A numerical orthogonal transformation method for reconstructing a wavefront by use of Zernike polynomials in lateral shearing interferometry is proposed. The difference fronts data in two perpendicular directions are fitted to numerical orthonormal polynomials instead of Zernike polynomials, and then the orthonormal coefficients are used to evaluate the Zernike coefficients of the original wavefront by use of a numerical shear matrix. Due to the fact that the dimensions of the shear matrix are finite, the high-order terms of the original wavefront above a certain order have to be neglected. One of advantages of the proposed method is that the impact of the neglected high-order terms on the outcomes of the lower-order terms can be decreased, which leads to a more accurate reconstruction result. Another advantage is that the proposed method can be applied to reconstruct a wavefront on an aperture of arbitrary shape from its difference fronts. Theoretical analysis and numerical simulations shows that the proposed method is correct and its reconstruction error is obviously smaller than that of Rimmer-Wyant method.
© 2012 OSA
In lateral shearing interferometry (LSI), the wavefront is interfered with a laterally shifted part of itself, thus eliminating the need of an extra reference wavefront. In addition, the two interference beams travel almost equivalent paths, which gives LSI an advantage that it is insensitive to mechanical vibration and environmental turbulence . Owing to these advantages described above, LSI has been used extensively in many applications [2,3]. However, the interferograms of LSI is directly related to the difference between the wavefront and its sheared copy rather than the wavefront under test itself. Thus, an inversion problem that reconstructs the original wavefront from two wavefront differences (difference fronts) in two perpendicular shear directions has to be resolved. A variety of approaches to solve this problem have been devised [4–19]. They are mostly categorized as either zonal reconstruction or modal reconstruction. In zonal reconstruction, the wavefront is directly evaluated at specific grid points [4–9]. While in modal reconstruction, the wavefront is expanded into a set of certain basis functions and its corresponding coefficients are evaluated [10–18]. As shown in , if the wavefront under test can be decomposed into a set of basis functions, modal reconstruction is superior to zonal reconstruction because of its better noise propagation properties and its efficient calculations.
For the modal reconstruction, Zernike polynomials are commonly used as basis functions for wavefront expansion [14–17]. It was introduced to analyze the lateral shearing interferograms first by Rimmer and Wyant (1975) . The two difference fronts in perpendicular shear directions as well as the original wavefront are expanded in terms of Zernike polynomials, the Zernike coefficients of difference front and original wavefront are related to each other by a shear matrix whose elements are functions of shear ratio. In fact, the number of terms of Zernike polynomials of a practical wavefront should be infinite in general case. However, the dimensions of the shear matrix cannot be extended to infinity. This contradiction leads to a result that the terms of Zernike polynomials of the original wavefront above a certain order have to be neglected because they cannot be represented by one or more elements of the shear matrix. These neglected terms are referred to as remaining high-order terms, and they have a negative effect on the outcomes of the lower-order terms. As can be seen below, this effect cannot be omitted even if the contributions of the remaining high-order terms to the original wavefront are slight compared with that of the lower-order terms. This problem was analyzed theoretically with matrix formulations by Herrmann , but he did not evaluate the error source that result from this problem. This error source was defined as remaining error and was evaluated theoretically and numerically by Dai , and the results showed that the remaining error can be reduced by use of Karhunen–Loève (K-L) functions instead of Zernike polynomials as basis functions to expand the wavefront. The K-L functions are optimal set of basis functions for atmospheric turbulence modal phase compensation. However, it is not employed in the field of optical testing.
In this paper, we propose a numerical orthogonal transformation method to reconstruct the wavefront from difference fronts based on Zernike polynomials. By using this method, the sensitivity of the outcomes of lower-order terms to the remaining high-order terms can be decreased, and a consequent result is that the remaining error is reduced and then the reconstruction accuracy is improved. This method can be implemented easily from Rimmer-Wyant method, and can be applied to reconstruct a wavefront on an aperture of arbitrary shape from its difference fronts. By theoretical analysis and numerical calculations, it is confirmed that the accuracy of the proposed method is superior to that of Rimmer-Wyant method.
