betaFeedback

WorksPhysics and Astronomy

Measurements, simulations, and models of the point-spread function of electron-beam lithography

Abstract

When a sample is exposed using electron-beam lithography, the electrons scatter deep and far in the substrate, resulting in unwanted deposition of dose at both the nano- and the microscale. This proximity effect can be mitigated by proximity effect correction provided that accurate and validated models of the point-spread function of the electron scattering are available. Most works so far considered a double-Gaussian model of the electron point-spread function, which is very inaccurate for modern electron-beam writers with high acceleration voltages. We present measurements of the process point-spread function for chemically semi-amplified resist on silicon and indium phosphide substrates using a 150 kV electron-beam lithography system. We find that the double-Gaussian model deviates from experiments by up to four orders of magnitude. We propose instead a model comprising the sum of a power-law and a Gaussian, which is in excellent agreement with simulations of the electron scattering obtained by a Monte Carlo method. We apply the power-law plus Gaussian model to quantify the electron scattering and proximity effect correction parameters across material stacks, processing, and voltages from 5 kV to 150 kV. We find that the power-law term remains remarkably constant, whereas the long-range dose contributions and the clearing dose are significantly affected by the substrate and the acceleration voltage. Combined, the results imply that reliance on the double-Gaussian approximation in proximity-effect correction has become a limiting factor for resolution in fabrication at the nanoscale.

Introduction

Electron-beam lithography (EBL) [1, 2] is central to modern nanofabrication as it is the workhorse for both direct-write prototyping in research and for the production of master masks for deep or extreme ultraviolet lithography [1, 3] and nanoimprint lithography [1, 4, 5]. The serial exposure of EBL is slower in comparison, although faster writing speeds are obtained in shaped-beam systems [6], multi-beam systems [7], and low-voltage systems [8]. For high-resolution lithography, the method of choice is single-shot Gaussian-beam EBL using high acceleration voltages [9, 10, 11].

The purpose of EBL is to deposit energy and induce chemical changes in an electron-sensitive resist [12], which, after development and subsequent processing such as etching and deposition, results in a mapping of the input mask onto the physical device. However, this mapping is distorted by three effects: Shot-filling errors, the proximity effect, and process effects, as illustrated in Figure 1.

Figure 1. Impact of shot filling and proximity effects on electron-beam lithography. (a) An input mask containing the intended pattern is rendered as (b) polygonal exposure primitives. In Gaussian-beam lithography, the polygons are exposed as arrays of shots on a discrete grid (orange dots). (c) The fabricated pattern deviates from the input mask due to three effects: Shot-filling errors, proximity effects, and process effects due to development as well as any subsequent etching and deposition.

Shot-filling errors arise when the input mask is fractured into simple polygons, which, in Gaussian-beam EBL, are further fractured into arrays of shots. In principle, shot-filling errors can be eliminated by using sufficiently small shot pitches, but in practice, limitations in most EBL systems prevent using shot pitches below a nanometer. Although some EBL systems offer features such as shaped beams or additional primitives such as circles, these features cannot fully circumvent shot-filling errors. The only robust solution is to ensure that the input mask consists only of rectangles, whose edges are aligned to the coordinate system of the EBL writer and whose side lengths are integer multiples of the shot pitch [13].

The proximity effect is much harder to correct for, as it cannot be removed at the design level, and proximity effect correction (PEC) is a hard inverse numerical problem. The issue of the proximity effect and the development of approaches to PEC have therefore been a central element of EBL research since its conception [14, 15, 16, 17, 18, 19, 20, 21, 22, 23]. The proximity effect depends sensitively on the material stack and exposure parameters, such as the acceleration voltage, VV, because it is governed by the spatial distribution of deposited energy, i.e., the electron point-spread function (PSF) [24, 25, 26, 27, 28, 29, 30, 31]. It can be expanded to a process-specific PSF (PPSF) to include additional effects, such as beam size, resist-development chemistry, and subsequent lithographic steps, e.g., etching [32] or deposition [33, 34]. It is important to stress that even the simplest practical example of EBL includes at least two effects that go beyond the PSF: The finite waist of the electron beam and the development. In practice, this means only the PPSF can be measured.

In an EBL point exposure, an electron beam impinges on a resist layer on a substrate, see Figure 2a. The physics underlying the electron scattering in the resist and the substrate is complex, involving secondary electrons and volume plasmons [35, 36, 37, 38], and the proximity effect is therefore commonly modeled using a stochastic Monte Carlo approach [39, 40, 41, 42, 43, 44] such as the simulations shown in Figure 2b and c. The resist is normally much thinner (a few hundred nanometers) than the scattering length (tens of microns), so it is often a good approximation to consider the PSF as being invariant of depth in the resist, in which case the PSF becomes a function of only the radius, rr, as shown in Figure 2d. We identify three distinct regions as indicated in the figure: II, for very short ranges, the PSF is relatively flat; IIII, for intermediate ranges, the PSF approximately follows a power-law [45, 46]; IIIIII, for long ranges, the PSF declines approximately with a Gaussian profile, or, as substrate density and electron scattering increases, an increasingly exponential-decay profile [40, 47].

Figure 2. The electron point-spread function (PSF) in electron-beam lithography. (a) Illustration in cylindrical coordinates of electrons impinging on a stack consisting of resist on a substrate. (b)–(c) Monte Carlo simulation of electron scattering and the corresponding deposited dose distribution from one million 150 kV electrons impinging on a resist-silicon stack. (d) When considering only the energy deposition in the resist, the PSF is only a function of the radius, and three distinct regions, II–IIIIII, can be identified as discussed in the main text.

Ignoring region IIII and separating dose contributions into incident and backscattering [48], may suggest a double-Gaussian (2G2\mathrm{G}) model which is common in the literature [16, 17, 18, 19, 20, 21, 22, 23, 24], but this model is an oversimplification. Nevertheless, since only ad-hoc methods exist today for short- and intermediate-range PEC, a common approach to PEC is to use automatic dose modulation for long-range effects [49, 50, 51] combined with manual shape correction for short-range effects [13, 52, 53, 54]. Although this relies on tedious trial and error for the short-range PEC, it can eventually be an effective strategy. However, this is not an optimal solution, and previous works have improved on the 2G2\mathrm{G} model by modifying the number of Gaussian or exponential terms [25, 55, 56]. Nevertheless, these improved models have not been able to accurately describe PPSFs across a broad range of experimental conditions. One reason is that the short-range Gaussian represents the incident electron beam [24], rather than the electron scattering. Fundamentally, the 2G model is only accurate in the long-range for lighter substrates and in the short-range when the incident beam is very broad. Improvements in EBL technology have moved toward narrower beams and higher VV, which has gradually made the 2G model a worse approximation of actual PSFs. Fits with multiple additional Gaussians can be accurate and useful for numerical calculations, but such fits inevitably include many parameters that impede assessments of resolution limits and quantitative comparisons between different processes across different hardware and material platforms. More accurate PSF models significantly improve PEC [57] and are thus necessary, especially at high VV. To this end, recent models of the PSF have included power laws [58, 59], but investigations across higher acceleration voltages and, more importantly, systematic comparisons to experiments were so far missing.

Here we present accurate measurements of the PPSF for chemically semi-amplified resist (CSAR) [60] on both silicon and indium phosphide (InP) substrates at 150 kV using single-shot exposures. We find excellent agreement with Monte Carlo simulations of the PSF except for distances below 10 nm, which indicates that the PSF and PPSF are identical except at very short distances. We propose an empirical power-law plus Gaussian (PLG) model of the PSF, and we demonstrate a greatly improved agreement with experiment over 10 orders of magnitude of deposited dose, corresponding to 5 orders of magnitude of distances from 1 nm–100 μ\mum for both silicon and InP substrates. The PLG model is validated to enable quantitative comparisons across exposure voltages 5 – 150 kV, which suggest that reliance on 2G-model-based PEC is a limiting factor in modern high-resolution EBL.

