Open Access
Add to BibSonomyAbstract
Scatterometry takes advantage of the sensitivity exhibited by optical diffraction from periodic structures, and hence is an efficient technique for lithographic process monitoring. A feature region measurement algorithm has been developed to extract accurately and quickly the relevant constitutive parameters from diffraction data. It is a method for efficiently determining grating structure by seeking the reflectance at some angles contains more information about the structure of the surface relief profile than the reflectance at other angles in a library data match process. The number of measurements and size of signature matching library will be reduced in a great percentage by performing the feature region algorithm.
©2006 Optical Society of America
1. Introduction
A lot of work has been carried out on optical techniques for CD (Critical Dimension) and overlay control to develop best metrology for 65 nm node (and beyond) microelectronics [1–3]. Complex processes and advanced production standards have a strong demand for fast, accurate, and precise inspection technology. The lithographic metrology has observed a substitute from mostly imaging techniques (i.e., CD-SEM, AFM) to non-imaging system, such as scatterometry. Scatterometry is generally considered a manufacturing metrology; however, it can also be used to support process development. It can be used to collect a large amount of profile information quickly enables it to more easily detect some problems that are difficult to discover using other metrology. Most existing scatterometers are designed to measure multiple incident angles at a single wavelength on periodic gratings. Recent year specular spectroscopic scatterometry (SSS) is designed to measure the zeroth-order diffraction response at a fixed angle of incidence and multiple wavelengths [4–6].
Scatterometry is an optical measurement technology based on the analysis of light scattered from a periodic array of features, such as line/space photoresist gratings or arrays of contact holes. It is a model-based metrology that determines measurement results by comparing measured light scatter against a model of theoretical scatter ‘signatures’. A diffracting structure consists of one or more layers that have lateral structure within the illuminated and detected area, resulting in diffraction of the reflected radiation. The structure geometry can significantly affect the reflection, making it possible to make optical measurements of structural features much smaller than the illuminating wavelength. The theoretical model consists of determining the diffraction efficiencies of the given grating using rigorous grating theory. The usual approach for the final part of the inverse problem is to generate a library of scatter signatures ahead of time. Then, when the raw signatures are measured, the library is searched for the best match to the data. However, the library building is time consuming. A common metric used as the best match criterion is the root mean square error (RMSE) which is taking the root of the mean square error of point-by-point (angle) comparison between the theoretical and measured data [3]. The parameters of the modeled signature that has the minimum RMSE is taken to be the parameters of the unknown signature. Conventional algorithms take the average of the difference between the experimental and modeled signatures across the full detection range. Recent study proposed real-time regression algorithm which results are unreliable [7]. Other study addressed the difficulty of inferring the structure geometry from measurement of diffraction due to highly nonlinear relationship between the properties of the diffracting structure and the diffraction measurements [8]. A linearized technique based on the mathematical matrix inversion from the reflectance departure was proposed to permit analysis of the departure of parameter values. However that linear approximation is valid over a limited range of parameter excursions.
In place of the need to generate lots of signature data to performing full library searches or complex mathematically inversion loop, we report a novel method for efficiently and accurately determining grating profiles using feature region matching in a discrepancy-enhanced library generation process. This method selects characteristic portions of the signatures wherever their measurement sensitivity exceeds the preset criteria and reforms a characteristic signature library for quick and accurate matching. By performing an analysis to determine which angles contain the most information, it is possible to reduce the number of measurements used in the inversion process. This results in an algorithm that is faster and more robust. This method does not need to modify existing measurement hardware or the grating target. It saves a great percentage of storage memory in the computer system, and also increases the measurement sensitivity.
2. Scatterometer-based device parameters measurement
2.1 Target details
Figure 1 shows a multilayer grating target. It includes two gratings, which have the same pitch and linewidth, in the top and bottom layer, respectively. An interlayer was fashioned between the top and bottom layer. The materials of the top grating, interlayer, bottom grating, and the substrate are Photo-Resist, Poly Si, SiO2 and Silicon, respectively. The pitch gratings have linewidth 400 nm which is the smallest possible of the aligner used for printing. It is retained its key design characteristics for an angular scatterometer measurement. The thickness and the optical parameters (refractive index, n, and extinction coefficient, k) of each layer were listed in Table 1. The wavelength of the incident beam is 632.8 nm.
