Numerical modeling of bright-field and dark-field imaging with spatially partially coherent light is considered. The illuminating field is expressed as a superposition of transversely shifted fully coherent elementary fields of identical form. Examples of imaging under variable coherence conditions demonstrate the computational feasibility of the model even when the coherence area of the illumination is in the wavelength scale.
© 2015 Optical Society of America
The bilinear nature of the object-image relationship under spatially partially coherent illumination implies that a direct solution of the imaging problem involves four-dimensional (4D) integrals [1,2], which leads to high computational effort. In non-separable imaging geometries, the problem generally becomes numerically prohibitive when the complexity of a two-dimensional object increases. Over several decades, various computational approaches have been developed to treat this problem [3–7], particularly in the field of projection lithography [8–10]. The computational task can be simplified by means of the coherent-mode decomposition of the cross-spectral density function (CSD) associated with the illuminating field , which reduces the 4D integrals into a sum of 2D integrals. Some initial steps in this direction have been taken [12,13], but all existing methods have some drawbacks as explained in .
We propose and demonstrate an alternative coherent-mode approach in which the incident CSD is represented in terms of spatially shifted and mutually uncorrelated elementary field modes [14–17]. This representation is applicable, for instance, to all imaging problems involving a spatially incoherent primary source and either critical or Köhler illumination of the object. Our numerical frequency-domain implementation involves the evaluation of a finite set of 2D integrals using the Fast Fourier Transform algorithm. We apply the method first to the classical edge-imaging problem and then evaluate partially coherent bright-field and dark-field (BFI and DFI) images of a 2D resolution target. In particular we demonstrate the feasibility of the method when the incident field is nearly incoherent.
In our formulation we assume Köhler illumination of the object and a telecentric imaging system illustrated in Fig. 1, but the method can be readily extended to other cases. As usual, we assume that the imaging takes place within the paraxial aplanatic region of the system.
Let us define the CSD of a stationary field at a plane z = constant as W(ρ1,ρ2) = 〈E* (ρ1)E(ρ2)〉, where ρ = (x,y), E(ρ) is a single field realization, and the brackets denote ensemble averaging. If a thin object defined by a complex-amplitude transmittance t(ρ′) is transilluminated with a field described by , the CSD of the field behind the object is
If the object is imaged by a system with coherent impulse response K(ρ,ρ′), the image-plane CSD takes the form
The necessary and sufficient condition for the realizability of a spatially partially coherent field is that its CSD can be expressed in the (genuine) form 19]. If is of this type, with kernel , Eq. (2) reduces to Eq. (3) with
The calculation of the image-plane spectral density, S(ρ) = W(ρ,ρ), now requires only 2D integrals: first the system response is evaluated for each using Eq. (4) and then Eq. (3) is applied to integrate over all values of . However, this does not immediately imply that the resulting numerical algorithm would be computationally feasible.
Let us next suppose that the incident CSD has the particular mathematical formEquation (5) is valid in both critical and Köhler illumination conditions (see ), and implies that the complex degree of spectral coherence Eq. (5) can also be written as Eq. (8) represents a genuine CSD with and .
Considering an isoplanatic region of an imaging system, such that K(ρ,ρ′) = K (ρ – ρ′), the image-plane CSD is
The spectral density of the field, given byEq. (10), and using Eq. (11) to construct the full partially coherent diffraction image. The expressions given above represent a reformulation of the Hopkins model . For example, Eq. (3) in  may be interpreted as a critical-illumination equivalent of our Eq. (11), although the former does not involve the elementary-field representation explicitly.
The pupil function P(ξ) = |P(ξ)|exp[ik0w(ξ)] accounts, as usual, for truncation and apodization of the field at the exit pupil, and w(ξ) contains the wave aberrations of the system. With this notation the elementary-field response is
We proceed to numerical demonstrations of the technique described above by applying it to both BFI and DFI. We assume an annular incoherent primary source with T(κ) = T0 when R1 ≤ |κ| ≤ R2 and zero otherwise. The CSD is obtained by Eq. (5) which corresponds to the van Cittert–Zernike theorem . Then, according to Eq. (6),Eq. (8), the incident elementary field is now
A circular incoherent source typically used in BFI is obtained by choosing R1 = 0 and dark-field imaging conditions can be modeled by choosing R1 ≥ R, where R defines the pupil of the optical system in normalized coordinates. The pupil function is assumed to be circular, with P(ξ) = 1 when 0 ≤ |ξ| ≤ R. In what follows, we normalize all quantities to R = k0 NA, where k0 is the vacuum wave number and NA is the numerical aperture of the imaging system.