Method: Measuring the point-spread function

Before discussing our experiment, we briefly outline the theory of EBL and the theoretical foundation for extracting the PSF, building in part on Refs. [24, 25]. The purpose of EBL is to map a mask, M(x,y)M(x,y), onto a physical device with the greatest possible fidelity. In Gaussian-beam EBL, the mask is rendered as a shot map

S(x,y,p)=p2∑i,jδ(x−ip)δ(y−jp)M(x,y),(1)S(x,y,p) = p^2 \sum_{i,j} \delta(x-ip)\delta(y-jp)M(x,y), \tag*{(1)}

where p=px=pyp=p_x=p_y is the shot pitch. It is desirable to choose pp much smaller than the smallest critical dimension of the mask to minimize shot-filling errors. Each shot is broadened according to the PSF, Ψ(r)\Psi(r), which is related to the density of energy deposited in the resist, E(r)E(r), as

Ψ(r)=E(r)2π∫E(r) r dr.(2)\Psi(r) = \frac{E(r)}{2\pi\int E(r)\,r\,dr}. \tag*{(2)}

The PSF is a probability density with normalization 2π∫Ψ(r) r dr=12\pi\int\Psi(r)\,r\,dr=1. We refer to Appendix § 11 for additional details on extracting the PSF from simulations of E(r)E(r). In practice, the energy is often replaced by an effective deposited charge dose density

D(r)=k−1E(r),(3)D(r) = k^{-1} E(r), \tag*{(3)}

with kk a charge-to-energy conversion factor that is constant for a given process and exposure condition. With these definitions, the deposited dose distribution is given by

D(x,y)=DinΨ(r)∗S(x,y,p),(4)D(x,y) = D_{\mathrm{in}}\Psi(r) * S(x,y,p), \tag*{(4)}

where ∗* denotes the convolution, and DinD_{\mathrm{in}} is the impinging charge dose density, which is directly related to the impinging shot charge, qin=p2Dinq_{\mathrm{in}} = p^2D_{\mathrm{in}}. Thus, if no PEC is applied, the convolution of S(x,y,p)S(x,y,p) with Ψ(r)\Psi(r) distorts D(x,y)D(x,y). The purpose of PEC is to transform S(x,y,p)S(x,y,p) and/or apply a spatial variation to DinD_{\mathrm{in}}, such that D(x,y)D(x,y) becomes as close to the input mask as possible. After development, the resist is fully cleared where

D(x,y)≥D0,(5)D(x,y) \ge D_0, \tag*{(5)}

with D0D_0 the clearing dose, which is the minimum dose required to clear the center of a polygon whose size is much larger than any length scales in the PSF. An advantage of this definition is that the clearing dose is pattern-independent, i.e., in the center of a sufficiently large polygon, D(x,y)D(x,y) becomes equal to DinD_{\mathrm{in}}. Note that D0D_0 depends on the material stack, acceleration voltage, and development conditions.

The exposed pattern can only be imaged after development, which implies that an experiment probes the PPSF rather than the PSF. It is useful to consider an isolated shot, where the shot map reduces to S(x,y,p)=p2δ(x,y)S(x,y,p) = p^2\delta(x,y) and Eq. (4) reduces to

D(r,p)=p2DinΨ(r)=qinΨ(r),(6)D(r,p) = p^2D_{\mathrm{in}}\Psi(r) = q_{\mathrm{in}}\Psi(r), \tag*{(6)}

Combined with Eq. (5), this ensures that the resist is fully developed out to a radius r0r_0, such that the PPSF can be obtained from a sequence of single-shot exposures with different qinq_{\mathrm{in}} using

Ψ(r0)=D0qin.(7)\Psi(r_0) = \frac{D_0}{q_{\mathrm{in}}}. \tag*{(7)}

Results and discussion

Measuring the clearing dose

Turning to the experiment, we first expose a set of large squares with different uniform doses as illustrated in Figure 3(a). Both the size of the squares, bb, and the distance between them, dd, are chosen to be much larger than the radius characterizing the long-range part of the PSF, β\beta. This ensures that the measured clearing dose is pattern-independent and that the deposited dose in the center is equal to the uniform incident dose. For an acceleration voltage of V=150V = 150 kV and a silicon substrate, then β∼60\beta\sim60 µm, so we choose b=250b = 250 µm. Figure 3(b-Figure 3c) shows bright- and dark-field optical-microscopy (OM) images of the exposed and developed pattern with insets for Din=225D_{\mathrm{in}} = 225 µC/cm2^2. The median of the normalized pixel intensity IpI_p is computed from the central cutouts of Figure 3(d-Figure 3e). This is used to determine the clearing dose, D0=225±5D_0 = 225 \pm5 µC/cm2^2, as the dose with no remaining resist, i.e., we observe no change in pixel intensity when increasing the dose further, based on an empirical condition for the gradient, ∇Ip≤0.05\nabla I_p \le0.05, being fulfilled for Din≥D0D_{\mathrm{in}} \ge D_0, see Figure 3(f).

Figure 3. Measurement of the clearing dose, D0D_0, for 150 kV exposures on a stack comprising 180 nm CSAR on silicon. (a) A mask containing squares with uniform incident doses indicated as the numbers in units of µC/cm². Both the size of the squares, bb, and the distance between them, dd, are chosen to be much larger than β\beta, where β\beta is the radius characterizing the long-range part of the PSF. These conditions ensure that in the center of the squares, the deposited doses are equal to the uniform incident doses. (b) Bright-field optical micrograph of the device after exposure and development, where the remaining resist thickness determines the color. (c) Dark-field optical micrograph of the same region as in (b). The overlapping subfields from the exposure are visible as a mesh for doses close to the clearing dose. (d)–(e), Cutouts of the central 50×50 μm250 \times50\,\mu\mathrm{m}^{2} regions of the exposed squares. (f) Extraction of the clearing dose as the minimum dose where the intensities from (d) and (e) are independent of dose, which is consistent across both the bright- and dark-field data sets.

Measurement of the process point-spread function from point exposures

Having determined D0D_0, we now consider the measurement of the PPSF according to Eq. (7). We therefore expose sequences of points as illustrated schematically in Figure 4(a), which shows two point exposures of charge qin,1<qin,2q_{\mathrm{in},1} < q_{\mathrm{in},2}, separated by a distance dd large enough to consider each exposure isolated, such that development removes resist out to radii r0,1<r0,2r_{0,1} < r_{0,2}. We then use scanning electron microscopy (SEM) and OM images of the developed holes to measure the radii r0r_0 as shown in Figure 4(b-Figure 4d). This method benefits from its conceptual simplicity but the resolution of SEM is insufficient to accurately measure holes resulting from impinging shot charges below qin=6q_{\mathrm{in}} = 6 fC. For high doses, a central pillar becomes cross-linked as can be seen in Figure 4(d) and may even topple over, which is apparent from a careful inspection of Figure 4(c). To extract the radii of the developed holes, we use image analysis as explained in Appendix § 12.

Figure 4. Measurement of the process point-spread function (PPSF). We consider the same parameters as in Figure 3, i.e., 150 kV exposures on a stack comprising 180 nm CSAR on silicon. (a) Illustration of two isolated points exposed with shot charges, qin,1<qin,2q_{\mathrm{in},1} < q_{\mathrm{in},2}, such that the resist is fully developed out to radii of r0,1<r0,2r_{0,1} < r_{0,2}. To avoid the two areas affecting each other through the proximity effect, they are separated by a distance d≫r0,2d \gg r_{0,2}. (b)-(c) Scanning electron microscopy (SEM) and (d) optical microscopy (OM) images of the developed patterns. The exposed holes are fitted to obtain the outline (colored 3/4 circles) and radii. (e) The deposited charge as a function of the measured radius, i.e., the PPSF obtained by varying the deposited charge (colored dots). Two different currents, II, are used to reduce the writing time for very large radii. Except for a few points at shorter radii, the analysis by OM and SEM give very similar results, which in turn agree very well with the PSF extracted from a Monte Carlo simulation (black dots). The deviations for radii below 50 nm could originate from process effects. By convolving the PSF with Gaussians of different standard deviation σ\sigma, simulated PPSFs that include the beam distribution are obtained (solid lines). (f) The cumulative density functions (CDFs) of the PPSFs emphasize the deviations, as well as the varying contributions to the deposited dose across the PPSFs.

The measured PPSF is shown in Figure 4(e). In order to reduce the writing time for the longer exposures, two different currents, I=0.1I = 0.1 nA and I=10I = 10 nA, are used for the experiment. We observe that the measured PPSF at I=10I = 10 nA is higher than at I=0.1I = 0.1 nA in the range of 5 nm–50 nm. We note that this cannot be explained solely by differences in beam waist resulting from the different apertures used for the two currents, because such broadening should be included as convolutions that would preserve the normalization. Instead, we observe a systematically higher PPSF for I=10I = 10 nA such that its integral is not preserved. This could indicate that the development rate depends weakly on the exposure current, which is not included in the theoretical framework outlined above, and could be attributed to radiation damage or similar effects. This shows that, at least for CSAR resist, one must be careful when switching between currents. In practice, lower currents are used for small features because of intrinsically smaller spot sizes and minimum-dwell-time limits, so we consider the PPSF at I=0.1I = 0.1 nA more representative at short ranges. Additional details on the data analysis is included in Appendix § 13.