Table 1. Grating structure details
2.2 Angular scatterometer details
We studied the effect of measurement configuration based on a 2-θ type angular scatterometer as shown in Fig. 1. The incident beam is scanned through a series of discrete incident angles denoted by θ. By varying the incident angle of incident light source and detection angle simultaneously, a diffraction efficiency of a zero order is obtained. These gratings diffract light only in the zeroth order as their pitch to wavelength ratio tends to smaller than one; all other orders are evanescent. Standard Euclidean distance measure, RMSE, is used for identifying the closest match. The library generation software is based on RCWT (Rigorous Coupled Wave Theory) by building the nominal stack in a graphical user interface and specifying the expected range for each parameter. It uses Maxwell’s equation and the appropriate boundary conditions to calculate the diffracted fields and does not employ any approximation except those resulting from numerical implementation. The accuracy of calculation of diffracted fields depends on the number of layer implemented in the algorithm.
3. Feature region algorithm
3.1 Initial model development
Scatterometry model based on RCWT was used to generate initial test library. It involves using the film stack information (thickness and dispersion of each film) and the grating information (pitch, height and profile characteristics) to create the theoretical diffraction model of the measured structure. This step creates full range (-47° to 47°) test library to study the general characteristics of the scanning signatures. The initial libraries were simplified with assumption of only one parameter was varied and the other two device parameters are constant. Table 2 lists the initial libraries generation details. The scanning signatures of the 1st library representing critical dimension ranging from 390 nm to 410 nm of the 800 nm pitch grating structure in a 1 nm steps are calculated. The scanning signatures of the 2nd library representing sidewall angle ranging from 88.5 to 91.5 degree of the grating structure in a 0.25 degree steps. The scanning signatures of the 3rd representing overlay ranging from 150 nm to 250 nm of the grating structure in a 2 nm steps, the average 1/4 pitch (200 nm) overlay is covered in middle of the test range. The trapezoidal profile is symmetric and the overlay not signed in our modeling work. Each signature represents a unique set of parameter values that describes the CD, sidewall angle and overlay. The RMSE of polarized signatures are calculated to indicate how a change in a particular parameter affects the signatures. Film thickness of each layer was input into the model setup as fixed constant which were commonly measured from specified thin film tool.
Table 2. Initial library generation details
3.2 Single parameter feature region selection
Figure 2 shows theoretical curves for varying parameters of CD, Sidewall (SA) and Overlay (OL) for the case of incident TE polarized light. For the curves in Fig. 2(a), the sidewalls are vertical (slope angle is 0 degree), the overlay between the top and bottom grating is 200 nm. The different curves are for different CD, ranging from 390 nm to 410 nm in increments of 5 nm. This group of curves clearly indicates that sensitivity to CD occurs for angular angles from 25 to about 29 deg and for angles greater than about 40 deg. These are the ranges for which the signatures separate from one another as the CD is changed. We examine the partial derivative of the reflectance with respect to CD as a function of θ for scanning from 0 to 47° as shown in Fig. 2(b). It indicates the varying sensitivity that the measurements possess to the angle of incidence. It has sharpest peak for an angle of incidence of about 16–19 deg and 26–28 deg which shows that if the derivative of reflectance were evaluated at a different location in CD, then the sensitivity will be different.
Figure 2(c) shows theoretical curves for different sidewall angle. The linewidth of the grating is 400 nm, and the overlay between the top and bottom grating is 200 nm. The different curves are for different side wall angle, ranging from 88.5 deg to 91.5 deg in increments of 0.75 deg. This group of curves clearly indicates that sensitivity to side wall angle occurs for angular angles from 13 to about 19 deg, from about 25 to 30 deg, and for angles greater than about 32 deg. These are the ranges for which the signatures separate from one another as the side wall angle is changed. We examine the partial derivative of the reflectance with respect to sidewall angle as a function of θ for scanning from 0 to 47° as shown in Fig. 2(d). It has sharpest peak for an angle of incidence of about 17–19 deg and 26–29 deg, and with increasing sensitivity for angular range greater than 35 deg.
Figure 2(e) shows theoretical curves for different overlay. The linewidth of the grating is 400 nm, and the sidewalls are vertical (slope angle is 0 degree). The different curves are for different overlay, ranging from 150 nm to 250 nm in increments of 20 nm. This group of curves clearly indicates that sensitivity to overlay occurs for angular angles from 0 to about 10 deg, from about 20 to 28 deg, and for angles greater than about 35 deg. These are the ranges for which the signatures separate from one another as the overlay is changed. We examine the partial derivative of the reflectance with respect to overlay as a function of θ for scanning from 0 to 47° as shown in Fig. 2(f). It has sharpest peak for an angle of incidence of about 20–28 deg, and are nearly flat for angular ranges smaller than 7 deg and greater than 32 deg.
