. A., “Implications of image plane line-edge roughness requirements on extreme ultraviolet mask specifications,” Proc. SPIE 7379, 73790O1-11 (2009). [5] Zweber, A. E., Gallagher, E., Sanchez, M., Senna, T., Negishi, Y., Konishi, T., McGuire, A., Bozano, L., Brock. P., Truong, H., “EUV Mask Line Edge Roughness,” Proc. SPIE 8322, 83220O1-10 (2012). [6] Pret, A. V., Gronheid, R., Engelen, J., Yan, P.-Y., Leeson M. J., Younkin, T. R. “Evidence of speckle in extreme-UV lithography,” Appl. Opt. 20(23) 25970-25978 (2012). [7] Yang, P.-Y., Zhang G., Gullikson, E. M., Goldberg, K. A., Benk, M. P., “Understanding EUV mask blank surface roughness induced LWR and associated roughness requirement,” Proc. SPIE 9422, 94220J1-12 (2015).
Proc. of SPIE Vol. 9776 97760H-8
Downloaded From: http://proceedings.spiedigitallibrary.org/ on 04/26/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx
[8] Goldberg, K. A., Mochi, I., Benk, M. P., Lin, C., Allezy, A., Dickinson, M., Cork, C. W., Macdougall, J. B., Anderson, E,. H., Chao, W., Salmassi, F., Gullikson, E., M., Zehm, D., Vytla, V., Cork, W., DePonte, J., Picchi, G., Pekedis, A., Katayanagi, T., Jones, M. G., Martin, E., Naulleau P. P., Rekawa, S. B., “The SEMATECH high-NA actinic reticle review project (SHARP) EUV mask-imaging microscope, ” Proc. SPIE 8880, 88800T1-9 (2013). [9] Mangat, P., Verduijn, E., Wood, O. R., Benk M. P., Wojdyla, A., Goldberg, K. A., “Mask blank defect printability comparison using optical and SEM mask and wafer inspection and bright field actinic mask imaging,” Proc. SPIE 9658, 96580E1-8 (2015). [10] Wintz, D. T., Goldberg, K. A., Mochi, I., Huh, S., “Photon flux requirements for extreme ultraviolet reticle imaging in the 22- and 16-nm nodes,” J. Micro/Nanolith. MEMS MOEMS 9(4), 041205-1-8 (2010) [11] McClinton, B. M. and Naulleau, P. P., “Mask roughness induced LER control and mitigation: aberrations sensitivity study and alternate illumination scheme,” Proc. SPIE 7969, 79691Z1-11 (2011)
Proc. of SPIE Vol. 9776 97760H-9
Downloaded From: http://proceedings.spiedigitallibrary.org/ on 04/26/2016 Terms of Use: http://spiedigitallibrary.org/ss/TermsOfUse.aspx