Figure 4(e) shows also the PSF obtained from Monte Carlo simulations. We observe excellent agreement except at very short radii, given that the data set extends over so many orders of magnitude. For the same reason, the apparent agreement may be misleading, as a few experimental points actually deviate from the simulation by more than an order of magnitude. In order to investigate whether the finite beam waist is responsible for the deviations between the simulated PSF and the measured PPSF at small radii, we convolve the PSF with Gaussians of different widths as shown in Figure 4(e). While this provides a slightly better agreement with the PPSF, the lack of data points below 4 nm make this comparison inconclusive.

The several orders of magnitude covered by the PPSF hide its nuances, some of which become clearer by also considering its cumulative density function (CDF),

Φ(r)=2π∫0rΨ(r~) r~ dr~,(8)\Phi(r) = 2\pi\int_{0}^{r} \Psi(\tilde{r})\,\tilde{r}\,\mathrm{d}\tilde{r}, \tag*{(8)}

which represents the fraction of the total dose deposited out to a given range. Importantly, simulations or models of the PPSF also need to accurately represent the CDF, which is complicated by the fact that seemingly minor changes in the PPSF, due to its scale, can significantly impact the CDF. The CDF estimated by trapezoidal numerical integration of the data in Figure 4(e) is shown in Figure 4(f). Notably, the integral of the measured PPSF, i.e., the maximum value of the CDF, is max⁡(Φ)=0.88\max(\Phi) = 0.88. Shot noise adds uncertainty to qinq_{\mathrm{in}}, but this noise averages out to less than 1% for all doses measured in our experiment. Nevertheless, multiple other factors could explain the deviation from the ideal value of unity: One is the above-mentioned current-dependency. A second is numerical errors associated with integrating a function over so many orders of magnitude. A third is errors in the measurement of r0r_0, which may be biased by, e.g., sidewall inclinations, line-edge roughness, and SEM artefacts. This is particularly pronounced for very small and very large radii. Finally, hydrodynamic effects may impact the development for the smallest radii as diffusion in and out of the tiny holes is slowed down, or impact the measurement of D0D_0 due to forces at the resist-substrate interface. In the rest of this paper, we normalize the measured CDFs as Φ/max⁡(Φ)\Phi/\max(\Phi) to evenly distribute errors from any of these effects.

Models of the process point-spread function at high acceleration voltages

For a quantitative description of the PPSF, the model most commonly used is the 2G model,

Ψ2G(r)=1π(1+η)[1α2e−r2α2+ηβ2e−r2β2](9)\Psi_{\mathrm{2G}}(r) = \frac{1}{\pi(1+\eta)}\left[\frac{1}{\alpha^2}e^{-\frac{r^2}{\alpha^2}}+\frac{\eta}{\beta^2}e^{-\frac{r^2}{\beta^2}}\right] \tag*{(9)}

where the standard deviations α/2\alpha/\sqrt{2} and β/2\beta/\sqrt{2} respectively represent the dose deposited by the forward-scattered electrons and the electrons back-scattered from the substrate into the resist, with η\eta describing their relative contribution. The advantages of the 2G model include simplicity and numerical speed. It is also possible to extract reasonable estimates of η\eta and β\beta for PEC by fitting the 2G model to the long-range PPSF or CDF while setting α=0\alpha=0. Figure 5 shows the same data as in Figure 4 compared to different models of the PPSF. The 2G model agrees well with the PPSF for large radii but fails at shorter radii, in particular in the intermediate range where it deviates by up to four orders of magnitude.

Figure 5. Comparison of theoretical models of the process point-spread function (PPSF). We consider the same parameters as in Figure 3, i.e., 150 kV exposures on a stack comprising 180 nm CSAR on silicon. (a) The measured PPSF (orange, yellow, and purple dots) is fitted with a power-law plus Gaussian (PLG) model (grey line) with excellent agreement. The commonly employed double-Gaussian (2G) model (black line) deviates from measurements by up to four orders of magnitude. The three regions I-III introduced in Figure 2(d) are well-defined in the PLG model: The radius at half maximum, γ=r1\gamma=r_1, the radius at which the two model terms are equal, r2r_2, and the characteristic radius of the long-range Gaussian term, β=r3\beta=r_3. (b) The corresponding cumulative density functions (CDFs) of the PLG and 2G models coincide for r>r2r>r_2.

On this background, we propose replacing the forward Gaussian with a power law, resulting in a power-law plus Gaussian (PLG),

ΨPLG(r)=1π(1+η)[ν−2(rγ)σ+er2(β/4)2+ηβ2e−r2β2](10)\Psi_{\mathrm{PLG}}(r) = \frac{1}{\pi(1+\eta)}\left[\frac{\nu^{-2}}{\left(\frac{r}{\gamma}\right)^\sigma+e^{\frac{r^2}{(\beta/4)^2}}}+\frac{\eta}{\beta^2}e^{-\frac{r^2}{\beta^2}}\right] \tag*{(10)}

where β/2\beta/\sqrt{2} remains the standard deviation of the long-range scattering, η\eta is now the ratio of long-to short- and intermediate-range scattering (both can be inserted directly into a 2G model), γ\gamma is the radius at half maximum of the forward distribution and marks the beginning of the power-law scaling, σ\sigma is the slope of the power law and ν\nu is the normalization constant of the power-law term. Our model builds on and extends previous efforts to model the PPSF with a power-law [58, 59], including an empirical Gaussian cut-off term in the denominator of the power-law term to ensure both that the long-range Gaussian dominates at r≫βr\gg\beta and to avoid a singularity at r=0 nmr=0\ \mathrm{nm}.

The parameters of the PLG model define the regions II-IIIIII: The short-range limit is r1=γr_1 = \gamma, and the radius at which the two model terms are equal, i.e., the intermediate- to long-range boundary is r2r_2. The regions are indicated in Figure 5 together with the characteristic radius of the long-range distribution, r3=βr_3 = \beta, and a fit of the PLG model to the measured PPSF, which show excellent agreement. Minor systematic deviations of the fit are present in the short-to-intermediate range and beyond β\beta, but such deviations are to be expected, since the PLG model remains empirical and cannot fully capture the underlying physics of electron scattering. In any case, the agreement over many orders of magnitude in both radii and PPSF is remarkable. For comparison, the 2G model in Figure 5 is chosen to have the same values of η\eta and β\beta as the PLG model fit, as well as a value for α\alpha that replicates the short-range behavior. We refer to Appendix § 14 for further details on the fitting procedure.

Importantly, the PLG model also accurately captures η\eta in the CDF. Here the maximum value of the power-law term, Φ=(1+η)−1\Phi= (1 + \eta)^{-1}, clearly distinguishes the short- to intermediate-range dose contribution from that of the long range. This means that we can consider regions II-IIII and IIIIII separately for PEC: When patterns are fractured into polygons of sizes larger than r2r_2, the proximity effect from one polygon to another is solely governed by the long-range Gaussian, and the proximity effect of the polygons onto themselves, i.e., the so-called self-dose, is solely governed by the short-to-intermediate ranges and reduces to (1+η)−1(1 + \eta)^{-1}. This is a too coarse fracturing for high-resolution lithography, but it can be applied away from polygon edges. Conversely, the long-range Gaussian does not contribute to the self-dose for polygons smaller than r2r_2. For practical applications, the PLG model can give an accurate estimate of the η\eta parameter, which is essential to many PEC algorithms, but its main purpose is as a simple model to allow for detailed comparisons between different material stacks, acceleration voltages, and process parameters.

Measurements of the 150 kV process point-spread functions for longer development, aluminum decharging layer and indium phosphide substrate

In order to compare the effects of different material systems and process parameters, we measure the PPSF in three additional experiments, each one modifying one parameter:

  1. Tripling the development time, which should reduce the clearing dose, D0D_0, but retain the same deposited dose distribution, D(r)D(r), and thus the same PSF.

  2. Adding a 20 nm aluminum layer on top of the resist to reduce any effects from deposited charges during exposure.

  3. Exchanging the substrate to InP, which should exhibit increased scattering in the substrate due to its higher density.

The measured PPSFs, their corresponding CDFs, and the fits of the PLG model to these experiments are shown in Figure 6, together with the fit of the PLG model to the experiment shown in Figure 3-5. The fit parameters for each of our experiments are summarized in Table 1.

D0D_0max(Φ\Phi)β\betaη\etaγ\gammaσ\sigmaνa\nu^{a}
μC cm−2\mu\mathrm{C}\,\mathrm{cm}^{-2}μm\mu\mathrm{m}nm\mathrm{nm}nm\mathrm{nm}
Si225(5)0.8859(4)0.67(5)15(2)2.69(7)24.9
Si Lb^{b}175(5)0.8956(3)0.75(4)17(2)2.71(7)27.3
Si Al225(5)0.8460(5)0.68(6)19(3)2.79(10)28.9
InP165(5)0.9323(1)1.16(13)12(2)2.52(9)22.2

Table 1. Measured clearing doses, D0D_0, integrals of the process point spread function (PPSF), max(Φ\Phi), and fit parameters of the power-law plus Gaussian (PLG) model, β\beta, η\eta, γ\gamma, σ\sigma, and ν\nu.

