INTERFACE '89 This paper was published in the proceedings of the KTI Microelectronics Seminar, Interface '89, pp. 209-215. It is made available as an electronic reprint with permission of KTI Chemicals, Inc. Copyright 1989. One print or electronic copy may be made for personal use only. Systematic or multiple reproduction, distribution to multiple locations via electronic or other means, duplication of any material in this paper for a fee or for commercial purposes, or modification of the content of the paper are prohibited. ,"f;~:.- OPTIMUM STEPPER PERFORMANCE THROUGH IMAGE MANIPULATION Chris A. Mack Department of Defense Fort Meade, Maryland Chris Mack received his B.S. degrees in Physics, length ranges. The new Micrascan I, the first ma- Chemistry. Electrical Engineering and Chemical chine to employ the long awaited "step-and-scan" Engineerin,';: in 1982. He joined the Microelec- technology, brings the possibility of variable tronics Research Laboratory of the Department of wavelength to stepper technology.2 The use of Defense in 1982 and began work in opticallitho,';:- excimer lasers as the light source for deep-UV raphy research. He has authored papers in the steppers brings with it the potential for generating area of optical lithography and has developed the an arbitrarily shaped illumination source by litho,,;:raphysimulation program PROLlTH. scanning the laser in the desired shape.3 ABSTRACT It is the goal of this paper to investigate the potential advantages of using flexible steppers for The advent of flexible steppers, allowing variation "image manipulation," that is, varying the aerial in the numerical aperture, partial coherence, and image of a given mask pattern by varying the possibly other optical parameters, allows new optical parameters of the stepper. Obviously, the opportunities for optimization. This paper will purpose of image manipulation will be to improve discuss a method for picking the optimum numeri- the lithographic process in some way. Thus, the cal aperture and partial coherence for a given mask first step will be to define some metric or metrics pattern. The intuitive solution, that the highest of image quality that can be related directly to possible numerical aperture is the best, is true only lithographic quality. Next, the available optical when the image is in perfect focus. For moderate parameters will be varied and the conditions which amounts of defocus, increasing the numerical give the best image (i.e., the maximum value of aperture may actually decreasethe quality of the the image metric) will be found. The three vari- aerial image resulting in degraded stepper perform- ables that will be used to manipulate the image ance. Thus, an optimum numerical aperture and will be the objective lens numerical aperture (NA), partial coherence will be defined for a given the partial coherence of the illumination (0'), and amount of defocus. The use of annular sources of the shape of the illumination source. illumination will be shown to give some improve- ment in high resolution performance. AERIAL IMAGE BEHAVIOR INTRODUCTION In previous work4.5the advantagesand disadvan- tages of various metrics for image quality were A trend is emergingin optical step-and-repeat discussed. It was shown that the slope of the log equipmenttoward greaterflexibility and user of the aerial image (called the log-slope) is directly control. The latestoffering from Ultratech Stepper proportional to the gradient of photoactive com- incorporatesa variable numericalaperture objec- pound and the relative gradient of development tive lens (ranging from 0.2 to 0.4) with variable rate. Through these relationships. the log-slope is partial coherenceof the illumination.1 Scanning the dominant feature of the aerial image which projectionprinters from Perkin-Elmerhave long determines exposure latitude. In general. the value beenable to pick amongmany possible wave- of the log-slope at the mask edge is sufficient to 209 charicterizethe image: Anothermetric which is A plot of nonnalizedlog-slope versus nonnal- appropriatefor spacesand line-spacepairs is the ized linewidth is shownin Figure 1 for equallines ratio of the maximum intensity (usuallythe center and spacesand a partial coherenceof 0.5. There of the space)to the intensity at the maskedge. are threeregions to this curve. Region(a) corre- This ratio is instrumentalin determiningthe spondsto high resolutionfeatures and is thus sidewallangle of the resistprofile. Thus, the two called the high resolutionregion. As expected,the potential metrics are image quality increasesalmost linearly with linewidth. In the low resolutionregion (c), the log-slopelevels off as expectedto a value corre- dIn! and /(edge)I(center) (1) spondingto an isolatededge. In the medium ax