Lithographically manufactured nanostructures are shrinking in size in an effort to keep up with Moore’s law. Extreme ultraviolet lithography (EUVL), with an imaging wavelength of 13.5 nm, is the next-generation lithography technology for high-volume manufacturing at the 14-nm half-pitch semiconductor device node and beyond.
The transition to EUVL not only brings challenges in terms of optics and mask design but also difficult metrology challenges. The continuous reduction in the size of semiconductor nanostructures has caused conventional techniques such as scanning electron microscopy and atomic force microscopy to start reaching their resolution limits.1,2 Inline scatterometry techniques have been receiving increasing recognition for their potential to deliver solutions beyond these resolution limits and on industrially feasible timescales, as can be seen by their mention as an emerging disruptive metrology technology in the 2017 publication of the International Roadmap for Devices and Systems (IRDS).3
In particular, critical-dimension small-angle x-ray scattering (CD-SAXS) has emerged as one promising scatterometry technique to extract the profile of line gratings with subnanometer resolution.4 CD-SAXS is a transmission-based scattering measurement where the sample is rotated in order to adjust the incident angle and extract the vertical profile. Sunday et al.5,6 showed that such a measurement could be used to reconstruct the cross-sectional profile (i.e., width, height, and sidewall angle) of the line gratings as a stack of trapezoids. Since then, this approach has also been extended to the study of line-edge roughness.7,8
Moreover, Sunday et al.9 also took advantage of the chemical sensitivity of resonant soft x-ray scattering (RSoXS) measurements to study diblock copolymer (PS-b-PMMA) gratings deposited on the ultrathin substrates with the transmission-based CD-SAXS approach. RSoXS leverages tunable “soft” x-ray (100 to 1500 eV) sources in order to dramatically enhance the scattering cross sections from heterogeneous materials when the x-ray photon energy is adjusted in order to coincide with a given material’s x-ray absorption edges. This combination of absorption spectroscopy and scattering enhances the scattering contrast due to subtle differences in chemistry that are difficult to detect with conventional “hard” SAXS. In recent years, RSoXS performed near the carbon K-edge has proven to be a very useful tool for soft matter researchers interested in defining structure with subnanometer spatial sensitivity by leveraging small differences in chemistry.10
In this paper, we applied the RSoXS technique using energies around the carbon K-edge to enhance the effective electronic density contrast due to chemical differences between the exposed and unexposed regions of a commercially available, chemically amplified EUV resist. Moreover, in contrast to conventional CD-SAXS, which has to be conducted in transmission geometry, this RSoXS measurement was performed in a reflection geometry on a standard Si wafer. This was only possible due to a recent advancement in the ability to confidently reconstruct three-dimensional (3-D) profiles from scattering data acquired in a grazing-incidence geometry (CD-GISAXS).11–13 CD-GISAXS enabled the use of the low-energy, or soft, x-rays needed for the enhancement of scattering contrast due to chemical heterogeneities within the latent image (i.e., before the development step ever took place), while avoiding the need to penetrate a standard Si wafer. This capacity to nondestructively extract a statistically significant cross-sectional representation of the latent image from EUV resists deposited on a conventional Si wafer will provide invaluable feedback to the EUVL community in elucidating the impact of the development step on the final shape generated.
Commercial chemically amplified EUV resists were studied in this paper. Throughout the process, minimal exposure to ambient or fluorescent light was ensured by storing samples in a completely dark sample holder and performing most of the necessary handling in a yellow-light class 100 clean-room environment. The resist was spin-coated onto standard undoped Si wafers in order to yield a film roughly 45-nm thick, as illustrated in Fig. 1(a). Next, a post-application bake was conducted for 60 s at 130°C. The exposure of the resists was performed with a Vistec VB300 Electron Beam Lithography System using a dose of to write a pattern over an area of , as shown in Fig. 1(b). Finally, a post-exposure bake was conducted at 110°C for 60 s.
Total electron yield near-edge x-ray absorption fine structure (TEY NEXAFS) spectra were collected at beamline 6.3.2 at the advanced light source (ALS).14 Carbon K-edge spectra were obtained for the photon energy ranging from 270 to 400 eV. The data were normalized to the incident photon beam intensity measured by a photodiode.
CD-GISAXS experiments were performed at beamline 22.214.171.124 at the ALS.15 A CCD camera was positioned at 50 mm of the sample and the measurement were done at the carbon K-edge, i.e., between 250 and 300 eV. The line grating samples were aligned with a rotation stage until the incoming x-ray beam was parallel to the patterned lines, which was verified when a symmetric semicircle of Bragg spots appeared on the detector, as shown in Fig. 2(a). These Bragg spots will trace out linear features known as Bragg rods when the moment transfer vector is changed by either rotating the sample perpendicular to the substrate (i.e., along the axis) or tilting the sample to different grazing angle of incidence (i.e., around the axis). In our case, the angle of incidence range was set to tilt from 0.1 deg to 20 deg during each exposure in order to record the traces as Bragg rods covering the whole detector. Since the full width at half maximum diameter of the beam was , the footprint of the beam on the sample will vary, respectively, between 70 and . Since the beamstop’s design was not optimized for this experiment and did not cover the whole CCD along the axis, the detector was shifted in order to avoid saturation from the direct and specularly reflected beams. Therefore, as illustrated in Fig. 2(b), only half of the reciprocal space was probed. However, the missing part of the reciprocal space, illustrated in Fig. 2(a), was the symmetric analog of the collected data. The two-dimensional (2-D) image, shown in Fig. 2(b), was recorded with a 10-s acquisition time with the sample rotating at 0.3 revolutions per minute (rpm) during the acquisition.
