sensitivity study of two high-throughput resolution metrics for photoresists

Sensitivity study of two high-throughput resolutionmetrics for photoresists

Christopher N. Anderson1,* and Patrick P. Naulleau2

1Applied Science and Technology Group, University of California at Berkeley, Berkeley, California 94720, USA2Center for X-Ray Optics, Lawrence Berkeley National Laboratory, 1 Cyclotron Road, Berkeley, California 94720, USA

*Corresponding author: [email protected]

Received 5 September 2007; revised 8 November 2007; accepted 9 November 2007;posted 9 November 2007 (Doc. ID 87231); published 20 December 2007

The resolution of chemically amplified resists is becoming an increasing concern, especially for lithog-raphy in the extreme ultraviolet (EUV) regime. Large-scale screening is currently under way to identifyresist platforms that can support the demanding specifications required for EUV lithography. Currentscreening processes would benefit from the development of metrics that can objectively quantify resistresolution in a high-throughput fashion. Here we examine two high-throughput metrics for resist reso-lution determination. After summarizing their details and justifying their utility, we characterize thesensitivity of both metrics to known uncertainties in exposure tool aberrations and focus control. For animplementation at EUV wavelengths, we report aberration and focus-limited error bars in extractedresolution of �1.25 nm rms for both metrics, making them attractive candidates for future screening anddownselection efforts. © 2008 Optical Society of America

OCIS codes: 070.0070, 100.3190, 110.3960.

1. Introduction

As lithography pushes to smaller and smaller featuresizes, the fidelity of chemically amplified resistsbecomes an increasing concern. This is especially truein the extreme ultraviolet (EUV) regime �� 13.5 nm�, where source power constraints place fur-ther limitations on resist sensitivity. For the past 2years, the ability to simultaneously achieve the res-olution, sensitivity, and line-edge-roughness (LER)required for EUV resist has been named in the toptwo highest-risk potential roadblocks to the commer-cialization of EUV lithography [1,2]. Currently thebest way to identify EUV resists showing promisefor future commercialization is through large-scalescreening and downselection based on printing per-formance in a variety of areas, including sensitivity,LER, and resolution. Of these three benchmarks, re-sist resolution is arguably the most important param-eter at the present time; it is also the most difficultparameter to objectively quantify.

It is reasonable to assume that resolution limits inchemically amplified resist are determined in largepart by acid diffusion. The measurement of this dif-fusion would be expected to provide an effective mea-sure of resolution. It has been suggested that theresist isofocal bias [3] and LER correlation length [4]measurements may be reasonable gauges of the aciddiffusion length. Recent findings [5], however, showthat these metrics do not extract resolutions that areconsistent with direct observation. One metric thathas had success in quantifying resist resolution in amanner consistent with direct observation is thatbased on the resist modulation transfer function(MTF) [6,7].

The drawback of the MTF approach is that it re-quires a large amount of exposure and scanning elec-tron microscopy (SEM) data for resolution extraction.For screening purposes requiring fast turnaroundtimes, the MTF metric has limited utility. In at-tempts to develop metrics suitable for high-volumescreening efforts, we will investigate two high-throughput resolution metrics that are based on thesame linear-systems principle of the MTF metric. Themetrics involve (1) measuring the size of printed con-tacts through dose [8] and (2) measuring the imaging

0003-6935/08/010056-08$15.00/0© 2008 Optical Society of America

56 APPLIED OPTICS � Vol. 47, No. 1 � 1 January 2008

fidelity of a corner of a large feature, i.e., the amountof rounding in the printed corner [5,9]. Once the util-ity of both metrics is justified, we determine the sen-sitivity of each metric to known uncertainties inexposure tool aberrations and focus control.