Figure 6. Process point-spread functions (PPSFs) and the corresponding cumulative density functions (CDFs) for 150 kV exposures on 180 nm CSAR with varying material stacks and processing parameters. Measured PPSFs (colored dots) are well reproduced by fits (black lines) with a power-law plus Gaussian (PLG). The clearing dose, D0D_0 extracted from a corresponding dose test (not shown) is indicated for each experiment. The PLG fit to the results presented in Figure 3-Figure 5 is shown for comparison (grey lines). (a)-(b) An increased development time lowers the clearing dose, but affects the PPSF minimally. (c)-(d) A 20 nm aluminum decharging layer on top of the resist during electron-beam exposure results in a negligible change. (e)-(f) An indium phosphide substrate instead of silicon significantly decreases the clearing dose and changes the PPSF: The long-range contribution falls off at shorter radii, but contributes relatively more to the deposited dose as seen from the CDF, which is attributed to increased electron scattering in the denser substrate.

a^{a} ν\nu values required for normalization with γ\gamma and σ\sigma best estimates.

b^{b} Longer development time of 180 s instead of 60 s.

In the experiment with a longer development time, more resist is dissolved, so each hole of a given shot dose, qinq_{\mathrm{in}}, is developed out to a larger radius, r0r_0. However, this is offset by a reduction in the clearing dose, D0D_0, such that the PPSF remains approximately the same, see Figure 6(a)-(b).

Adding the decharging layer results in a slight broadening of the short-range PPSF, as shown in Figure 6(c)-(d). The difference between currents remains. On the one hand, this indicates that this difference does not originate from charging effects and, on the other hand, that the effect of the thin decharging layer is negligible for exposure of thin resists at high acceleration voltages.

The long-range part of the PPSF is significantly different for the InP substrate as shown in Figure 6(e)-(f), whereas the short- to intermediate-range distributions remain largely invariant between substrates. The decrease in β\beta highlights that, without the power-law cutoff term in the PLG model, the power-law term would dominate beyond β\beta and result in a poor fit. Additionally, η\eta increases significantly, which is evident from the CDF, and also leads to a decrease in D0D_0. Essentially, the long-range scattering is stronger but shorter due to the increased electron scattering on the heavier elements in the InP substrate.

The effect of acceleration voltage on the process point-spread function

We further explore the applicability of the PLG model for various acceleration voltages, VV. We simulate the PSFs for 180 nm CSAR on silicon with V=[5V=[5-150]150] kV and convolve them with a Gaussian with a full width at half maximum of 4 nm in order to estimate the PPSFs. The result is shown in Figure 7(a) with the corresponding CDFs shown in Figure 7(b). We observe that the short-range contribution is much less dependent on VV than the long-range part. A fit of the PLG model for 150 kV is included as an example, and we refer to Appendix § 14 for fits at other voltages and for comparisons with the popular resists PMMA and HSQ. The fits capture the PPSFs accurately across all voltages and resists, and we extract the parameters β\beta, η\eta, γ\gamma, σ\sigma, and ν\nu as a function of VV, as shown in Figure 7(c)-(Figure 7g).

Figure 7. The dependence of the process point-spread function (PPSF) on acceleration voltage, VV, in the range V=5V=5-150 kV for 180 nm CSAR on silicon. (a) Estimate of the PPSF for different voltages (colors). These are calculated by convolving the point-spread functions obtained from Monte Carlo simulations with a Gaussian of full width at half maximum 4 nm approximating the finite beam waist. The fit to the power-law plus Gaussian (PLG) model for V=150V=150 kV is also shown (grey line). (b) The cumulative density functions (CDFs) corresponding to the PPSFs. (c)-(g) The fitting parameters of the PLG model as a function of acceleration voltage. (h) The clearing dose calculated from the normalization of each simulation, normalized to D0(V0)D_0(V_0) where V0=150V_0=150 kV, cf. Eq. (11). We observe that β\beta and D0D_0 increase with VV, while the remaining parameters, η\eta, γ\gamma, σ\sigma, and ν\nu remain approximately constant for V>50V>50 V (black dots) and vary significantly for V<50V<50 V (gray dots), which we attribute to three-dimensional effects as discussed in the main text.

We note two distinct regimes, wherein the parameters η\eta, γ\gamma, σ\sigma, and ν\nu remain approximately constant for V>50V > 50 kV but vary significantly for V<50V < 50 kV. Conversely, β\beta follows an approximately quadratic dependence across all acceleration voltages. At the lower voltages, β\beta reduces from a few microns to a few hundred nanometers, such that the PPSFs likely vary significantly within the depth of the resist. Thus, we attribute the behavior at V<50V < 50 kV to the fact that we average out the depth dependence of the PSFs in the output of our simulations. In contrast, when V>50V > 50 kV the scattering length is high compared to the resist thickness and the only variation with VV is the increase in β\beta, which simply shows that the same fraction of long-range dose is distributed across a larger area. The consequence of this is that PEC can be simpler for large VV because the long-range scattering is approximately constant for patterns much smaller than β\beta. However, the larger range also implies that PEC must correct for proximity doses over a much larger area. The remaining PLG model parameters stay constant at high VV and highlight that the short- to intermediate-range proximity effects remain relevant for nanostructure fabrication. This is problematic for PEC based on the 2G model, as it assumes the intermediate range is negligible.

Lastly, it should be noted that increasing VV also reduces the total energy deposited by scattering, Etot(V)=2π∫E(r,V) r drE_{\mathrm{tot}}(V)=2\pi\int E(r,V)\,r\,\mathrm{d}r, which is the normalization constant of the PSF simulations, and inversely related to the clearing dose as

D0(V)D0(V0)=Etot(V0)Etot(V)(11)\frac{D_0(V)}{D_0(V_0)}=\frac{E_{\mathrm{tot}}(V_0)}{E_{\mathrm{tot}}(V)} \tag*{(11)}

with V0V_0 some fixed voltage. Hence, simulations of the electron scattering give estimates of D0(V)D_0(V), as shown in Figure 7(h) with V0=150V_0=150 kV. The nearly linear dependence of D0(V)D_0(V) means that high acceleration voltages enable finer dose control at the cost of speed, which can be necessary in order to achieve increasingly smaller feature sizes.

The effect of the short- to intermediate range point-spread function on lithography

Finally, we consider how our findings impact EBL at different scales. The trend towards narrower beam widths and higher VV results in substantial deviations from the 2G-model at 150 kV for feature sizes on the order of tens- to hundreds of nanometers, i.e., all nanoscale structures. The impact of this deviation for patterns with feature sizes 10 nm-1 μ\mum is visualized in Figure 8, which shows D(x,y)D(x,y) of masks convolved with the 2G or PLG PPSFs from Figure 5. The exposed areas are much smaller than r2r_2, so the dose deposited from long-range scattering is almost entirely outside the visible areas of the plots, giving an approximately constant reduction of the deposited dose. This means that only the short- to intermediate-range of the PPSFs give rise to spatial variations in the deposited dose. In all cases, the areas that would be cleared of resist after development (black lines) do not match the masks (white lines), because the PPSFs distribute dose away from the points of exposure — the fundamental problem that PEC addresses and corrects for.

Impact of the point-spread function on electron-beam lithography at different length scales

Figure 8. Impact of the point-spread function on electron-beam lithography at different length scales. (a-d) Deposited dose density, D(x,y)D(x,y), for increasing feature sizes when the mask of an isolated feature (exposure within white outline) is convoluted with the double-Gaussian (2G) process point-spread function (PPSF) from Figure 5. The threshold at which resist is fully cleared after development, D0=D(x,y)D_0 = D(x,y), is indicated when the incident dose is Din=D0(1+η)D_{\mathrm{in}} = D_0(1+\eta) (full black line) and Din=1.1D0(1+η)D_{\mathrm{in}} = 1.1D_0(1+\eta) (dashed black line), respectively. The factor (1+η)(1+\eta) compensates that the long-range scattering deposits dose almost entirely outside of the visible areas. (e-h) The same mask convoluted with the power-law plus Gaussian (PLG) PPSF from Figure 5. Less dose is deposited in the range tens-of-nanometers, but more in the range hundreds-of-nanometers, leading to a more pronounced proximity effect. As feature sizes increase, both the disparity between PPSFs and distortions due to the proximity effect become less visible. (i-l) Enclosed versions of the mask (exposure from white line to border) convoluted with the same 2G PPSF, (m-p) and PLG PPSF. Both isolated and enclosed features experience similar distortions, but the contrast in D(x,y)D(x,y) is worse for enclosed features, making them particularly difficult to resolve in lithography.