2. Numerical orthogonal transformation
It is supposed that a wavefront under test is represented by the first fringe Zernike polynomials  asEq. (1) the pistonwas omitted because it is of no concern. Assuming that the two lateral shearing interfergrams in two perpendicular shear directions have been obtained within the two overlap regionsandon a series of discrete points, as shown in Figs. 1(a) and 1(b). And the shear ratioin the two shear directions is assumed identical in this paper, thus the number of measurement pointsfor the two difference fronts is the same. The two difference fronts on these discrete points can be evaluated from the two lateral shearing interferograms.
The difference fronts are also expanded by Zernike polynomials. For the moment, we analyze the difference front when the shearing is in thedirection. Then we haveEq. (1) and the expression of Zernike polynomials in Cartesian coordinates into Eq. (2), the relationship between the Zernike coefficients of the difference front and the original wavefront can be derived. It is written in matrix form for simplicity as
However, the Zernike coefficients of the difference fronts are not independent of each other, because Zernike polynomials are orthogonal basis functions over unit circle but not orthogonal over the overlap regionof two circular beams where the difference front lying in. Thus, it can be expected that the use of orthogonal polynomials instead of Zernike polynomials to expand the difference fronts and a new shear matrix to relate the Zernike coefficients of the original wavefront and the orthogonal coefficients of the difference fronts may reduce the reconstruction error. Indeed, as can be seen below, the reconstruction of the original wavefront with orthogonal transformation results in a smaller remaining error and a more accurate reconstruction result.
2.1. Calculation of the numerical orthonormal polynomials
Although some methods have been presented to construct orthogonal polynomials analytically for non-circle pupils in simple shape, e.g. annular, hexagon, ellipse and rectangle [21–26], it is still a challenging work to construct a set of orthogonal polynomials over the overlap region of two inference beams analytically, especially when the wavefront under test is in a more complicated region, e.g. annular region with a spider that emerged in the Extreme Ultraviolet Lithography . In fact, what we needed is a set of polynomials that are orthogonal over the discrete points of the difference front data set withinrather than polynomials that are orthogonal over the full region of.The orthogonal polynomials over the discrete points can be obtained conveniently from numerical calculation by computer regardless of the shape of the region. Since Zernike polynomials form a complete set, the numerical polynomials that are orthonormal over the discrete points withinare available from the linear combination of Zernike polynomials. The Zernike polynomials and the setare related to each other according toEquation (4) can be written in matrix form as24] which can be extended to discrete domain conveniently. Multiply both sides of Eq. (5) by and, respectively, we obtain
Becauseare orthonormal over the data set, so we have, whereis identity matrix. Use this condition to Eq. (6), we get
Equation (9) can be solved foruniquely with, e.g, Cholesky decomposition due to the fact that is symmetric positive definite matrix , and then conversion matrixcan be determined fromThe expressions of Zernike polynomials in Cartesian coordinate are known, and then the matrix can be calculated by substitute the coordinate values of the measurement points into the expressions of Zernike polynomials. Therefore, the numerical orthonormal polynomials can be determined from Eq. (5) with the matrixand the conversion matrix.
2.2 Derivation of the corresponding shear matrix
By expanding the difference front within with the fist numerical orthonormal polynomials, the corresponding orthogonal coefficients are obtained as vector. Since the fact that each term in the setis a linear combination of the Zernike polynomials, the difference front expanded with the orthonormal polynomials is identical to that expanded with corresponding set of Zernike polynomials. Then, we have a matrix formula,Eq. (5) to Eq. (10), the relationship between the orthonormal coefficients and the Zernike coefficientscan be obtained as
Here, denotes the shear matrix that relates the Zernike coefficients of the original wavefront and the orthonormal coefficients of difference front.
2.3 Wavefront reconstructionEquations (19) and (20) are the representation of Rimmer-Wyant method and the proposed method, respectively.
We note that a similar method by using elliptical orthogonal transformation has been presented in Ref. 16. In this method, the difference front indirection withinwas reduced to an elliptical region. And the elliptical Zernike polynomials which are orthogonal over this elliptical region were generated from Zernike polynomials by coordinate transformation. Then the reduced difference front was expanded into the obtained elliptical Zernike polynomials. The shear matrix that relates the elliptical Zernike coefficients of difference front and the Zernike coefficients of the original wavefront were derived from a double integral. The difference front indirection withinwas analyzed following the same way.