The sampling distances in the two lateral directions are chosen based on the properties of the illumination and the object. To obtain adequate convergence, the sampling distance of the object is taken as r/10, where r = 0.61/R is the resolution limit of the imaging system. The sampling of the incident elementary function e0(ρ′) must be sufficiently dense such that there are several sampling points between |ρ′| = 0 and |ρ′| = σ/R, where σ (the dimensionless number) indicates the first zero of e0(ρ′) that, in view of Eq. (15), can be associated with the transverse coherence length. In our examples the sampling distance of the elementary field was also taken to be r/10, which simplifies the simulation but is not obligatory since one may always apply interpolation. We then scan the incident elementary field with a discrete set of points across the object t(ρ′) to obtain the transmitted elementary fields. Each of these fields is propagated through the system by the FFT-algorithm to find the set of image-plane elementary fields , which are superposed incoherently by the discrete version of Eq. (11). Since the elementary fields are uncorrelated, the algorithm can be easily parallelized.
Because the FFT algorithm is used, the computation time of our method is proportional to M2N2 logN2, where M2 is the total number of elementary fields needed to represent the CSD correctly and N2 is the number of sampling points required to cover the spatial region where the incident elementary field differs significantly from zero. Considering BFI, we roughly have M ∝ 1/σ and N ∝ σ. Hence there is no significant dependence of the computational effort on the degree of spatial coherence. In DFI the effective extent of the elementary field is larger than in BFI (the extreme case being R2 → R1, which represents a thin primary-source ring). In DFI we therefore need to sample the elementary field over an area several times larger than in BFI.
As a first example we consider the problem of bright-field imaging of a straight edge [23, 24] with a circular primary source and pupil, treated semi-analytically in . In Fig. 2, the black line shows the ideal geometrical image of the edge and the other curves illustrate diffraction images under different coherence conditions. The red curve with R2 = 0.1R is practically indistinguishable from the fully coherent case and the green curve (R2 = R) represents the case in which the image of the incoherent primary source matches the pupil.
Tables 1 and 2 illustrate the convergence of our numerical model. Here we consider the value of the image intensity at the geometrical image point of the edge (point Rx = 0 in Fig. 2@@). Both the distance between adjacent elementary fields and the size of the elementary-field window affect convergence. The former is characterized by a parameter Δx and the latter by a parameter L. The results converge when Δx is reduced and L in increased. In Table 1 we show convergence results when the density of the elementary fields is increased and the elementary-field window is chosen large enough (L = 4), while Table 2 demonstrates the convergence when the density of the elementary fields is high (Δx = 0.1) and the elementary-field window is increased. Irrespective of the degree of coherence of illumination, a value Δx = 0.1 is seen to be adequate. Because of the oscillatory nature of the elementary field modes, the results fluctuate somewhat when L is increased, but L = 4 is sufficient in bright-field imaging for most practical purposes.
We proceed to apply our approach to partially coherent imaging of the 2D resolution target illustrated in Fig. 3. The transverse scale of the object is expressed in dimensionless units Rx and Ry. In the BFI case, the white areas of the object have a complex-amplitude transmittance t = 1 and the black areas are opaque (t = 0). In DFI, we assume a low-contrast object with t = 0.99exp(iπ/4) and t = 1 for the white and black regions, respectively.
Figures 4 and 5 illustrate the effect of varying the degree of spatial coherence in the imagery of the resolution target in BFI and DFI conditions, respectively. In BFI the resolution is best with least coherent illumination, and strong fluctuations emerge when the coherence is high. In DFI we see the (expected) edge-enhancement effect particularly well in the largest object features. Table 3 summarizes the simulation parameters together with the required computations times.
The combined effect of spherical aberration and defocus, defined by a wave aberration functionFigs. 6 and 7 for BFI and DFI, respectively. In these examples b = 1, i.e., we assume one wave of pure spherical aberration at the edge of the aperture. The top views represent images in the geometrical focal plane (pure spherical aberration), and the improvement in image quality by balancing spherical aberration with defocusing is clearly visible in the middle and bottom views. The simulation parameters are identical to the corresponding values of R1 and R2 in Table 3. The effect of aberrations on imaging process has been analyzed in  and the simple results for the 1D case that only include paraxial defocus can be found in work done by S. Subramanian .