The relative difference between mask and result is larger for smaller patterns, since the length scale of the deviation is set by the PPSF, and it is therefore independent of the feature size and remains on the order of 100 nm or below. This corresponds to the length scale at which the PPSFs change significantly and the deviation between PPSFs for 2G- and PLG-models is largest at 150 kV. The deviation between the mask and the resulting pattern is significantly larger for the PLG model, and, since the PLG model is a much better model of the PPSFs than the 2G model, this shows that the 2G-model significantly underestimates the magnitude of the proximity effect. This, in turn, implies that PEC based on the 2G-model would not only underestimate the proximity effect, but it would also lead to distorted patterns with poor agreement with the mask.

Conclusions

We have measured the PPSF that characterizes the proximity effect in EBL. The PPSF, obtained from exposures of single shots and normalized by the pattern-independent clearing dose, shows excellent agreement with Monte Carlo simulations of the PPSF and CDF, although we observe deviations in the regions limited by metrology or the number of data points.

When applying the common 2G model of the PPSF, it deviates from experiments by more than four orders of magnitude because it disregards the intermediate range. We therefore propose a PLG model that shows excellent agreement with all the experiments and Monte Carlo simulations of the PPSFs, independent of material stacks, development time, and acceleration voltages in the range of 5–150 kV. This indicates that the PLG model can capture the key features of the PPSFs regardless of all microscopic details, while simultaneously offering the advantages of a simple model with interpretable parameters. This allows comparisons between exposures, processes, and material platforms in order to obtain insights into the electron scattering and PEC.

We find that it is only meaningful to assign η\eta to distinctly separate the long-range dose contribution of the PPSF from the short- to intermediate-range contributions when the electron acceleration voltage is high compared to the resist thickness and electron scattering in the substrate. For a stack consisting of 180 nm resist on silicon, this separation is excellent for V≥50V \ge50 V. The boundary is captured by the radius at which the two terms of the PLG model are equal, i.e, when the long-range approximation deviates from the PPSF by a factor of two, which far exceeds the forward-scattering approximation of the 2G model. The complicated behavior of the PPSF at low voltages, which is likely due to three-dimensional effects, calls for further simulations and experiments. The clearing dose scales nearly linearly with voltage, enabling finer dose control at the cost of speed.

In the high-voltage regime, the short- to intermediate-range PPSF is accurately described by a power law and it is largely unaffected by voltage, material stack, and development time. However, η\eta remains constant for a given substrate while β\beta increases with an approximately square dependence, implying that the relative long-range dose contribution remains the same, only distributed across longer distances. This potentially simplifies PEC because the long-range dose is approximately constant for distances ≪β\ll\beta, although larger areas need to be accounted for. However, at the short- to intermediate range, the 2G model is increasingly inaccurate at higher voltages. Thus, the tendency towards the intuitively and mathematically simple 2G model in PEC has become a limiting factor, specifically in high-resolution EBL, where PEC based on accurate PPSFs from numerical simulations, many-Gaussian approximations, or PLG-like models could improve device fabrication.

Acknowledgments

S.S. acknowledges Bingrui Lu for valuable discussions.

Funding

We gratefully acknowledge financial support from the Villum Foundation Young Investigator Program (Grant No. 13170), Innovation Fund Denmark (Grants No. 0175-00022 – NEXUS), the European Research Council (Grant No. 101045396 – SPOTLIGHT), the Horizon Europe Research and Innovation Programme (Grant No. 101098961 – NEUROPIC, and the Danish National Research Foundation (Grant No. DNRF147 – NanoPhoton).

Conflicts of interest

N.B.H., K.S.K., M.A., T.S., and C.A.R. declare no conflict of interests. S.S. declares his partial ownership of Beamfox Technologies ApS.

Author contributions

Nikolaj Balslev Hougs Writing—original draft, Writing—review and editing, Visualization, Methodology, Formal analysis, Data curation, Conceptualization.

Kristian Sofus Knudsen Writing—original draft, Writing—review and editing, Visualization, Methodology, Formal analysis, Data curation, Conceptualization.

Marcus Albrechtsen Writing—original draft, Writing—review and editing, Methodology, Data curation, Conceptualization, Supervision.

Taichi Suhara Methodology, Data curation, Conceptualization.

Christian Anker Rosiek Writing—review and editing, Conceptualization.

Søren Stobbe Writing—review and editing, Methodology, Conceptualization, Project administration, Funding Acquisition, Supervision.

Data availability

Data from the article is freely available at Zenodo, https://zenodo.org/records/15691545.

Fabrication

Samples consisting of 180 nm AllResist (Strausberg, Germany) AR-P 6200 (CSAR) on silicon and InP substrates are exposed using a 150 kV Elionix (Tokyo, Japan) ELS-F150 electron-beam writer. Each sample contains both point and area exposures: The pattern of point exposures is one array of shot charges qinq_{\mathrm{in}} ranging from 1 fC to 5 μC at pitch d=140d = 140 μm and beam current I=10I = 10 nA, and an additional array with qinq_{\mathrm{in}} from 1 fC to 10 pC with d=5d = 5 μm and I=0.1I = 0.1 nA. The pattern of area exposures is an array of squares of side lengths b=250b = 250 μm, receiving a uniform exposure dose DinD_{\mathrm{in}} in the range 180 μC · cm−2^{-2} to 235 μC · cm−2^{-2}.

The samples are developed in single-puddle Zeon Corporation (Tokyo, Japan) ZED-N50 developer for 60 s before the developed radii, r0r_0, are measured using scanning electron microscopy (SEM) and optical microscopy (OM). One sample with silicon substrate is instead developed for 180 s. Another sample with silicon substrate has a 20 nm aluminium decharging layer deposited on top of the resist before exposure. The layer is removed again after exposure, but prior to development of the resist, by development in single-puddle MicroChemicals GmbH (Ulm, Germany) AZ 726 MIF for 60 s.

Obtaining the electron point-spread function from Monte Carlo simulations of electron scattering

For this work we have used a Geant4-based Monte Carlo simulation of electron scattering [44, 61, 62]. In such a simulation, we consider a resist of depth z0z_0 on a substrate as in Figure 1(a) of the main text. When simulating the electron scattering, it is typically the energy density deposited in the resist, E~(r,z)\tilde{E}(r,z), with 0<z<z00 < z < z_0, that is obtained. In general, the electron point-spread function (PSF) is then extracted as

Ψ~(r,z)=E~(r,z)2π∬E~(r,z)r dr dz=E~(r,z)E~tot.(12)\tilde{\Psi}(r,z) = \frac{\tilde{E}(r,z)}{2\pi\iint\tilde{E}(r,z)r\,\mathrm{d}r\,\mathrm{d}z} = \frac{\tilde{E}(r,z)}{\tilde{E}_{\mathrm{tot}}}. \tag*{(12)}

In cases of thin resist, the PSF is considered invariant in the vertical direction and approximated as an area density

Ψ(r)≈∫0z0Ψ~(r,z) dz≈z0Ψ~(r,zres),(13)\Psi(r) \approx\int_{0}^{z_0} \tilde{\Psi}(r,z)\,\mathrm{d}z \approx z_0\tilde{\Psi}(r,z_{\mathrm{res}}), \tag*{(13)}

with zresz_{\mathrm{res}} an arbitrary depth within the resist. This approximation is included in the output of our simulations, except those shown in Figure 2(b) and (c) of the main text. The same approximation can be similarly applied to E~(r,z)→E(r)\tilde{E}(r,z) \to E(r) in order to obtain Eq. (1) of the main text. Note that this changes the units from [Ψ~]=m−3[\tilde{\Psi}] = \mathrm{m}^{-3} and [E~]=eV⋅m−3[\tilde{E}] = \mathrm{eV} \cdot\mathrm{m}^{-3} to [Ψ]=m−2[\Psi] = \mathrm{m}^{-2} and [E]=eV⋅m−2[E] = \mathrm{eV} \cdot\mathrm{m}^{-2}.

The beam distribution can be included at the simulation stage, but we recommend simulating the deposited energy and PSF from electron scattering of a δ\delta-function beam, Ψ~scat(r,z)\tilde{\Psi}_{\mathrm{scat}}(r,z), which an arbitrary beam distribution, Ψbeam(x,y)\Psi_{\mathrm{beam}}(x,y), can be convolved to find the process-specific PSF (PPSF)

Ψ(x,y)=z0Ψ~scat(r,zres)∗Ψbeam(x,y).(14)\Psi(x,y) = z_0\tilde{\Psi}_{\mathrm{scat}}(r,z_{\mathrm{res}}) * \Psi_{\mathrm{beam}}(x,y). \tag*{(14)}

Any further fabrication steps, such as development or etching, can then be accounted for by further convolutions with appropriate functions, in order to simulate an arbitrary PPSF.

