1、Controlling Metrology Bias: Using a CD AFM to Calibrate Other Metrology SystemsIntroduction”CD measurements must account for side wall shape. Process control such as (stepper) focus, exposure and etchbias will require greater precision and 3D capability.” A restatement of this ITRS 2001(Metrology ch
2、apter, page 7) requirementis found in the ITRS 2002 update, Table98a: ”Measurement tool performanceneeds to be independent of targetshape, material and density (of features).”The ITRS Roadmap clearly underscoresthe necessity of measuring shape, aswell as critical dimensions, as workprogresses below
3、100nm. There areseveral reasons for this requirement: In order for device manufacturing to be profitable when criticaldimensions are below 100nm,metrology error must be less than 2nm. Alone, this 2nm precision is notenough, as it does not account for thetarget shape, material, and densitymentioned i
4、n the ITRS Roadmap. Due to an inherent lack ofprecision/capability, cross-sectionSEM is unable to serve as an absolutereference system.Effects traditionally requiring cross-sectionSEM calibration for nondestructive inlinemetrology include (1) sidewall angle, themetrology component of iso-dense bias,
5、Wall-Angle Variation through Feature Density (Iso-Dense Bias) in Etched Polysilicon Structuresacross chip-feature shape variation and(2) proximity effects related to materialproperties and local topographies.Use of precision to evaluate CD SEM(and, now, optical scatterometry*)cannot reliably measure
6、 these effects;therefore, CD SEM is insufficient as astand-alone metrology method below100nm. And although cross-sectionSEMs have provided an absolutereference for actual process calibrationin the past, this trusted method is nottruly viable as a single-reference system.Leading process developers ha
7、verecognized these shortcomings. Industry consortia such as InternationalSematech are actively pursuingalternative strategies for evaluating CDmetrology systems that incorporate CD AFM as a reference measurementsystem. Going forward, the term”linewidth (CD)” will gradually bereplaced by ”shape” (or
8、its components:top/middle/bottom CD, wall angles,linewidth variation) as this more complexmetric becomes the industry standard forlithography and etch control. The use ofappropriate reference measurement willbroaden the traditional role of precision,enabling it to account for variations inshape, mat
9、erial, and density.* Despite manufacturer claims that opticaltechniques accurately account for wall angle,independent studies (such as Reference #2,cited later in this piece) prove otherwise.Figure 1. The Dimension X3D provides directmeasurement of seven key shape parameters.Linewidth variabilityTop
10、 CDMiddle CDBottom CDDepthSidewallanglesSidewallroughnessandprofilesCons Very labor intensive Very long TAT Not precise/repeatable Very limited sampling Labor intensive Long TAT Not precise/repeatable Very limited sampling Labor intensive* Generally no material differentiation Inherent offset* Inher
11、ent offset No material differentiation Cannot confirm structure shape* Accurate High resolution Confirm structure shape Material differentiation Somewhat accurate Confirm structure shape Material differentiation with proper decoration Precise/repeatable Confirm structure shape Larger sampling possib
12、le Nondestructive Precise/repeatable Fast Largest sampling possible NondestructiveTEMX-sectionSEMCD AFMCD SEMProsTable 2. Comparison of metrology techniques (IBM).Table 1. Long-term repeatability study.Middle Width12341231231231233233.15234.1053231.7605233.0368233.4155233.4268232.9218233.0113232.075
13、5233.311233.295234.03632.03226.68226.2083227.1425226.6538227.0045226.7618227.4638227.2223227.4645227.324226.855227.65431.261.41207.028207.0193206.7155207.8208207.6905207.0788207.3438206.9103206.7375207.301207.445207.23831.04255.506254.9193256.1905256.0978256.2655256.1268256.1778255.6423255.1085255.8
14、21256.095255.68831.33268.56268.9993269.1695269.3098269.5825269.0598269.3938269.4523270.1075269.733269.225269.14331.18Site 1 Site 2 Site 3 Site 4 Site 5Nominal Dose + Dose + Focus + Focus +Day TrialPooled 3-The Dimension X3D AFMThe new Dimension X3D from VeecoInstruments is a 3D-metrology atomicforce
15、 microscope capable of makinghighly detailed measurements in threeaxes. It uses flared silicon tips as smallas 65nm to measure spaces as smallas 76nm and supports bothcharacterization and process control.Wall-Angle Variation throughFeature Density (Iso-Dense Bias)in Etched Polysilicon StructuresMuch
16、 has been done to try to extendCD SEM to account for effects such aswall angle; optical scatterometrycompanies now claim to be able toprovide calibration for sidewall angles.However, recent publications1,2,3document both CD SEMs and opticalscatterometry systems lack ofcorrelation with CD AFMs for wa