Results and Discussion
Near-Edge X-Ray Absorption Fine Structure
The first step of the analysis is to determine whether there is any effective electronic density contrast induced in the EUV resist by the exposure step. To do so, we apply the near-edge x-ray absorption fine structure (NEXAFS) technique to record the imaginary part of the refractive index, . Both fields of exposed and unexposed resists are measured at the ALS’s beamline 6.3.2. The extracted is shown in Fig. 3(a).
From the imaginary component of the refractive index, the real component is calculated using the Kramers–Kronig relations.16 Therefore, the full refractive index can be calculated from measuring only the imaginary partFig. 3(a).
The overall features of both and are identical and only small variations can be observed between the two spectra. This leads to the conclusion that the chemical environment of the carbon atoms in the resist is not strongly modified by the EUV light exposure. However, the slight difference in intensity of the peaks observed at 288 and 295 eV, illustrated on the inset of Fig. 3(a), indicates a difference in the density of chemical states associated with that specific energy, which may correspond to the scission, or breaking, of specific carbon bonds in the resist that results from the EUV exposure and post-exposure bake steps.
For the RSoXS, the contrast between two chemical states with electronic density equal to and can be expressed as a sum of the square of the difference of and
From both the delta and beta measured and calculated for the exposed and unexposed resists, we have calculated three different electronic density contrasts and plotted them in Fig. 3(c): between the exposed resist and vacuum (referred as Cev in the rest of the paper) (for which and are equal to 0); between the unexposed resist and vacuum (referred as Cuv); and between the exposed resist and the unexposed resist (referred as Ceu). The similarity between both and of the exposed and unexposed resists leads to a weak electronic density between resists. To better visualize this, a zoomed scale is presented in the inset of Fig. 3(c). More interestingly, for most energies, Ceu is negligible in comparison to Cev and Cuv, and except at specific energies around 280 and 287 eV, the three different electronic density contrasts have the same order of magnitude, making the relative contribution of Ceu to the observed scattering data more significant. Therefore, we have focused on collecting RSoXS measurements for the energy range from 280 to 290 eV.
Resonant Soft X-Ray Scattering
To record the Fourier rods, also known as Bragg rods, coming from the gratings, we scan the incident angle of the sample from 0.1 deg to 20 deg. We then decouple the various spatial contributions to the measured scattering features from the following expression of the vector:Fig. 4 and are displayed in the pixel space, since each pixel will have a different coordinate in the and referential for each incident angle.
The Bragg rods can be observed consistently across all four images, regardless of incident energy, confirming the presence of an in-plane structure inside the film. The presence of Bragg rods is unrelated to the presence of contrast between the exposed and unexposed resists. Indeed, in the hypothesis of any Cue equal to 0, the film of resist can be considered as a uniform film with a given height. In this scenario, the scattering signal will be concentrated at the specular plane, at . Therefore, only modulations of the reflected beam intensity would appear; however, our specific setup did not record this reciprocal space region. Not only does the presence of Bragg rods allow rejecting the assumption of an homogeneous film of resist, but it also reveals the existence of periodic features along the surface of the resist induced by the slight shrinking of the resist under EUV light, as illustrated in Fig. 5. Therefore, the distance between two consecutive Bragg rods can be used to extract the distance between two EUV exposures and therefore the pitch of the distance between two unexposed resist “lines.” A value of 188.47 nm is extracted.
Figure 4 clearly shows that only some incident energies produce significant modulations along the lengths of the Bragg rods (i.e., ), while others do not. To illustrate this phenomenon, vertical one-dimensional cuts are done and are annotated in Fig. 4(a) taken at 283 eV. The first, second, and third Bragg rods are shown in Fig. 4(e).
Unlike the Bragg rods, the presence or absence of the modulations can be connected to the electronic density contrast calculated from the NEXAFS measurements. Our previous study on the CD-GISAXS method demonstrated that these modulations provide in-depth information on line gratings. This is because the intensity of the recorded Bragg rods is modulated by the Fourier transform of the shape of the individual lines.17
Consequently, the modulations at 283 and 286 eV reveal the existence of the electronic density contrast between the exposed and unexposed resists and are in agreement with a film thickness of 45 nm, as illustrated in Fig. 5. Conversely, the absence of modulations can be explained if there is no contrast between the resists. In that specific case, only a subnanometer height grating at the surface of the film is seen by the beam, coming from the shrinkage of the resist.