Image analysis to extract the radii of developed holes

We estimate the radii of holes in the developed resist using an automatic image-analysis process, seeking consistent estimates of radii for the hundreds of images we analyze. The input data are scanning electron microscopy (SEM) or dark-field optical microscopy (OM) images, such as Figure 4(b)–(d) in the main text. The result is deterministic given the same input image and parameters for the analysis but may be affected by the imaging settings and the slope of the sidewalls. In dark-field OM, topographical variations such as edges or roughness appear bright. In SEM, features appear brighter when the electron beam penetrates material adjacent to an edge, in which case secondary electrons can escape and be detected [63]. Based on this, the inner transition from dark to bright is chosen as the estimate for the developed hole radius, although the true radius may be slightly offset or, in some cases, gradual. In any case, a rigorous definition of how to extract the hole radius is needed. The following is a summarized description of the automatic image-analysis process:

  1. Denoising: Noise in the input image severely hinders the following steps if not removed. Sharp edge features can be preserved while removing high-frequency noise by denoising using anisotropic diffusion [64], as shown in Figure 9(a).

Figure 9. Image-analysis process to determine the radii of developed holes. (a) First, the input image is denoised by anisotropic diffusion (100 iterations, local contrast K=0.05K = 0.05, and conduction coefficient weight λ=0.2\lambda= 0.2) [64], (b) then a Sobel filter which produces a discrete approximation of the magnitude of the gradient is applied. (c) Thresholding of the gradient image and removing select pixel groups allows isolation of the circle of interest, whose center is found using a Circular Hough Transform. (d) Lastly, the pixel intensities of the gradient image, integrated wrt. radius and normalized, gives peaks corresponding to the inner and outer edge of the bright circle in the input image, where the inner peak is estimated as the resist-substrate interface.

  1. Pre-processing: A gradient filter (Sobel) applied to the denoised image results in a gradient image as shown in Figure 9(b). A threshold is applied to binarize the gradient image, then all pixels not connected to the ring-shaped inner edge of the circular hole are removed, as shown in Figure 9(c).

  1. Circle-finding: A Circular Hough Transform, a slow but accurate shape-detection algorithm, is performed on the binary image for various test radii to estimate the circle center and radius.

  1. Refined estimate: The pixel intensities on the gradient image are integrated radially from the estimated center of the circle, giving graphs as in Figure 9(d) that are resilient against minor errors in the initial center estimates. The final radius is estimated as the peak corresponding to the inner edge of the hole.

Additional details on the data sets and analysis

We use a low current of I=0.1I = 0.1 nA for the point exposures shown in Figure 4–Figure 6 except for the largest doses where a higher current of I=10I = 10 nA was chosen to reduce the write time. The two SEM data sets agree well except at very small radii, because the different exposure currents result in physically different hole sizes. Alternatively, fast and easy measurements devoid of charging artefacts can be obtained from OM, which agrees well with the SEM data for large radii but becomes too inaccurate for smaller radii. We therefore discard the data points for I=10I = 10 nA below 50 nm and the OM data below 2 μm2\,\mu\mathrm{m} when calculating the cumulative density functions (CDFs) and fitting models. In Figure 4(f), the OM data point at 7 μm7\,\mu\mathrm{m} is inconsistent with the corresponding SEM data point at 25 μm25\,\mu\mathrm{m} and is likewise discarded. The discrepancy stems from the different contrast mechanisms in the two imaging methods in the case of a hole with a particularly low and rough sidewall slope.

Power-law plus Gaussian model fits for different acceleration voltages and resists

We first note that power laws are generally difficult to fit well [46]. Since the PPSF spans several orders of magnitude from nanometer to micrometer scale, one issue is that simple least-squares fitting is biased towards data points at lower radii, as these have higher values and are more dense, thus producing far larger errors. The issue stemming from the significant range of the vertical axis can be avoided by fitting the CDF, but this is troublesome without an analytical CDF, as is the case with the PLG model. Instead, we fit to the PPSF in log-log space to minimize the relative residuals

R=∑jr^j2Δr^j2[Ψ^j(r^j)−Ψ^PLG(r^j)]2,(15)R = \sum_j \hat{r}_j^2 \Delta\hat{r}_j^2 [\hat{\Psi}_j(\hat{r}_j) - \hat{\Psi}_{\mathrm{PLG}}(\hat{r}_j)]^2, \tag*{(15)}
Δr^j=(r^j+1−r^j−1)/2,(16)\Delta\hat{r}_j = (\hat{r}_{j+1} - \hat{r}_{j-1})/2, \tag*{(16)}

where r^\hat{r} indicates log⁡10(r)\log_{10}(r) and the weights Δr^j\Delta\hat{r}_j approximate areas of rings with radii r^j\hat{r}_j and widths Δr^j\Delta\hat{r}_j, as a log-scale approach to the weights used by Rishton and Kern [25]. However, the power-law term of the PLG model is only normalized by selecting a value for ν\nu, so directly fitting the parameters can give incorrect ν\nu and η\eta. For this reason, we first obtain an initial fit that is close to, but not normalized. Then we scale ν\nu to obtain a normalized but incomplete fit. Lastly, we fit to the CDF to obtain η\eta, by keeping the remaining parameters from the incomplete fit constant.

Figure 10 shows examples of such PLG model fits to simulated PPSFs and corresponding CDFs for acceleration voltages V=5V = 5-150 kV. The fit parameters are included in Table 2, along with the normalization constant of the simulations. Notably, the two model terms are poorly distinguishable at 5-10 kV, making estimates for η\eta in this range less interpretable as the ratio between a distinct long-range contribution and other dose contributions.

Figure 10. Fits to Monte-Carlo simulations of the PPSF with the power-law plus Gaussian (PLG) model. (left) Simulated electron point-spread functions in 180 nm CSAR on silicon with increasing electron acceleration voltage, VV. (right) Corresponding cumulative density functions. The PLG model provides excellent fits in all cases. The vertical gray lines indicate β\beta and the horizontal lines indicate the fraction of dose deposited at short to intermediate range, 1/(1+η)1/(1+\eta). The line thicknesses indicate one standard deviation of the uncertainty in the estimated values. At low VV, there is a less distinct separation between power-law and Gaussian behavior. As VV increases, the model parameter β\beta grows and η\eta stabilizes.

ResistVVEtot/e−E_{\mathrm{tot}}/e^-β\betaη\etaγ\gammaσ\sigmaνa\nu^{\mathrm{a}}
CSAR512410.32(2)0.38(4)1.68(6)2.01(2)5.1
CSAR106170.81(1.0)0.53(6)1.86(9)2.37(4)3.9
CSAR252422.57(2)0.72(4)1.85(5)2.58(2)3.1
CSAR501228.27(9)0.72(5)1.50(5)2.50(2)2.9
CSAR1006627.7(3)0.73(4)1.41(4)2.45(1.3)2.9
CSAR1505157.2(11)0.72(3)1.40(4)2.43(1.2)2.9
PMMA1506057.5(20)0.73(3)1.40(8)2.48(2)2.7
HSQ1502556.7(23)0.81(7)1.4(1.0)2.42(3)3.0

Table 2. Point-spread-function parameters obtained from simulations of electrons impinging on a stack comprising 180 nm resist on silicon at different acceleration voltages, VV; Energy deposited per electron, Etot/e−E_{\mathrm{tot}}/e^-, and fit parameters of the power-law plus Gaussian (PLG) model, β\beta, η\eta, γ\gamma, σ\sigma, and ν\nu.

Two additional popular resists, PMMA and HSQ, are compared to that of CSAR (equivalent to ZEP) in Figure 11 and Table 2, based on simulations of the PPSFs and CDFs at V=150V = 150 kV. Their primary difference with respect to electron scattering is their densities, which depends significantly on the specific chemical composition. The values used here are ρCSAR/ZEP=1.053g/cm3\rho_{\mathrm{CSAR/ZEP}} = 1.053\mathrm{g}/\mathrm{cm}^{3}, ρPMMA=1.18g/cm3\rho_{\mathrm{PMMA}} = 1.18\mathrm{g}/\mathrm{cm}^{3}, and ρHSQ=0.64g/cm3\rho_{\mathrm{HSQ}} = 0.64\mathrm{g}/\mathrm{cm}^{3}. The PPSFs and CDFs only deviate slightly: HSQ is lighter and thus has relatively reduced scattering, both in general and at distances shorter than the resist thickness, which results in a larger η\eta and less energy deposited per electron.