2. Aerial Image Modeling Limitations

The exact processes that describe how an incidentaerial image is converted to a latent deprotectionprofile are not well understood at the present time.Owing to the difficulty in tracking and predicting thecomplicated chemical diffusion, and stochastic pro-cesses that occur during exposure, bake, and devel-opment, several authors [5,10] have turned to asimplified point-spread function (PSF) linear systemsapproach to modeling resist exposure dynamics. Theidea behind this approach is to lump all of the com-plicated chemical and diffusion processes during anexposure and bake them into an effective blur func-tion that converts the incident aerial image to alatent deprotection profile through a simple convolu-tion process. Due to the simplicity of this model, ex-tracting the effective resist blur from printing data isa relatively straightforward process.

Models based on the linear-systems PSF approachare convenient because they provide an intuitive linkto the resist resolution limit. As with many resistmodels, their success relies on the ability to accu-rately predict the aerial image incident at the wafersurface. In practice, limited knowledge of the exper-imental conditions in any given exposure hampersour ability to accurately model the aerial image. Withexposure tools constantly pushing the limits of theirimaging optics, the sensitivity of the aerial image tosmall changes in aberrations and focus is a possibleconcern for the success of PSF-based metrics. As anillustrating example we consider these effects for animplementation at EUV wavelengths. The parame-ters we describe correspond to the those found at theSEMATECH Berkeley microfield exposure tool(MET) printing facility [11].

The focus steps in a typical focus-exposure-matrix(FEM) at the Berkeley facility are on the order of50 nm. Assuming that nominal focus is somewhere inthe FEM, the random variable associated with aerialimage defocus of the best-focus row in the FEM isuniform on the interval ��25, 25� nm with a stan-dard deviation of �14 nm. For high-resolution expo-sure tools such as the 0.3 numerical aperture (NA)SEMATECH Berkeley MET, this magnitude of defo-cus may be enough to noticeably reduce the aerialimage contrast at high spatial frequencies.

Regarding aberrations, the rms error in interfero-metrically measured aberrations of the SEMATECHBerkeley MET optic is 0.1545 nm [12], correspondingto an �10%–20% error bar in reported Zernike coef-ficients used in aerial image modeling software. Witherror bars of that magnitude one would expect anupper limit on the ability to accurately model theaerial image as printed feature sizes shrink towardthe diffraction limit of the imaging optic. With expo-sure condition knowledge always limited to some ex-

tent, it is of interest to investigate the impact this hason the ability of PSF-based resolution metrics to ex-tract credible resolution numbers.

3. Two High-Throughput Resolution Metrics

The utility of a resolution metric is arguably bestdescribed by its ability to produce robust, credibleresults that agree with observed resolution limits. Inpractice, metric utility can also be characterized byan efficiency measure, or by the amount of exposuretool use hours, SEM images, and modeling supportthat is required for resolution extraction. In efforts tofind high-throughput resolution metrics suitable forlarge-scale screening and resist downselection, westudy metrics that require �10 SEM images andhave relatively low overhead in terms of modelingsupport needed for blur extraction.

A. Contact Metric

The first metric we discuss is based on measuringthrough-dose contact printing results [8]. The ex-tended Nijboer–Zernike method described in [8] re-quires an entire FEM of SEM images to extract thediffusion length. We choose to study a simplified con-tact metric that does not isolate the effects of spher-ical aberrations and tool focus in determining theresolution. The reduced complexity of this modelcomes at the price of possibly increasing the sensitiv-ity of the metric to aberrations and focus. It turns out,however, that the simplified metric has a low enough

Fig. 1. Fifty nanometer contact metric: (a) Modeled deprotectionimage cross sections for resist blurs of 0, 5, . . . , 35 nm. Lightercolors are larger blurs. (b) Experimental PD versus relative dosecurves for MET1K and EUV2D resists.

1 January 2008 � Vol. 47, No. 1 � APPLIED OPTICS 57

sensitivity to reasonable levels of focus and aberra-tion uncertainty that its use is worthwhile. The met-ric we study involves measuring the printed diameter(PD) of through-dose contacts at nominal focus andcomparing these data to modeled PD versus dosecurves at nominal focus for varying degrees of resistblurs, assuming the simple PSF-blur model. Using aleast-squares method we find the modeled blur thatmost closely resembles the experimental data.