However, the obtained elliptical Zernike polynomials are orthogonal over the full elliptical region, but not orthogonal over the discrete points at which the difference front are measured. Moreover, the applications of this method are limited to the situation that the wavefront under test is circular because the overlap region of the two interference beams cannot be approximated by an ellipse when the wavefront under test is generated from a non-circle aperture such as annular aperture.
For the proposed method, the numerical orthogonal transformation is implemented directly on the discrete points of the difference front data set, and the obtained numerical polynomials are orthogonal over these discrete points rather than the full overlap region. In addition, the orthogonality of the obtained numerical polynomials is not influenced by the shape of the region. Thus, the proposed method can be applied on apertures of arbitrary shape.
The fact that the outcomes of the lower-order terms are influenced by the remaining high-order terms was pointed out and illustrated by an example shown in Ref.16. However, the authors did not explain the effect of orthogonal transformation on the reduction of this influence. This effect will be made clear with a cross-coupling formula derived in the following section.
3. Impact of remaining high-order terms
Here, we assume the number J stands for infinity. Therefore, any practical wavefront can be represented completely by Eq. (1). Now, returning to Eq. (10), under the assumption that J stands for infinity, the column number of the matrixis extended to infinity. If the matrixand the vectorare splited into two blocks, a formula can be obtained from Eq. (10) as
Here, andare the two blocks containing the firstand the remaining columns of the matrixrespectively. andare the firstand the remaining elements of the vector.
3.1 Least-square fitting of difference fronts
Since the difference front cannot be fitted with infinite terms of Zernike polynomials, we assume it is fitted by the firstterms, and then the corresponding coefficients can be obtained by means of least-square fitting,Eq. (21) into Eq. (22), we obtain
3.2 Splitting of the shear matrixes
To facilitate the derivation, Eq. (3) is written in another form as
Here, the shear matrixwith infinite rows and columns is splited into four blocks by a horizontal line under therow and a vertical line behind thecolumns. So, the block is matrix, and the dimensions of other blocks can also be deduced easily. In addition, is becauseis an upper triangular matrix. Vectorsandare arrays of the firstand the remaining elements of vector, respectively.
The shear matrixcan also be splited into four blocks in the same way, and a similar expression as Eq. (24a) for the analysis of the difference front inshear direction can be obtained as
3.3 Reconstruction with finite-dimensional shear matrixes
As discussed in Sec.1, the wavefront cannot be reconstructed completely due to the fact that the dimensions of the shear matrix are finite. As Eq. (19), the estimation of the firstZernike coefficients of the original wavefront except piston can be obtained by shear matrix, and the estimation coefficientsand asEqs. (23a) and (23b) into Eq. (26), we obtain
is referred to as cross-coupling matrix, and it represents the impact of the remaining high-order terms on the outcomes of the lower-order terms when the wavefront is reconstructed by use of Eq. (19). Note that this cross-coupling matrix is different from the cross-coupling matrix derived by Herrmann  and the cross-talk matrix derived by Dai .
3.4 Same analysis of numerical orthogonal transformation method
Although Eq. (28) is deduced from the situation that the Zernike polynomials are used as a basis to expand the difference fronts, it can also be used in the case that the numerical orthonormal polynomials are used as a basis to expand the difference fronts. To analyze the proposed method, by imitating Eq. (28) we obtain
Similar toand, we haveand. Note that, due to the orthonormality ofand, we have
Substituting Eqs. (30). (a) and (b) to the expression of, the result of is obtained, and then substituting this result to the expression of, we obtain.
Obviously, bothandare zero matrixes when, under this condition we have , that is the wavefront can be reconstructed without error regardless of the orthogonality of the basis functions that are used to expand the difference front. However, as discussed in Sec.1, this condition cannot be met in practical situation. Consequently, it is inevitable that the estimations of lower-order coefficients are affected by the high-order coefficients as long as.This problem is inherent deficiency of the modal methods which cannot be prevented by the proposed method. However, it can be alleviated by the proposed method, because the level of the cross coupling of is far below than that ofas demonstrated below.