Figure 11. Fits to Monte-Carlo simulations of the PPSF with the power-law plus Gaussian (PLG) model for different resists. Left column: Simulated electron point-spread functions in 180 nm resist on silicon with 150 kV electron acceleration voltage. Right column: The corresponding cumulative density functions. The PLG model provides excellent fits in all cases. The vertical gray lines indicate β\beta and the horizontal lines indicate the fraction of dose deposited at short to intermediate range, 1/(1+η)1/(1+\eta). The line thicknesses indicate one standard deviation of the uncertainty in the estimated values. The three resists were simulated with densities ρCSAR/ZEP=1.053g/cm3\rho_{\mathrm{CSAR/ZEP}}=1.053\mathrm{g/cm^3}, ρPMMA=1.18g/cm3\rho_{\mathrm{PMMA}}=1.18\mathrm{g/cm^3}, and ρHSQ=0.64g/cm3\rho_{\mathrm{HSQ}}=0.64\mathrm{g/cm^3} for a number of incident electrons 1e61\mathrm{e}6, 1e51\mathrm{e}5, and 1e51\mathrm{e}5, respectively. The least dense resist, HSQ, shows less scattering at distances shorter than the resist thickness, resulting in a larger η\eta.

References

References

  1. [1]Smith B, Rice B, Groves T and Resnick D 2014 Nanolithography (Woodhead Publishing)
  2. [2]Bojko R J, Li J, He L, Baehr-Jones T, Hochberg M and Aida Y 2011 Electron Beam Lithography Writing Strategies for Low Loss, High Confinement Silicon Optical Waveguides J. Vac. Sci. Technol. B 29 06F309DOI
  3. [3]Biswas A, Bayer I S, Biris A S, Wang T, Dervishi E and Faupel F 2012 Advances in Top-Down and Bottom-Up Surface Nanofabrication: Techniques, Applications & Future Prospects Adv. Colloid Interface Sci. 170 2–27DOI
  4. [4]Chou S Y, Krauss P R and Renstrom P J 1996 Nanoimprint Lithography J. Vac. Sci. Technol. B 14 4129
  5. [5]Asano M, Yoshikawa R, Hirano T, Abe H, Matsuki K, Tsuda H, Komori M, Ojima T, Yonemitsu H and Kawamoto A 2017 Metrology and Inspection Required for Next Generation Lithography Jpn. J. Appl. Phys. 56 06GA01DOI
  6. [6]Ikeno R, Maruyama S, Mita Y, Ikeda M and Asada K 2016 Electron-Beam Lithography with Character Projection Technique for High-Throughput Exposure with Line-Edge Quality Control J. Micro/Nanolithogr., MEMS, MOEMS 15 031606DOI
  7. [7]Servin I, Pourteau M L, Pradelles J, Essomba P, Lattard L, Brandt P and Wieland M 2017 Progress and Process Improvements for Multiple Electron-Beam Direct Write Jpn. J. Appl. Phys. 56 06GC03DOI
  8. [8]Greve M M and Holst B 2013 Optimization of an Electron Beam Lithography Instrument For Fast, Large Area Writing at 10 kV Acceleration Voltage J. Vac. Sci. Technol. B 31 043202DOI
  9. [9]Mankiewich P M, Jackel L D and Howard R E 1985 Measurements of Electron Range and Scattering in High Voltage E-Beam Lithography J. Vac. Sci. Technol. B 3 174DOI
  10. [10]Jones G A C, Blythe S and H A 1987 Very High Voltage (500 kV) Electron Beam Lithography for Thick Resists and High Resolution J. Vac. Sci. Technol. B 5 120
  11. [11]Tennant D M 2013 Progress and Issues in E-Beam and Other Top Down Nanolithography J. Vac. Sci. Technol. A 31 050813DOI
  12. [12]Gangnaik A S, Georgiev Y M and Holmes J D 2017 New Generation Electron Beam Resists: A Review Chem. Mater. 29 1898–1917DOI
  13. [13]Albrechtsen M, Vosoughi Lahijani B, Christiansen R E, Nguyen V T H, Casses L N, Hansen S E, Stenger N, Sigmund O, Jansen H, Mørk J and Stobbe S 2022 Nanometer-Scale Photon Confinement in Topology-Optimized Dielectric Cavities Nat. Commun. 13 6281DOI
  14. [14]Eisenmann H, Waas T and Hartmann H 1993 PROXECCO—Proximity Effect Correction by Convolution J. Vac. Sci. Technol. B 11 2741DOI
  15. [15]Mack C A 2001 Electron-Beam Lithography Simulation for Mask Making: VI. Comparison of 10- and 50-kV GHOST Proximity Effect Correction Photomask and Next-Generation Lithography Mask Technology VIII vol 4409 (SPIE) pp 194 – 203DOI
  16. [16]Parikh M 1978 Self-Consistent Proximity Effect Correction Technique for Resist Exposure (SPECTRE) J. Vac. Sci. Technol. 15 931–933DOI
  17. [17]Parikh M 1979 Corrections to Proximity Effects in Electron Beam Lithography. I. Theory J. Appl. Phys. 50 4371–4377DOI
  18. [18]P Kern D 1980 A Novel Approach to Proximity Effect Correction Electron and Ion Beam Science and Technology, Ninth International Conference vol 80 pp 326–339
  19. [19]Owen G and Rissman P 1983 Proximity Effect Correction for Electron Beam Lithography by Equalization of Background Dose J. Appl. Phys. 54 3573–3581DOI
  20. [20]Paykovich J M 1986 Proximity Effect Correction Calculations by the Integral Equation Approximate Solution Method J. Vac. Sci. Technol. B 4 159DOI
  21. [21]Abe T and Takigawa T 1989 Proximity Effect Correction for High-Voltage Electron Beam Lithography J. Appl. Phys. 65 4428–4434DOI
  22. [22]Owen G 1990 Methods for Proximity Effect Correction in Electron Lithography J. Vac. Sci. Technol. B 8 1889DOI
  23. [23]Murai F, Yoda H, Okazaki S, Saitou N and Sakitani Y 1992 Fast Proximity Effect Correction Method Using a Pattern Area Density Map J. Vac. Sci. Technol. B 10 3072DOI
  24. [24]Chang T H P 1975 Proximity Effect in Electron-Beam Lithography J. Vac. Sci. Technol. 12 1271–1275
  25. [25]Rishton S A and Kern D P 1987 Point Exposure Distribution Measurements for Proximity Correction in Electron Beam Lithography on a Sub-100 nm Scale J. Vac. Sci. Technol. B 5 135DOI
  26. [26]Stevens L, Jonckheere R, Froyen E, Decoutere S and Lanneer D 1986 Determination of the Proximity Parameters in Electron Beam Lithography Using Doughnut-Structures Microelectron. Eng. 5 141–150DOI
  27. [27]Duan H, Winston D, Yang J K W, Cord B M, Manfrinato V R and Berggren K K 2010 Sub-10-nm Half-Pitch Electron-Beam Lithography by Using Poly(Methyl Methacrylate) As a Negative Resist J. Vac. Sci. Technol. B 28 C6C58–C6C62DOI
  28. [28]Dai Q, Lee S Y, Lee S H, Kim B G and Cho H K 2011 Experiment-Based Estimation of Point Spread Function in Electron-Beam Lithography: Forward-Scattering Part Microelectronic Engineering 88 3054–3061DOI
  29. [29]Manfrinato V R, Cheong L L, Duan H, Winston D, Smith H I and Berggren K K 2011 Sub-5keV electron-beam lithography in hydrogen silsesquioxane resist Microelectron. Eng. 88 3070–3074DOI
  30. [30]Winston D, Ferrera J, Battistella L, Vladár A and Berggren K K 2012 Modeling the point-spread function in helium-ion lithography Scanning 34 121–128DOI
  31. [31]Manfrinato V R 2015 Electron-Beam Lithography Towards the Atomic Scale and Applications to Nano-Optics Ph.D. thesis Massachusetts Institute of Technology
  32. [32]Donnelly V M and Kornblit A 2013 Plasma Etching: Yesterday, Today, and Tomorrow J. Vac. Sci. Technol. A 31 050825DOI
  33. [33]Hess D W and Jensen K F 1989 Chemical Vapor Deposition Adv. Chem. 221 199–263