In our specific implementation of EUV resist test-ing using a 0.3 NA exposure tool, we choose to mea-sure a contact size of 50 nm with a pitch of 150 nm. InFig. 1(a), modeling data show that increased resistblurring leads to reduced contrast in the latent depro-tection image. The result is that under the simplethresholding model for resist, poorer performing re-sists (larger blurs) have smaller relative dose rangesfor which contacts can print. Figure 2 shows through-dose contact printing for Rohm and Haas EUV2D andMET1K resists with equal (15%) relative dose stepsbetween exposures. The corresponding PD versus rel-ative dose curves for these data [see Fig. 1(b)] showthe mentioned reduction in relative dose latitudefor the poorer performing resist (EUV2D). Here wewish to address the degree to which the slope differ-ences in the PD versus relative dose curves can beused to quantify differences in the intrinsic resolutionof the two resists.

B. Corner Metric

The second resolution metric we consider is based onthe measurement of the imaging fidelity of a corneron a large feature [5,9]. In Fig. 3(a), modeling datasuggest that higher levels of resist blur lead to in-creased corner rounding in the final binarized resistimage. Experimental support of this result is seen bycomparing the amount of corner-rounding present inresists that are well known to have different resolu-tions. Figures 3(b) and 3(c) show SEM images of thecorner of a 700 nm elbow at dose to size [13] taken atidentical magnifications in EUV2D and MET1K re-sists. It is well-known that MET1K supports higherresolutions than EUV2D [5]; the larger corner round-ing present in the EUV2D platform is consistent withthe predictions of the PSF model. We extract resolu-tion with this metric by comparing the amount ofexperimental corner rounding in a large, isolated fea-ture to the amount of corner rounding in the equiv-alent modeling data with varying degrees of resistblur.

To quantify the amount of rounding in a givencorner, we have developed and tested three differentmethods and selected the one with the least sensitiv-ity to noisy experimental data. We settled on a metricthat uses the removed area to indirectly compute aneffective corner radius. As shown in Fig. 4, we usein-house software to take radial lineouts of an exper-imental (or modeled) dose-to-size image to extract aradius versus angle profile of the experimental (mod-eled) corner edge. By extrapolating the flat parts ofthe elbow out to the ideal (nonrounded) corner loca-tion we are able to generate a radius versus angleprofile for the ideal corner edge and compute the areathat has been removed in converting the aerial imageto a printed resist image.

4. PSF-Based Resolution Metrics: Sensitivity to Focusand Aberrations

All PSF-based resolution metrics require aerial im-age modeling for resolution extraction. In this paper

Fig. 2. Through-dose 50 nm contact printing in (a) MET1K and(b) EUV2D resists with 15% relative dose steps between exposures.

Fig. 3. (a) Modeled rounding of a 700 nm elbow corner for varyingdegrees of deprotection blur. Blur numbers are given in nanome-ters, FWHM. (b), (c) Equal magnification SEM images of a 700 nmelbow corner at dose to size [9] in MET1K and EUV2D resists,respectively.


all aerial images are generated using in-house soft-ware that supports arbitrarily defined optical aber-rations, tunable defocus, and customizable pupil fills[14]. For our specific implementation at EUV wave-lengths, the base modeling parameters are set tomatch the experimental conditions of the SEMATECHBerkeley MET so that modeling data can be directlycompared to experimental data obtained at the Berke-ley facility. This includes a 5� reduction imagingconfiguration, � � 0.35–0.55 annular pupil fill, andoptical aberrations set to match the aberrationsmeasured during the initial interferometricSEMATECH Berkeley MET tool alignment [12].

A. Corner Metric

For the implementation at EUV wavelengths, ourobject of choice is a dark-field 700 nm elbow (the el-bow is bright). To model the 700 nm elbow, we as-sume an idealized binary mask as an approximationto the realistic e-beam-written, multilayer-coatedEUV mask. We believe these approximations to bereasonable owing to the fact that in terms of cornerfidelity, the feature size of interest is very large rel-ative to the 13.5 nm wavelength being used in theexposure tool. Furthermore, the sub-50 nm resolu-tion of the e-beam process used to fabricate the 5�mask (3.5 �m elbow) ensures that mask making tech-nology supports a much higher resolution than theMET optic.