4. Numerical simulation
4.1 Simulation condition
To confirm the proposed method and the theory analysis discussed above, numerical simulations were implemented. A digitized wavefront filtered by a circle pupil was generated over a square grid by the first 20 fringe Zernike polynomials, that was .The corresponding coefficients were generated randomly, but the last five elements were multiplied by.This attenuation followed the assumption that all the terms that gave significant contributions to the wavefront were contained in the first 15 terms, and the contributions of the last five terms were very weak compared with those of the first 15 terms. Noise-free difference fronts data in two perpendicular shear directions was calculated with the same shear ratio of. The test wavefront and the difference fronts in two directions are shown in Figs. 2(a) to 2(c). The Zernike coefficients of the test wavefront are shown in Fig. 2(d). For comparison, the wavefront were reconstructed from the two difference fronts by Rimmer-Wyant method as Eq. (19) and by the proposed method as Eq. (20), respectively.
4.2 Reconstruction without remaining high-order terms
First, we consider the reconstruction without the remaining high-order terms, that is, with . In this case, all of the Zernike coefficients of the original wavefront are represented by one or more elements of the shear matrixand. The dimensions of both the two shear matrixesandwere.The results of the evaluation of the Zernike coefficients of the original wavefront are shown in Table 1 . When, the evaluated coefficients of both Rimmer-Wyant method and the proposed method are identical to the input coefficients. These results make it clear that both the two methods are capable to reconstruct the wavefront without error under the condition that all the terms of original wavefront are included in the analysis, that is there is no remaining high-order terms. Unfortunately, this condition can only be met in simulation because the existence of remaining high-order terms is inevitable in the analysis of a practical wavefront in general case, as discussed in Sec. 1.
4.3 Reconstruction under the impact of remaining high-order terms
To examine the impact of the remaining high-order terms on the outcomes of the lower-order terms, the wavefront was reconstructed under the condition that, that is, the last five terms were the remaining high-order terms. The dimensions of both the two shear matrixes andwere shrunk to. The results of the evaluation of the Zernike coefficients of the original wavefront are also shown in Table 1. The value of “Percentage Error” column was calculated by the formula of, where is the input coefficient, and is the corresponding evaluated value by the two methods. The calculation results of are also diagramed in Fig. 3(d) for visualization. It can be clearly seen from the “Percentage Error” column of Table.1 that all of the absolute values of the coefficients error that were retrieved by the proposed method are smaller than those by Rimmer-Wyant method. And the proposed method retrieved the input coefficients with small differences of less than in most case, while these differences are higher thanin some case by Rimmer-Wyant method.
To facilitate the analysis, the wavefront under test is divided into two parts. One part is denoted by which represents the contributions of the first 15 terms, while the other part denoted bydescribes the contributions of the remaining 5 terms. In this simulation, we evaluate the reconstruction accuracy ofThe wavefront was reconstructed by use of the evaluated Zernike coefficients shown in “K=15”column of Table 1. The notations of and were assigned to represent the reconstruction results of Rimmer-Wyant method and the proposed method, respectively. The contour plots of these wavefronts and the reconstruction errors of the two methods are shown in Fig. 3. The reconstruction error of the proposed method is obviously smaller than that of Rimmer-Wyant method, as can be clearly seen from Figs. 3(e) and 3(f).
4.4 Comparison of RMS and PV value
The root mean square (RMS) and peak-to-valley (PV) value were used to characterize the reconstruction accuracy. The RMS and PV value of the original wavefront and the reconstruction error of the two methods are shown in Table 2 . The RMS reconstruction error isfor Rimmer-Wyant method and for the proposed method. In other words, the remaining error of the proposed method is aboutof that of Rimmer-Wyant method. Moreover, we know from Table 2 that the PV reconstruction error of the proposed method is about of that of Rimmer-Wyant method. In brief, the reconstruction accuracy of the proposed method is superior to that of Rimmer-Wyant method, which can also be confirmed from Figs. 3(a) to 3(c). Note that the reconstruction accuracy will change with some parameters, such as the number of the Zernike polynomialsused in the reconstruction and the shear ratio, but the fact that the proposed method is more accurate will not change.