In conclusion, we have approached the problem of partially coherent imaging by means of of an elementary-field decomposition of the incident field. Here we have assumed an incoherent primary source at the input plane of the condenser. However, spatially partially coherent primary sources can be treated by the elementary-field approach without added difficulty, provided that the source CSD obeys the Schell model [14–17]. In order to bring out the essential concepts in simple terms, we presented the method in the domain of Fourier optics. Extensions to non-paraxial aplanatic systems (that obey the sine law) are straightforward: certain well-known geometrical apodization factors arise, which depend on the conjugate ratio . In the non-paraxial domain the scalar treatment of the elementary illumination field must also be replaced by its electromagnetic counterpart . In our paraxial-domain analysis the results were presented in dimensionless coordinates (xR,yR). However, when features in the object structure are in the wavelength scale, the imagery depends on the wavelength and the transmission-function approach implied by (1) must be replaced by rigorous electromagnetic diffraction analysis. In this domain we would first evaluate the scattering matrix of the object using, e.g., the Fourier Modal Method . Once this is done, the elementary-field response at each center point can be readily evaluated in the plane-wave basis and the rest of the imaging analysis proceeds as described above.
The work was supported by the Academy of Finland ( 252910) and the strategic funding of the University of Eastern Finland.
References and links
1. H. H. Hopkins, “The concept of partial coherence in optics,” Proc. Royal Soc. 208, 263–277 (1951). [CrossRef]
2. J. W. Goodman, Statistical Optics (Wiley, 2000).
4. J. van der Gracht, “Simulation of partially coherent imaging by outer-product expansion,” Appl. Opt. 17, 3725–3731 (1994). [CrossRef]
5. K. Yamazoe, “Computation theory of partially coherent imaging by stacked pupil shift matrix,” J. Opt. Soc. Am. A 25, 3111–3119 (2008). [CrossRef]
7. K. Yamazoe, “Two models for partially coherent imaging,” J. Opt. Soc. Am. A 29, 2591–2597 (2012). [CrossRef]
8. A. E. Rosenbluth, S. J. Bukofsky, M. Hibbs, K. Lai, R. N. Singh, and A. K. K. Wong, “Optimum mask and source patterns to print a given shape,” J. Micro/Nanolith., Microfab., Microsyst. 1, 13–30 (2002).
9. R. L. Gordon and A. E. Rosenbluth, “Lithographic Image Simulation for the 21st Century with 19th-Century Tools,” Proc. SPIE 5182, 73–87 (2003). [CrossRef]
10. R. L. Gordon, “Exact Computation of Scalar, 2D Aerial Imagery,” Proc. SPIE 4692, 517–531 (2002). [CrossRef]
11. E. Wolf, “New theory of partial coherence in the space-frequency domain. Part I: Spectra and cross-spectra of steady-state sources,” J. Opt. Soc. Am. 72, 343–351 (1982). [CrossRef]
13. A. S. Ostrovsky, O. Ramos-Romero, and M. V. Rodríguez-Solís, “Coherent-mode representation of partially coherent imagery,” Opt. Rev. 3, 492–496 (1996). [CrossRef]
14. A. Burvall, A. Smith, and C. Dainty, “Elementary functions: propagation of partially coherent light,” J. Opt. Soc. Am. A 26, 1721–1729 (2009). [CrossRef]
15. F. Gori and C. Palma, “Partially coherent sources which give rise to highly directional laser beams,” Opt. Commun. 27, 185–188 (1978). [CrossRef]
17. J. Tervo, J. Turunen, P. Vahimaa, and F. Wyrowski, “Shifted-elementary-mode representation for partially coherent vectorial fields,” J. Opt. Soc. Am. A 27, 2004–2014 (2010). [CrossRef]
20. M. Born and E. Wolf, Principles of Optics, 7th ed. (Cambridge University, 1999), Sect. 10.6.2. [CrossRef]
22. J. W. Goodman, Fourier Optics, 3rd ed. (Roberts & Company, 2005), Chap. 6.
23. R. E. Kinzly, “Images of coherently illuminated edged objects formed by scanning optical systems,” J. Opt. Soc. Soc. Am. 56, 9–11 (1966). [CrossRef]
24. B. Möller, “Imaging of a straight edge in the partially coherent illumination in the presence of spherical aberration,” Opt. Acta 15, 223–236 (1968). [CrossRef]
25. B. M. Watrasiewicz, “Theoretical calculations of images of straight edges in partially coherent illumination,” Opt. Acta: Int. J. of Opt. 12, 391–400 (1965). [CrossRef]
26. D. S. Goodman and A. E. Rosenbluth, “Condenser aberrations in Köhler illumination,” Proc. SPIE 922, 108–134 (1988). [CrossRef]
28. L. Novotny and B. Hecht, Principles of Nano-Optics (Cambridge University., 2006). [CrossRef]
29. H. Kim, J. Park, and B. Lee, Fourier Modal Method and Its Applications in Computational Nanophotonics (CRC Press, 2012).