1. Corner Metric Focus SensitivityFocus sensitivity is examined by generating throughfocus aerial images in 10 nm focus steps spanning��50, 50� nm from the nominal focus plane. The nom-inal plane of best focus is defined as the image planewith the least corner rounding at dose to size withbase modeling parameters. The aerial images fromeach defocus plane are then blurred by convolutionwith an assumed resist PSF [7,8] at full width athalf-maximum (FWHM) blur levels ranging from5–60 nm. For each level of defocus, we measure thecorner rounding for the entire series of blurred depro-tection images and generate a corner rounding versus

PSF-blur curve. In this manner, we build up round-ing versus deprotection blur curves at each modeleddefocus.

Figure 5 shows corner rounding versus blur curvesfor five focus steps spanning a ��50, 50� nm range ofdefocus values. To work out the focus-induced errorbars for the extracted PSF blur, we pick a midrangecorner radius (radius � 90 nm; blur � 30 nm) andtake the statistics of the blur values from differentfocus curves that intercept the corner radius value(see Figs. 6(c) and 6(d); although they are aberrationstudy plots they show the method used for computingthe blur values). At �30 nm of blur, we report focus-limited errors in extracted blur of 1.65 nm peak tovalley and 0.77 nm rms.

2. Corner Metric Aberration SensitivityAberration sensitivity is investigated by generatingaerial images at nominal focus with varying degreesof random noise added to the first 37 Zernike coeffi-cients used in the base-aberrations optical model (ex-

Fig. 4. Left, screenshot of software used to extract corner rounding from an experimental SEM image. Right, radial lineouts for the ideal(nonrounded) and actual (rounded) corner edges.

Fig. 5. Modeled rounding versus blur curves for the cornermetric focus study. There are five focus curves associated with��50, �25, 0, 25, 50� nm of defocus.


cluding piston, tilt, and focus). We study four rmsnoise levels, 0%, 10%, 20%, and 30%, and for eachnoise level (except 0%) we generate ten random ab-erration maps and their corresponding aerial images.For each aerial image at a given rms noise level, wegenerate deprotection images through blur, measurethe corner rounding at dose to size, and produce afamily of rounding versus blur curves; one curve foreach of the ten aberration maps in a particular noiselevel.

Figure 6 shows the families of rounding versus blurcurves for rms aberration noise levels of 10%, 20%,and 30%. Aberration-limited error bars are workedout exactly as they are for focus [see Figs. 6(c) and6(d)] at an average blur of �30 nm. Table 1 summa-rizes the results. We note that blur error bar magni-tudes (in an absolute sense) have a slight dependenceon blur owing to the fact that smaller blurs do a betterjob mapping small aerial image changes into thedeprotection image.

B. Contact Metric

1. Contact Metric Focus SensitivityFor the contact metric, our modeled object is a 300� 300 nm patch of 50 nm, 150 nm pitch, dark-field

contacts (the contacts are bright), which throughsampling effects is equivalent to modeling an infinite2D array of contacts. We again assume an idealizedbinary mask. To generate blurred aerial imagesthrough focus, we follow a procedure identical to thatoutlined in Subsection 4.A.1. For contacts, we definethe nominal plane of best focus as the plane with thehighest average aerial image contrast when the ab-errations are set to match our base standard.

For each modeled focus-blur combination, we usein-house software to measure the PD through doseand generate a PD versus relative dose curve. Rela-tive dose is obtained by normalizing absolute dose to

Fig. 6. Modeled rounding versus blur curves for the corner metric aberration study. There are ten random aberration curves at each rmsnoise level: (a) 10%, (b) 20%, and (c) 30%. (d) Zoomed 30% noise plot centered around the radius of 90 nm. Solid curves indicate modeledradius versus blur data for the ten aberration maps in the 30% noise level. The intersections of the horizontal dashed line at radius � 90 nmwith the ten modeled curves are traced down with vertical dashed lines to show the range of blurs that might produce a rounding of 90 nmassuming a 30% rms uncertainty in optical aberrations.