Fig. 2. Angular sensitivity curves for several geometrical profile parameters: (a) critical dimension (c) sidewall angle (e) overlay. Angular sensitivity through partial derivatives of reflectance with respect to several profile parameters: (b) critical dimension (d) sidewall angle (f) overlay.
Above simulations show the measurement sensitivity are different at different locations of scanning angles. The range of incidence angles with smaller derivative of reflectance provides little information because the reflectivity measurements are not sensitive to variations of parameter. Conventional algorithms take the average RMSE from full range of scanning angles between the experimental and modeled signatures, which actually neutralizes the greatest sensitivity presenting in part of the measurement range.
3.3 Performance of feature region inversion process
In the current invention only those parts of the signatures that change significantly with the measured parameter are used to calculate the quality of the fit to the experimental data. Those characteristic signature portions form a new library for quicker and more accurate matching.
Figure 3 shows the measurement sensitivity based on RMSE calculation of full scan range (0°–47°) and the feature region 26°–29°. We calculated the RMSE of each adjacent signature with the designed incremental steps which is 1 nm for linewidth, 0.1° in sidewall angle and 10 nm in overlay. By comparing the two curves in Fig. 3(a), we see the differences in the sensitivity to linewidth for the full scan range and feature region which is especially notable around the nominal linewidth 400 nm. It also states that the measurement resolution is much enhanced by implementing the feature region algorithm for library matching. Figure 3(b) shows the difference in the sensitivity to sidewall angle for the full scan range and feature region 26°–29° which is especially notable around the sidewall angle around 89, 90 and 91. Figure 3(c) shows the difference in the sensitivity to overlay for the full scan range and feature region 26°–29°. It states that the measurement resolution is inversely proportional to the same increment in the larger overlay range.
3.4 Multi-parameter feature region selection
In order to establish the priority sequence for feature region selection, we summarized the absolute value of reflectance derivatives of all those three parameters which are shown in Fig. 3(b), (d) and (f) respectively. Figure 4 shows the summarized curve which is used for multiparameter feature region selection, the election sequence is determined as 26°–29°, 40°–47°, 15°–19° etc according to their summarized score. The range of 26°–29° is selected as the first priority feature region for it has more turning points and higher summarized score for the enhanced library matching process. Figure 5 shows the comparison of 3D RMSE map calculated from structure overlay 200 nm and 205 nm of ranging CD (390–410 nm, 1 nm step) and SA (88.5°–91.5°, 0.25° step). It shows the feature matching of 26°–29° has much higher measurement sensitivity in almost all the varying range which is up to 3 times comparing to full range (0°–47°) measurement sensitivity showing in RMSE map. Figure 6 show the matching of 40°–47° feature region also has higher measurement sensitivity across all the varying range.
Fig. 3. RMSE of each adjacent signature of (a) CD (b) SA (c) OL in 1 nm, 0.1° and 10 nm steps respectively.
Fig. 4. feature region summarized curve, the feature region selection sequence is determined as 26°–29°, 40°–47°, 15°–19° etc according to their summarized score.
Fig. 5. feature region(26°–29°) vs. full-range 3D RMSE map calculated from overlay 200 nm and 205 nm of varying CD (390–410 nm, 1 nm step) and SA (88.5°–91.5°, 0.25° step).
Fig. 6. feature region(40°–47°) vs. full-range 3D RMSE map calculated from overlay 200 nm and 205 nm of varying CD (390–410 nm, 1 nm step) and SA (88.5°–91.5°, 0.25° step).
Figure 7 shows the flow chart for generating the characteristic signatures library and multiparameter matching process. Designed grating structure parameters were used to calculate diffraction profiles over a range of incrementally changing parameters. We selected a characteristic region of the calculated diffraction profiles wherein a change in an incrementally increasing parameter causes a change exceeding a threshold value among the diffraction profiles and reform them as a new library. We scan the measurement parameters and select all the possible angles that yield the best sensitivity. The parameters can be freely varied and among possible printing combination. We select the feature regions with priority that all three parameters have generally similar character of greater measurement sensitivity. For example, 26–29° is the first priority region for data matching in this case; and the 40–47° are the second priority region and etc. If the matching results from the first feature region inversion process are more than one solution; then the signatures of those combination parameters are processed into the second priority feature region for further matching. Scatterometry data taken over the feature regions were seriously compared with the reformed new library until the best unique match was found. The combination of device parameters of the modeled signature that agrees most closely with the measured signature is taken to be the parameter of this measured target. The matching library is reduced up to 85% through implementing this feature region matching algorithm if the unique matching solution was found the first feature region matching process. The measurement resolution is increased up to 3 times through the signature discrepancy enhanced process.