Therefore, we can only approximate from the intensity along the Bragg rods at 285 and 287 eV that the difference in thickness, , between exposed and unexposed resists is , as illustrated in Fig. 5. Such approximation is due to the fact that the first minimum of the form factor of such thin line should appear at , which is not accessible with the current instrumental setup (limited to ).
The electronic intensity contrast variation between resists observed with RSoXS follows the trend observed in the NEXAFS data. Indeed, the relative evolution of Cue with Cev and Cuv, leads to the same conclusions. Indeed, when and Cuv, annotated with dash lines in Fig. 3(c), at 283 and 286 eV, for example, modulations along the Bragg rods can be observed. This differs starkly from the behavior at 285 and 287 eV, when and Cuv; no modulations along the Bragg rods are observed.
In addition, both height contributions to the form factor are weighted by Cue and Cuv. Therefore, at 283 and 286 eV, both form factors are visible on the Bragg rods, even though the form factor of the shallow gratings did not show any feature in that -range. At 285 and 287 eV, since and Cuv, the form factor of the shallow gratings strongly dominates, even though some very weak modulations can be observed at 287 eV.
In-Depth Profile Reconstruction
To extract a detailed height profile of the latent image, the modulations along the Bragg rods at 283 eV are plotted in Fig. 4(e) and are fitted with an inverse algorithm. The height profile of the latent image can thus be described as a stack of trapezoids. The simulation of the form factor and subsequent reconstruction of the line shapes are done under the distorted-wave Born approximation (DWBA). The DWBA extends the Born approximation to include both reflective and refractive events at the film–substrate interface.18,19
The cross section of the depth profile of the latent image is approximated by stacking five trapezoids. The cross section can be described with parameters, where is the number of trapezoids and the parameters are each trapezoid’s height, (kept identical for all the trapezoids); a sidewall angle per trapezoid (assumed symmetric); and the bottom width of the trapezoid underneath.
To converge to the best cross-sectional profile of the line gratings, we used a genetic algorithm to minimize the difference between the form factor recorded experimentally and the simulated one.11 Based on the study by Hannon et al.,20 we applied a covariance matrix adaptation evolutionary strategy (CMAES).
The results of the CMAES are shown in Fig. 5, corresponding to the fit presented in Fig. 4(e). The best-fit profile comes from a stack of seven trapezoids, with each trapezoid having a height of 6.23 nm, with an overall line-height of 43.60 nm, and a bottom linewidth of 87.20 nm. Each of the sidewall angles, from the bottom to the top are, respectively, equal to 71.8 deg, 81.7 deg, 71.5 deg, 88.9 deg, 92.5 deg, 90.7 deg, and 91.5 deg.
The fitted form factors show a good agreement with the experimental data; they mimic the position of the minima and the periodicity of the modulations. However, the amplitude is not perfectly reproduced. Our current simulation of a monodisperse line profile over the fully exposed area introduces some error in the reproduction of the modulation amplitude seen experimentally. Moreover, the possible depth variation of the electronic density contrast need also to be considered to improve the fit. A more detailed explanation on the simulation of the CD-GISAXS patterns recorded at different incident angles will be reported in a coming study and a statistical distribution of the depth-profile model will be given.
The overall profiles also show the presence of a footing feature (i.e., a wider width at the bottom of the line at the substrate interface). However, no rounding effect at the top of the line can be observed. The next step of the study will be a comparison of the in-depth profile before and after the development of the exposed resist, in order to determine the significance and nature of both of these features.
In this study, we demonstrated the capability of the RSoXS technique to reconstruct the 3-D shape of the latent image in a EUV resist, before the development step took place.
The slight shrinkage of the resist under the EUV light, coupled with the chain scission of some carbon bonds, induced the appearance of a relatively weak electronic density contrast between the exposed and unexposed resists. Nevertheless, with a combination of NEXAFS and RSoXS measurements, we were able to enhance the effective electronic density contrast between the developed and undeveloped resists using energies around the carbon K-edge in order to detect the latent image on an EUV resist.
This study has given a first glimpse into the potential of the RSoXS technique to elucidate the structure of the exposed regions of the resists before development. Further studies using this approach are underway, such as quantifying the impact of the development step on the final profile and the impact of the soft x-rays, in order to provide insight to understand the physical and chemical phenomena underlying EUV lithography.
We acknowledge the support from the LBNL LDRD program, the DOE Early Career Award program, and the Center for Advanced Mathematics in Energy Research Applications (CAMERA), funded jointly through DOE’s ASCR and BES programs. This work used resources of U.S. DOE Office of Science User Facilities, supported under contract Nos DE-SC0012704, DE-AC02-06CH11357, and DE-AC02-76SF00515. Work at the Molecular Foundry was supported by the Office of Science, Office of Basic Energy Sciences, of the U.S. Department of Energy under Contract No. DE-AC02-05CH11231. Terry McAfee was partially supported by NSF Grant No. 1626566.