Table 1. Aberration-Limited Error Bars for the Corner RoundingMetrica

rms Noise Level(%)

Error Peak-to-Valley(nm)

Error �(nm)

10 1.20 0.4120 2.25 0.7030 3.40 1.11

aData are taken at a modeled blur of �30 nm. We note valueswill increase slightly for smaller blurs and decrease slightly forlarger blurs.


dose to size; each focus-blur combination has aunique dose where this occurs. Figure 7 shows thefamily of these curves for five focus steps spanning

��50, 50� nm of defocus and for blurs ranging from 0to 35 nm in 5 nm steps. The series of focus curveswithin a given blur level are plotted with the samegray level, and there are different gray levels for eachblur level. Note the curves in the 0 and 5 nm blurlevels overlap heavily. We do not process the largerblurs (40, 50, 60 nm) for the contact metric as theyseverely reduce the contrast in the deprotection im-age.

To work out the focus-limited error bars we use aleast-squares approach. For each defocused trialcurve in the 30 nm blur level we find the blur whosenominal-focus blur versus relative dose curve mostclosely resembles the defocused trial curve in terms ofleast-squared error (LSE). By cataloging the range ofnominal blurs spanned by the series of defocused trialcurves we report focus-limited error bars of 1.70 nmpeak to valley and 0.83 nm rms.

2. Contact Metric Aberration SensitivityAerial images are generated following the procedureoutlined in Subsection 4.A.2. For each of the ten im-ages at a given noise level (10%, 20%, and 30%) wegenerate PD versus relative dose curves for blursranging from 0 to 35 nm in 5 nm steps and plot themso that the series of ten random aberration curveswithin a given blur level is plotted with the same gray

Fig. 8. Modeled PD versus relative dose curves for the contact metric aberration study. Each gray level corresponds to one blur. Thereare eight blurs spanning 0–35 nm in 5 nm steps. Each blur level contains ten curves, one from each of the ten randomly generated opticalaberration maps within a given aberration noise level. Plots (a)–(d) are associated with rms aberration noise levels of 0%, 10%, 20%, and30%, respectively. Note the heavy overlap of the 0 and 5 nm blur level curves.

Fig. 7. Modeled PD versus relative dose curves for the contactmetric focus study. Each gray level corresponds to one blur. Thereare eight blurs spanning 0–35 nm in 5 nm steps. Each blur levelcontains five focus curves associated with ��50, �25, 0, 25, 50� nmof defocus. Note the heavy overlap of the 0 and 5 nm blur curves.


level, different gray levels for each blur level. Figure8 shows the families of curves for rms aberrationnoise levels of 0%, 10%, 20%, and 30%. To work outthe aberration-limited error bars for a given noiselevel, we use the same least-squares approach as inthe focus study except that the target curves we fit tohave changed as a result of aberrations, not defocus.Table 2 summarizes the aberration-limited error barsfor the contact metric.

5. Discussion

As mentioned above, the rms error in interferomet-rically measured aberrations of the SEMATECHBerkeley MET optic puts 10%–20% rms error bars onZernike coefficients used for aerial image modeling.This information suggests that error bars from the20% rms aberration noise level should provide anupper limit on resolution errors originating from in-complete aberration knowledge. Keeping the focussteps in the exposure tool �50 nm limits defocus to a��25, 25� nm range, indicating that the focus-limitederror bars we report, which are based on a ��50, 50�nm defocus range, provide an upper limit on resolu-tion errors originating from tool focus uncertainty.

Overall, modeling suggests that at the Berkeleyfacility both resolution metrics are slightly more vul-nerable to uncertainties in aberrations than to un-certainties in focus. With the naive assumption ofindependence we can add the focus and aberrationerror bars in quadrature to report an estimate of thetotal error in extracted blur for the two metrics. Foran implementation at EUV wavelengths, we reporttotal error bars in the extracted PSF width of�2.79 nm peak to valley and 1.04 nm rms for thecorner metric and 3.32 nm peak to valley and1.23 nm rms for the contact metric.