4.5 Evaluation of the cross-coupling matrix
To explain the reason why the reconstruction accuracy of the proposed method is superior to that of Rimmer-Wyant method under the impact of remaining high-order terms, the cross-coupling matrixes andof the two methods were evaluated. The two cross-coupling matrixes and are shown in Figs. 4(a) and 4(b). As discussed in Sec.1, the cross-coupling matrix manifests the impact of the remaining high-order terms on the outcomes of the lower-order terms. For example, as shown in Fig. 4(a), the high-order coefficient has affections on the estimation of all the lower-order coefficients, especiallyand, when Rimmer-Wyant method is used. When the proposed method is used, the coefficientjust affects onandwith more slight level, as shown in Fig. 4(b).
On the other hand, the cross-coupling matrix also manifests the sensitivity of the outcomes of the lower-order terms to the remaining high-order terms. For example, as shown in Fig. 4(a), the evaluation of the coefficientis sensitive to almost all of the five remaining terms, especially toand, when Rimmer-Wyant method is used. When the proposed method is used, the evaluation ofis not sensitive to any one of the five remaining terms, as can be seen from Fig. 4(b). Thus the calculation error ofof the proposed method is far below than that of Rimmer-Wyant method which can also be confirmed from Table 1.
Anyway, it can be clearly seen from Fig. 4 that the level of cross coupling ofis far below than that of .Therefore the sensitivity of the outcomes of lower-order terms to the remaining high-order terms can be decreased by the proposed numerical orthogonal transformation method which leads to a smaller remaining error and a more accurate reconstruction result than Rimmer-Wyant method.
4.6. Comparison of the computation time
To compare the computation time of the proposed method and Rimmer-Wyant method, several simulations were implemented in different sample sizes. The computation time of the proposed method was divided into the following two parts: one part is the time to do the numerical orthogonal transformation and the other part is the time to reconstruct the wavefront with the new shear matrixes and numerical orthogonal polynomials. The total reconstruction time is the summation of the two parts. It can be clearly seen from Table 3 that the proposed method is faster than Rimmer-Wyant method. The reason is that the time to obtain the orthogonal coefficients of difference fronts is shorter than the time to obtain the Zernike coefficients of difference fronts as follows: the orthogonal coefficients of the difference front in x direction can be obtained directly by, while the Zernike coefficients are obtained by means of least-squares fitting as which needs more computation time. The time difference is longer than the time to operate the numerical orthogonal transformation. Note that our calculation was performed on a personal computer equipped with a 2.80GHZ Pentium-4 processor and the software was MATLAB (Version 7.8.0).
4.7. Simulation with a general wavefront
Although all of the test wavefronts used above were devised in a special case, the obtained conclusions can be promoted reasonably to general situation. To support this viewpoint, another simulation was implemented. In this simulation, the test wavefront was constructed by use of the first 36 fringe Zernike polynomials and the corresponding coefficients were decided from the experiment results of Ref. 28. The coefficients of tilt and defocus were set to be zero. The difference fronts data in the two directions were calculated with the same shear ratio . The wavefront was reconstructed by use of the first 31 fringe Zernike polynomials and the last 5 terms played the parts of remaining high-order terms. We note that the coefficients of the last 5 terms were comparable to that of the first 31 terms in this case. The RMS reconstruction error was 11.29% for Rimmer-Wyant method and 2.98% for the proposed method.
The number of terms of Zernike polynomials of a practical wavefront is infinite in general case, but the dimensions of the shear matrix which relate the Zernike coefficients of the original wavefront and difference fronts cannot be extended to infinity. As a result, the outcome of the lower-order terms affected by the remaining high-order terms is inevitable when the modal method is used to reconstruct a practical wavefront from its corresponding difference fronts. Nevertheless, this effect can be decreased by use of orthogonal transformation which results in a smaller reconstruction error. This is just the primary idea of the proposed method which indeed leads to a more accurate result than Rimmer-Wyant method. In theory, the prescription described in this paper can be applied to reconstruct a wavefront on an aperture of arbitrary shape from its difference fronts, and it can also be applied when other polynomials are used as basis functions to expand the wavefront under test.
This work was supported by the Grant from the National Natural Science foundation of China under no. 60938003.