4. Summary
We have proposed a new approach to the scatterometer-based measurement of determining surface relief profile information from scatterometric data. We simulated reflecting signature for geometries in which all three parameter values differed from the design geometry values. We found it is capable of determining the correct multiparameter value in relatively few regions matching process. We conducted simulated experiments for geometries in which all three parameter values differed from the designed nominal geometry values. The method proposed here eliminated the need to create a large database which comparison are made and improve the measurement speed. It also improves the measurement sensitivity, and hence the measurement accuracy.
For future work, we will continue experimental work run through this algorithm/comparing the algorithm effectiveness & accuracy with full-angle range approach. We will study whether the performance of our developed algorithm is applicable to variant process configurations. Also, we will see whether the sensitiviness of measurements is also the highest around the feature region angles (e.g. sensitivity to state of polarization, shape of incident beam, noise, surface roughness).
References and links
1. C. H. Ko, Y. S. Ku, N. P. Smith, D. M. Shyu, S. C. Wang, and S. H. Lu, “Comparisons of overlay measurement using conventional bright-field microscope and angular scatterometer,” in Metrology, Inspection, and Process Control for Micro-lithography XIX, R. M. Silver., ed., Proc. SPIE 5752, 987–966 (2005). [CrossRef]
2. C. M Ke, S. S. Yu, Y. H. Wang, Y. J. Chou, J. H. Chen, B. H. Lee, H. Y. Chu, H. T. Lin, T. S. Gau, C. H. Lin, Y. C. Ku, and B. J. Lin, “90 nm Lithography Process Characterization using ODP Scatterometry Technology,” in Metrology, Inspection, and Process Control for Microlithography XVIII, R. M. Silver; ed., Proc. SPIE 5375, 597–604 (2004). [CrossRef]
3. M. Littau, C. J. Raymond, C Gould, and C. Gambill, “Novel implementations of scatterometry for lithography process control,” in Metrology, Inspection, and Process Control for Microlithography XVI, D. J. Herr, ed., Proc. SPIE 4689, 506–516 (2002). [CrossRef]
4. J. A. Kim, S. J. Kim, S. B. Chin, S. H. Oh, D. Goo, S. J. Lee, S. G. Woo, H. K. Cho, W. S. Han, J. T. Moon, C. J. Raymond, M. E. Littau, B. Youn, and C. J. Sohn, “Successful application of angular scatterometry to process control in sub-100-nm DRAM device,” in Metrology, Inspection, and Process Control for Microlithography XVIII, R. M. Silver; ed., Proc. SPIE. 5375, 541–549 (2004). [CrossRef]
5. W. Yang, R. L. Webb, S. Rabello, J. Hu, J. Y. Lin, J. Heaton, M. Dusa, A. de Boef, M. van der Schaar, and A. Hunter, “A novel diffraction based spectroscopic method for overlay metrology,” in Metrology, Inspection, and Process Control for Microlithography XVII, D. J. Herr, ed., Proc. SPIE. 5038200–207 (2003). [CrossRef]
6. H. T. Huang, G. Raghavendra, A. Sezginer, K. Johnson, F. Stanke, M. Zimmerman, C. Cheng, M. Miyagi, and B. Singh, “Scatterometry-Based Overlay Metrology,” in Metrology, Inspection, and Process Control for Microlithography XVII, D. J. Herr, ed., Proc. SPIE 5038126–137 (2003). [CrossRef]
7. J. Opsal, H. Chu, Y. Wen, Y. C. Chang, and G. Li, “Fundamental solution for real-time optical CD metrology,” in Metrology, Inspection, and Process Control for Microlithography XVI, D. J. Herr, ed., Proc. SPIE 4689, 163–176 (2002). [CrossRef]
8. E. M. Drege, J. A. Read, and D. M. Byrne, “Linearized inversion of scatterometric data to obtain surface profile information,” Opt. Eng. 41, 225–236 (2002). [CrossRef]