In the study presented here we have assumed arelatively simple PSF-based model for resist expo-sure dynamics. We believe, however, that the re-ported error bars should be applicable in moreadvanced resist models owing to the fact that theerror bars are associated with aerial image predictionlimitations that are not unique to this metric.

In practical terms of resist screening, the metricsdescribed here have a limited ability to compare tworesists on an absolute scale. Consider two resists Aand B exposed 6 months apart. If the difference inextracted blurs is within one error bar, the metricsdescribed here cannot reliably say which resist hasthe better resolution. If, however, resists A and B are

exposed on the same day, it is very likely that theywill see almost identical tool aberrations. In thismanner error bars can be reduced to the subnano-meter rms focus-limited regime, enabling highlyaccurate relative comparisons useful in resist param-eter optimization. In addition, averaging can be im-plemented to mitigate statistically additional formsof error: human, corner-rounding extraction, toolmalfunctions, etc., further reducing tool-limited errorbars.

We note that for our implementation at EUV wave-lengths, modeling suggests that both metrics will notwork for PSF blurs �5 nm. For the contact metric(see Figs. 7 and 8) it is easy to see that the curves forthe 0 and 5 nm blur levels overlap in all of the defo-cused and aberrated trials. This indicates that inpractice one cannot tell the difference between resistswith blurs �5 nm with this metric. Although it isslightly less obvious, the breakdown of the cornermetric at 5 nm is also evident from the data pre-sented here. We mentioned before that the error barsincrease (decrease) slightly with smaller (larger) re-sist blurs. For blurs �5 nm, the error bars become asizable portion ��50%� of the blur value, renderingboth metrics useless in this regime.

We note, however, that it is reasonable to assume aminimal fidelity loss in converting the aerial image toa deprotection profile when the resist blur is muchsmaller than the PSF blur of the imaging optic. Theresolution of the SEMATECH Berkeley MET optic is�22 nm [11]. That said, we expect to see very lit-tle difference in printing for resists with blurs�5–10 nm for our EUV implentation. Of course, asthe resolution of the lithography tool improves, wewould expect the utility of both resolution metrics toextend into the sub-5 nm blur regime. For example,we would expect to have no problem measuring a5 nm resist blur with an 8 nm e-beam tool providedthe 8 nm tool aberrations are characterized wellenough to allow the aberration-limited error bars tosupport this measurement.

The sources of error described here can be catego-rized into a group of errors that limit the ability toaccurately model the aerial image at the wafer sur-face in a given exposure. As we have seen, thesemodel-limiting error sources put constraints on thecredibility of resolution numbers extracted with thecorner and contact metrics. In addition to these errorsources, there are other errors that affect the credi-bility of extracted resolution numbers. One exampleto study is how SEM focus affects the measured PDand corner-rounding numbers. Another example tolook at is how tool dose errors affect the slopes of themeasured PD versus dose curves (note this only ap-plies to the contact metric). We are also interested instudying how LER impacts our analysis software interms of consistently delivering the same PD androunding numbers for different copies of the samecoded feature. Another useful thing to study would bethe shot-to-shot variations in extracted PD and cor-ner rounding for a full FEM of identically preparedfeatures. In a way, this study could essentially lump

Table 2. Aberration-Limited Error Bars for the Contact Metrica

rms Noise Level(%)

Error Peak-to-Valley(nm)

Error �(nm)

10 1.40 0.4620 2.85 0.9130 4.90 1.53

aData are taken at a modeled blur of �30 nm. We note valueswill increase slightly for smaller blurs and decrease slightly forlarger blurs.


the effects of SEM focus, tool focus, and LER into afull process error bar.

6. Summary

In this paper we have studied two high-throughputPSF-based photoresist resolution metrics, justifiedtheir utility, and characterized their sensitivity toknown uncertainties in exposure tool aberrations andfocus control. In our specific implementation at EUVwavelengths with exposure conditions matchingthose at the SEMATECH Berkeley MET printing fa-cility, modeling suggests that PSF-based resolutionmetrics have focus- and aberration-limited error barsin extracted resolution of �3 nm peak-to-valley and1.25 nm rms. As the PSF-based metrics consideredhere require minimal exposure data and relativelylow overhead in terms of modeling support and SEMimages, they are attractive platforms for large-scaleresist screening and downselection efforts.