References and links
1. D. Malacara, Optical Shop Testing, 3rd ed, (CRC Press, Taylor& Francis, 2007).
3. Y. Zhu, K. Sugisaki, M. Okada, K. Otaki, Z. Liu, J. Kawakami, M. Ishii, J. Saito, K. Murakami, M. Hasegawa, C. Ouchi, S. Kato, T. Hasegawa, A. Suzuki, H. Yokota, and M. Niibe, “Wavefront measurement interferometry at the operational wavelength of extreme-ultraviolet lithography,” Appl. Opt. 46(27), 6783–6792 (2007). [CrossRef] [PubMed]
5. D. L. Fried, “Least-square fitting a wave-front distortion estimate to an array of phase-difference measurements,” J. Opt. Soc. Am. 67(3), 370–375 (1977). [CrossRef]
6. R. H. Hudgin, “Wave-front reconstruction for compensated imaging,” J. Opt. Soc. Am. 67(3), 375–378 (1977). [CrossRef]
7. B. R. Hunt, “Matrix formulation of the reconstruction of phase values from phase differences,” J. Opt. Soc. Am. 69(3), 393–399 (1979). [CrossRef]
8. J. Herrmann, “Least-squares wave front errors with minimum norm,” J. Opt. Soc. Am. 70(1), 28–35 (1980). [CrossRef]
9. X. Liu, Y. Gao, and M. Chang, “A partial differential equation algorithm for wavefront reconstruction in lateral shearing interferometry,” J. Opt. A, Pure Appl. Opt. 11(4), 045702 (2009). [CrossRef]
10. S. Okuda, T. Nomura, K. Kamiya, H. Miyashiro, K. Yoshikawa, and H. Tashiro, “High-precision analysis of a lateral shearing interferogram by use of the integration method and polynomials,” Appl. Opt. 39(28), 5179–5186 (2000). [CrossRef] [PubMed]
12. J. Herrmann, “Cross coupling and aliasing in modal wavefront estimation,” J. Opt. Soc. Am. 71(8), 989–992 (1981). [CrossRef]
13. K. R. Freischlad and C. L. Koliopoulos, “Modal estimation of a wave front from difference measurements using the discrete Fourier transform,” J. Opt. Soc. Am. A 3(11), 1852–1861 (1986). [CrossRef]
14. M. P. Rimmer and J. C. Wyant, “Evaluation of large aberrations using a lateral-shear interferometer having variable shear,” Appl. Opt. 14(1), 142–150 (1975). [PubMed]
15. W. Shen, M. W. Chang, and D. S. Wan, “Zernike polynomial fitting of lateral shearing interferometry,” Opt. Eng. 36(3), 905–913 (1997). [CrossRef]
18. G.- Dai, “Modal wavefront reconstruction with Zernike polynomials and Karhunen-Loève functions,” J. Opt. Soc. Am. A 13(6), 1218–1225 (1996). [CrossRef]
19. W. H. Southwell, “Wave-front estimation from wave-front slope measurements,” J. Opt. Soc. Am. 70(8), 998–1006 (1980). [CrossRef]
20. J. C. Wyant and K. Creath, Basic Wavefront Aberration Theory for Optical Metrology, Vol. XI of Applied Optics and Optical Engineering Series (Academic, 1992), 28.
21. V. N. Mahajan, “Zernike annular polynomials for imaging systems with annular pupils,” J. Opt. Soc. Am. 71(1), 75–85 (1981). [CrossRef]
27. M. Hasegawa, C. Ouchi, T. Hasegawa, S. Kato, A. Ohkubo, A. Suzuki, K. Sugisaki, M. Okada, K. Otaki, K. Murakami, J. Saito, M. Niibe, and M. Takeda, “Recent progress of EUV wavefront metrology in EUVA,” Proc. SPIE 5533, 27–36 (2004). [CrossRef]
28. Y. Zhu, S. Odate, A. Sugaya, K. Otaki, K. Sugisaki, C. Koike, T. Koike, and K. Uchikawa, “Method for designing phase-calculation algorithms for two-dimensional grating phase-shifting interferometry,” Appl. Opt. 50(18), 2815–2822 (2011). [CrossRef] [PubMed]