17、ll-angle information.CD AFMs are used as referencemeasurement systems4, especially forcalibration of other metrology systems, for a number of reasons: CD AFMs are not biased by shape,material, or feature density, all ofwhich destabilize measurementsprovided by CD SEMs and optical scatterometry. CD A
18、FMs are uniquely flexible in their ability to nondestructivelymeasure many shape parameters and sizes, including re-entrant profiles (Figure 1). CD AFMs are able to measurestatistically significant numbers of sample sites automatically.Since etch profiles can vary betweenisolated and nested lines (c
19、ausing shiftsin CD as measured by CD SEM),control of etch processes in state-of-the-art semiconductor fabs requiresdetailed information about sidewallshape. Direct measurement of etchshape by AFM, cross-section SEM,and TEM is the only way toconfidently correlate etch with theelectrical performance o
20、f isolatedtransistors, Kelvin probes, and finisheddevices. CD AFM, CD SEM, TEM,and cross-section SEM are often usedin the industry today to create areference system that calibratesstructures for tool evaluations or tomaintain process calibration within a production facility.Table 1 details a long-te
21、rm repeatabilitystudy based on 492 measurements onpolysilicon lines using a CDR70 tip(70nm nominal tip width). The lines wereprinted as part of a focus/expose-matrix etch experiment.Recent advances in CD AFM metrologyenable a greater dependence on thismethod than previously possible. Table 2,which i
22、s reprinted from M. Sendelbach and C.N. Archie2, compares four metrology techniques.* Labor intensive. This comment is based on a previous generation of laboratory AFMs. The X3D AFM isdesigned to support the fab environment for production and process calibration.* Inherent offset. This comment refer
23、s to the introduction of tip shape into previous CD AFM results. The X3DAFMs automated tip calibration eliminates the offset.* This comment has been added to the table by the authors of this application note.151050200 400 600 800 1000Figure 5. “Bottom-top” width bias for etched polysilicon linesthro
24、ugh pitch.Pitch, nmWall Angle versus PatternDensity of Polysilicon LinesThe Dimension X3D can accurately andrepeatably measure wall angle as afunction of varying line pitch in post-etch polysilicon (Figures 2 and 3). Etched polysilicon lines were measuredwith the X3D; four different linewidthswere s
25、ampled, each at four differentpitches. The sidewall angles, as well asthe difference between top and bottomCD, are distinctive functions of pitch,with 100nm ground-rule lines (pitch =260nm) giving shallower sidewallangle and almost 15nm differencebetween top CD and bottom CD.Figure 2. This 200 600nm
26、 pitchshows isolated and dense features.Figure 3. In this X3D trial, four different linewidths were sampled, each atfour different pitches.Three examples of etched polysilicon wall anglesinteracting with metrology systems will be studied here.A CD AFM serves as the reference measurementsystem in eac
27、h example.90.089.589.088.588.087.587.0200 400 600 800 1000Figure 4. Polysilicon sidewall angle through pitch.Pitch, nmThe sidewall angle is a sensitivefunction of pitch, especially atminimum pitches and for lines near the 100nm ground rule. The sidewallangle changes by 1.5 per 100nm,indicating that
28、chips printed withrelaxed ground rules will likely havedifferent polysilicon etch profiles thanthose at the nominal pitch. During thecycle of ramping a technology node,the etch process will need to becharacterized after each mask setchange as the minimum feature sizeshifts towards smaller lines (Fig
29、ure 4).The difference between bottom andtop width is also a concern in this etchprocess, especially with the re-entrantprofile. CD SEM measurement willhave a strong scattering peak at thetop corner, be blind to the re-entrantportion, and be broadened by thebottom width. While library-basedinterpreta
30、tion of complex SEM signalshas been proposed3, this approachrequires the parameterization of cornerrounding and slope of the (presumablystraight) sidewall, as well as input offeature height. In the AFM measure-ment, the top-bottom width differenceis measured directly, independent offilm stack, profi
31、le shape, transitions, orre-entrant profiles. The results show a4nm width change between pitches of250 350nm independent of linesize, but correlated to pitch (Figure 5).100nm lines120nm lines160nm lines100nm lines120nm lines160nm linesFigure 6. Cross-section data from 193nm photoresist lines/spaces.