The authors are greatly indebted to Paul Denham,Ken Goldberg, Brian Hoef, Gideon Jones, and JerrinChiu of the Center for X-Ray Optics (CXRO) at Law-rence Berkeley National Laboratory for expert supportwith the exposure tool as well as the entire CXROengineering team for building and maintaining theEUV exposure tool. We also thank Jim Thackeray andKathleen Spear of Rohm and Haas for resist support.The authors are grateful for support from the NationalScience Foundation EUV Science and Technology Cen-ter. This research was also supported by SEMATECHand performed at Lawrence Berkeley National Labo-ratory using the SEMATECH MET exposure facilityat the Advanced Light Source. Lawrence Berkeley Na-tional Laboratory is operated under the auspices of theDirector, Office of Science, Office of Basic Energy Sci-ence, of the U.S. Department of Energy.

References and Notes1. S. Wurm, “EUV lithography development in the United

States,” presented at the Fourth International EUV Lithog-

raphy Symposium, San Diego, Calif., 7–9 November 2005,proceedings available from SEMATECH, Austin, Tex.

2. S. Wurm, “EUV lithography update,” presented at theFifth International Symposium on EUV Lithography, Bar-celona, Spain, 15–18 October 2006, proceedings availablefrom SEMATECH, Austin, Tex.

3. G. M. Schmid, M. D. Stewart, C. Wang, B. D. Vogt, M. Vivek,E. K. Lin, and C. G. Willson, “Resolution limitations in chem-ically amplified photoresist systems,” Proc. SPIE 5376, 333–342 (2004).

4. G. F. Lorusso, P. Leunissen, M. Ercken, C. Delvaux, F. V. Roey,and N. Vandenbroeck, “Spectral analysis of line width rough-ness and its applications to immersion lithography,” J. Micro-lithogr., Microfab., Microsyst. 5, 033003 (2006).

5. P. Naulleau and C. Anderson, “Lithographic metrics for thedetermination of intrinsic resolution limits in EUV resists,”Proc. SPIE 6517, 65172N (2007).

6. J. Hoffnagle, W. D. Hinsberg, F. A. Houle, and M. I. San-chez, “Characterization of photoresist spatial resolution byinterferometric lithography,” Proc. SPIE 5038, 464–472(2003).

7. T. Brunner, C. Fonseca, N. Seong, and M. Burkhardt, “Impactof resist blur on MEF, OPC, and PD control,” Proc. SPIE 5377,141–149 (2004).

8. P. Dirksen, J. Braat, A. J. E. M. Janssen, A. Leeuwestein, H.Kwinten, and D. Van Steenwinckel, “Determination of resistparameters using the extended Nijboer–Zernike theory,” Proc.SPIE 5377, 150–159 (2004).

9. R. Jones and J. Byers, “Theoretical corner rounding analysisand mask writer simulation,” Proc. SPIE 5040, 1035–1043(2003).

10. C. Ahn, H. Kim, and K. Baik, “Novel approximate model forresist process,” Proc. SPIE 3334, 752–763 (1998).

11. P. Naulleau, “Status of EUV micro-exposure capabilities at theALS using the 0.3-NA MET optic,” Proc. SPIE 5374, 881–891(2004).

12. K. Goldberg, P. Naulleau, P. Denham, S. Rekawa, K. Jackson,E. Anderson, and J. A. Liddle, “At-wavelength alignment andtesting of the 0.3 NA MET optic,” J. Vac. Sci. Technol. B 22,2956–2961 (2004).

13. As an external control, all experimental and modeling cornerdata are taken at and around the dose where the coded 100 nmfeatures print at 100 nm.

14. Note that commercial modeling packages such as PROLITH andSOLID E could also be used.


sensitivity study of two high-throughput resolution metrics for photoresists

Documents