resolutionregion (b), however,the behavioris less than obvious. Here,the effects of coherencecause This work will use the log-slopeas the image the image log-slopeto peak at aboutk=0.85. To metric. To begin, the behaviorof this metric will emphasizethe role of coherencein this behavior, be investigatedas a function of severalimaging Figures 2a) and b) showlog-slope linewidth curves variables. In any physical systemwith more than for coherentand incoherentlight, respectively. two variables,the key to understandingthe behav- The dramaticbehavior of the coherentlight can be ior of the systemlies in choosingan appropriate derivedanalytically, and for equallines and spaces graphical representationof the data. A good takesthe fonn choice of graphscan makecomplicated behavior easyto visualize. There are two graphswhich will iJlnI prove useful for understandingimage behavior: the --N-4 (3) -ax- k log-slopeversus linewidth curve and the log-slope defocuscurve. where N is the numberof diffraction orders making The log-slopelinewidth curve will showthe it throughthe lens. At k=O.5,the:1:1 diffraction effect of increasinglinewidth on the quality of the orderspass through the lens makingN=2. Thus, aerial image. Intuitively, one would expectthat in- the log-slopeat this point has a value 16 ~m-1. creasinglinewidth would improve the image Betweenk=O5 and k=1.5 the log-slopegoes as Ilk. quality, leveling off at somevalue for very large At bI.5, the :1:3diffraction ordersalso make it features. In fact, the behavioris somewhatmore throughthe lens. Thus, N=4 and thereis a dra- complicated. Before showingthe results,the curve matic jump in the value of the log-slopeat this can be madeuniversal by scalingall dimensions point. Likewise, as the linewidth continuesto by NAf).. where A is the wavelength.Thus, the x increase,more diffraction orderspass through the position and the linewidth (LW) becomenormal- lens and thereis a discontinuousjump at each ized to value of k correspondingto a new diffraction order. For incoherentillumination, thesedisconti- -xNA nuities are effectively averagedout, resulting in a x=-x- smoothervariation. k = LWNA (2) The effect of defocuson the imagecan be shown ).. in the log-slopedefocus curve.s This curve shows the degradationof image quality as the imagegoes Note that the nonnalized linewidth k is just the out of focus. Figure 3 showsthree log-slope constant in the Rayleigh resolution equation. defocuscurves corresponding to a given linewidth and three different numericalapertures. 210 ...'- (Note: The curves of Figure 3 as they first ap- optimum NA for these 0.6 micron line-space pairs peared in reference5were incorrect. An error in the is significantly different for the two focus budgets computer program used to generate the curves of interest (curves (c) and (d)). Generating mul- gave incorrect results for high numerical aperture tiple curves like Figure 4 for different linewidths simulations. This error has since been corrected and partial coherences, the optimum optical and the new model has been verified by extensive parameters can be detennined within the range of comparisons with SAMPLE and Doug Goodman's our hypothetical stepper. The results are given in proprietary program IMAGE.) Table I. The importantfeature of Figure 3 is that the It is interestingto note that the optimum NA for curvescross. When decidingon the bestvalue of thesefocus budgetsnever exceeds 0.52 down to NA for this linewidth, one must specifythe focal half-micronfeature sizes. Thereseems little rangefor which the line is to be printed. If a range impetusfor building higher NA steppersunless the of :t2 microns or less OOFis required,the high focus budgetcan be significantly reduced. Also, NA lens is defmitely preferred. However,higher largerfeature sizes require lower NA steppers.For amountsof defocusshow that the lower NA lenses example,a state-of-the-art0.7 micron production may give betterresults. Another way to view this processwith a well controlled:tl micron focus effect is to plot log-slopeversus NA for different budgetwould perform betteron a 0.43 NA stepper amountsof defocus,as shownin Figure4. One than a more costly higher NA machine. Note also can seethat the optimumvalue of NA dependson that only very high resolutionfeatures require the the amountof defocus. partialcoherence to be greaterthan 0.3. Using the threetypes of curvesdescribed above, Annular Source it is possibleto characterizethe effects of variable optical parameterson the image. In general,the The optimum stepper configurations given in optimum set of conditions will dependon the Table I assume a standard circular illumination amountof defocusspecified. This amountcorre- source with a size determined by the partial coher- spondsto the total amountof focus