32、Figure 7. Linewidth analysis is presented for ten scans, replete with summary data andline-by-line detail for depth, top/middle/bottom widths, and left and right sidewalls.The X3D demonstrates that patterndensity correlates to sidewall-anglevariation for etched polysilicon lineswhen printed close to
33、 the ground-ruleline/pitch. The CD AFM resultspresented here are for sidewalls lessthan 90, although the AFM can alsomeasure the special case of re-entrantfeatures (greater than 90 sidewalls), asshown in Figure 6. Standard linewidthanalysis gives top/middle/bottomwidths as well as sidewall angles fo
34、rboth sidewalls. The data set reportedhere comes from the standard sidewall-analysis results.It is important to note that eachmeasurement site produces the datashown in Figure 7. Thus, whenproduction monitoring requires wallangle, CD, and depth, a single CDAFM metrology recipe can replace the multip
35、le metrology steps needed toobtain this data using other methods(e.g., cross-section SEM, CD SEM, 1D AFM, film metrology). Throughputresults of up to 65 sites per hour havebeen obtained with the X3D.CD SEM Bias Dependence on Polysilicon Line Pitch: V. Ukraintsev (Texas Instruments),SPIE, 2003Figure
36、8 shows a data set with CDSEM bias correlated to pattern density(CD SEM bias is defined here as mid-derivative line by CD SEM minus CDAFM at 10% of line height). In thiscase, wall angle is not explicitly shown. V. Ukraintsev1 states in his analysis:”Figure 8 shows probably the mostdrastic case of sa
37、mple-to-sample CDSEM bias variation. SEM bias variationas high as 15nm has been detected.These data represent measurementsdone on periodic structures of100nm polysilicon lines placed atvarious pitches. The line orientation onthe wafer vertical vs. horizontalapparently affects SEM bias as well.Throug
38、h pitch optical proximitycorrection (OPC) is a very common lineCD adjustment used by every chipmanufacturer. CD SEM is currently usedfor the input data collection for theOPC models. Should the SEM datapresented in the figure be used as anOPC model input the physicaldimensions of the line through pit
39、ch willvary as much as (10 15)nm. The realdanger of the situation is that noFigure 8. CD SEM biasversus pitch for horizontaland vertical lines.300 400 500 600 700 800 900 1000 1100 12002520151050Pitch, nmHorizontal poly-Si linesVertical poly-Si linesAveragedproblem would be detected if the sameSEM i
40、s used to monitor the quality ofthe OPC. Therefore, the problem willbecome evident only at a later stage of the process development whenelectrical data are available to traceacross chip CD non-uniformity.”The CD SEM bias could well describethe relationship between the width asmeasured at a fixed hei
41、ght (10%)to the scatterometry data at the variousheights above the base, as evidencedby the changing TMU at each verticalheight. This lack of direct correlationshows two things: All scatterometer systems demonstratea nonlinear response through patterndensity variation with respect to CDAFM (slope of
42、 plots of CD AFMversus scatterometer = 0.65 0.75, linear response would have a slope of 1).* There is a variation in response for a given scatterometer system at different feature heights. Thus, it can be concluded that highlyrepeatable (precise) optical scattero-meter results cannot be correctlyint
43、erpreted with any certainty below 8 15nm without using a referencesystem that is accurate as well as precise. What are the consequences? First,characterization of process variation forsmall features using scatterometry results(without calibration to an absolutereference system) would lead toincorrec
44、t quantitative conclusions. Forinstance, a standard focus/exposematrix used to characterize etchprocess window would always show asmaller process window than actuallyexists. This statement holds true for allOptical Scatterometry Pattern-Density Dependency: M. Sendelbach and C.N. Archie (IBM Research
45、),SPIE, 2003Recently, IBM evaluated fourscatterometry systems using a referencemeasurement system consisting of a CDAFM calibrated by TEM and/or cross-section SEM. This study, published atSPIE Microlithography 2003, shows awide range of optical scatterometryresponses to etched polysilicon lines, 40
46、90nm, printed with a fixed pitch.IBM used the TMU (Total MeasurementUncertainty) method to evaluate thesesystems. TMU gives an uncertaintyvalue in nanometers that incorporatesthe measurement uncertainty for boththe reference measurement system andthe tool-under-test, yielding resultssimilar to ordin
47、ary least-squares-fit andR2 comparisons.Figure 10 (from page 28 of thepresentation) shows that the TMU variesas a function of (1) tool and (2) heightfrom the base of the feature. As in theexample from Texas Instruments for CDSEM versus CD AFM, the scatterometerresult is compared to the CD AFM atmult
48、iple heights above the feature base.Wall-angle changes directly observedwith the AFM do not correlate directlyFigure 10. These scatterometry results show varying uncertainty as a function of CDmeasurement height on samples with varying wall angles as measured by CD AFM.TMU* AllTMU* 10%TMU* 50%TMU* 7
49、5%181614121086420 A B C Dabove the feature bottom and the mid-width of a feature with varying-slopesidewall. Ukraintsevs data aboveshows 5nm bias change between250 350nm pitch. This changecompares to the previous studys 4nmtop-to-bottom width change within thesame pitch range on polysilicon lines.J.S. Villarrubia3 of NIST discusses thephenomenon in some detail. He states:”if normal manufacturing processvariation results in some randomness inedge shape sic, wall angle, there willbe a random component of error, ak