  34. [34]Ahmad W, Wolfbeis O S, Hahn Y B, Alshareef H N, Torsi L and Salama K N 2018 Deposition of Nanomaterials: A Crucial Step in Bioensor Fabrication Mater. Today Commun. 17 289–321DOI
  35. [35]Kanaya K and Okayama S 1972 Penetration and Energy-Loss Theory of Electrons in Solid Targets J. Phys. D: Appl. Phys. 5 43–58DOI
  36. [36]Messina G, Paoletti A, Santangelo S and Tucciarone A 1992 Electron scattering in microstructure processes Riv. Nuovo Cim. 15 1–57DOI
  37. [37]Manfrinato V R, Wen J H, Jang L, Yang Y, Hobbs R G, Baker B, Su D, Zakharov D, Zaluzec N J, Miller D J, Stach E A and Berggren K K 2014 Determining the Resolution Limits of Electron-Beam Lithography: Direct Measurement of the Point-Spread Function Nano Lett. 14 4406–4412DOI
  38. [38]Randall J N, Owen J H G, Lake J and Fuchs E 1989 Next Generation of Extreme-Resolution Electron Beam Lithography J. Vac. Sci. Technol. B 37 061605
  39. [39]Saitou N 1973 Monte Carlo Simulation for the Energy Dissipation Profiles of 5–20 keV Electrons in Layered Structures Jpn. J. Appl. Phys. 12 941–942DOI
  40. [40]Hawryluk R J, Hawryluk A M and Smith H I 1974 Energy Dissipation in a Thin Polymer Film by Electron Beam Scattering J. Appl. Phys. 45 2551–2566
  41. [41]Adesida I, Everhart T E and Shimizu R 1979 High Resolution Electron-Beam Lithography on Thin Films J. Vac. Sci. Technol. 16 1743–1748DOI
  42. [42]McMillan J A, Johnson S and MacDonald N C 1989 Simulation of Electron Beam Exposure of Submicron Patterns J. Vac. Sci. Technol. B 7 1540–1545DOI
  43. [43]Johnson S and MacDonald N C 1989 A Program for Monte Carlo Simulation of Electron Energy Loss in Nanostructures J. Vac. Sci. Technol. B 7 1513DOI
  44. [44]Agostinelli S, Allison J, Amako K, Apostolakis J, Araujo H, Arce P, Asai M, Axen D, Banerjee S, Barrand G, Behner F, Bellagamba L, Boudreau J, Broglia L, Brunengo A, Burkhardt H, Chauvie S, Chuma J, Chytracek R, Cooperman G, Cosmo G, Degtyarenko P, Dell’Acqua A, Depaola G, Dietrich D, Enami R, Felicello A, Ferguson C, Fesefeldt H, Folger G, Foppiano F, Forti A, Garelli S, Giani S, Giannitrapani R, Gibin D, Gómez Cadenas J, González I, Gracia Abril G, Greeniaus G, Greiner W, Grichine V, Grossheim A, Guatelli S, Gumplinger P, Hamatsu R, Hashimoto K, Hasui H, Heikkinen A, Howard A, Ivanchenko V, Johnson A, Jones F, Kallenbach J, Kanaya N, Kawabata M, Kawabata Y, Kawaguti M, Kelner S, Kent P, Kimura A, Kodama T, Kokolun R, Kossov M, Kurashige H, Lamanna E, Lampén T, Lara V, Lefebure V, Lei F, Liendl M, Lockman W, Longo F, Magni S, Maire M, Medernach E, Minamimoto K, Mora de Freitas P, Morita Y, Murakami K, Nagamatu M, Nartallo R, Nieminen N, Nishimura T, Ohsubo K, Okamura M, O’Neale S, Oohata Y, Paech K, Perl J, Pfeiffer A, Pia M, Ranjard F, Rybin A, Sadilov S, Di Salvo E, Santin G, Sasaki T, Savvas N, Sawada Y, Scherer S, Sei S, Sirotenko V, Smith D, Starkov N, Stoecker H, Sulkimo J, Takahata M, Tanaka S, Tcherniaev E, Safai Tehrani E, Tropeano M, Truscott P, Uno H, Urban L, Urban P, Verderi M, Walkden A, Wander W, Weber H, Wellisch J, Wenaus T, Williams D, Wright D, Yamada T, Yoshida H and Zschiesche D 2003 Geant4—a Simulation Toolkit Nucl. Instr. Meth. Phys. Res. A 506 250–303
  45. [45]Tennant D M, Doran G E, Howard R E and Denker J S 1988 Electron Scattering Distribution in InP at 50 kV J. Vac. Sci. Technol. B 6 426–431DOI
  46. [46]Clauset A, Shalizi C R and Newman M E J 2009 Power-Law Distributions in Empirical Data SIAM Rev. 51 661–703DOI
  47. [47]Boere E, Van Der Drift E, Romijn J and Rousseeuw B 1990 Experimental Study on Proximity Effects in High Voltage E-Beam Lithography Microelectron. Eng. 11 351–354
  48. [48]Saitou N, Munakata C and Honda Y 1972 Change of Apparent Sensitivity of an Electron Resist Due to Backing Materials Jpn. J. Appl. Phys. 11 1061–1062
  49. [49]Yu H, Yu J, Sun F, Li Z and Chen S 2007 Systematic Considerations for the Patterning of Photonic Crystal Devices by Electron Beam Lithography Opt. Commun. 271 241–247
  50. [50]Bickford J R, Lopez G, Belic N and Hofmann U 2014 Hydrogen Silsesquioxane on SOI Proximity and Microloading Effects Correction From a Single 1D Characterization Sample J. Vac. Sci. Technol. B 32 06F511
  51. [51]Florez O, Arregui G, Albrechtson M, Ng R C, Gomis-Bresco J, Stobbe S, Sotomayor-Torres C M and García P D 2022 Engineering Nanoscale Hypersonic Phonon Transport Nat. Nanotechnol. 17 947–951DOI
  52. [52]Lee S, Jacob J C, Chen C, McMillan J A and MacDonald N C 1991 Proximity Effect Correction in Electron-Beam Lithography: A Hierarchical Rule-Based Scheme—PYRAMID J. Vac. Sci. Technol. B 9 3048
  53. [53]Sewell H 1978 Control of Pattern Dimensions in Electron Lithography J. Vac. Sci. Technol. 15 927–930
  54. [54]Eunsung Seo E S and Ohyun Kim O K 2000 Dose and Shape Modification Proximity Effect Correction for Forward-Scattering Range Scale Features in Electron Beam Lithography Jpn. J. Appl. Phys. 39 6827
  55. [55]Wind S J, Rosenfield M G, Pepper G, Molzen W W and Gerber P D 1989 Proximity Correction for Electron Beam Lithography Using a Three-Gaussian Model of the Electron Energy Distribution Journal of Vacuum Science & Technology B: Microelectronics Processing and Phenomena 7 1507–1512
  56. [56]Liu C H, Ng H T, Ng P C W, Tsai K Y, Lin S J and Chen J H 2008 A Novel Curve-Fitting Procedure for Determining Proximity Effect Parameters in Electron Beam Lithography Lithography Asia 2008 vol 7140 pp 367 – 376
  57. [57]Liu C H, Ng P C W, Shen Y T, Chien S W and Tsai K Y 2013 Impacts of Point Spread Function Accuracy on Patterning and Proximity Effect Correction in Low-Voltage Electron-Beam–Direct-Write Lithography J. Vac. Sci. Technol. B 31 021605
  58. [58]Albrechtsen M and Stobbe S 2019 Power-Law Short Range Point-Spread Function In Electron-Beam Lithography 45th International Conference on Micro and Nano Engineering (Rhodes, Greece) p A21
  59. [59]Mao Q, Zhu J, Cheng X and Wang Z 2025 Proximity effect correction in electron beam lithography using a composite function model of electron scattering energy distribution Discov. Nano. 20 84
  60. [60]Schirmer M, Büttner B, Syrowatka F, Schmidt G, Köpnick T and Kaiser C 2013 Chemical semi-amplified positive e-beam resist (csar 62) for highest resolution 29th European Mask and Lithography Conference p 88860D
  61. [61]Allison J, Amako K, Apostolakis J, Araujo H, Arce Dubois P, Asai M, Barrand G, Capra R, Chauvie S, Chytracek R, Cirrone G, Cooperman G, Cosmo G, Cuttone G, Daquino G, Donszelmann M, Dressel M, Folger G, Foppiano F, Generowicz J, Grichine V, Guatelli S, Gumplinger P, Heikkinen A, Hrivnacova I, Howard A, Incerti S, Ivanchenko V, Johnson T, Jones F, Koi T, Kokolun R, Kossov M, Kurashige H, Lara V, Larsson S, Lei F, Link O, Longo F, Maire M, Mantero A, Mascalino B, McLaren I, Mendez Lorenzo P, Minamimoto K, Murakami K, Nieminen P, Pandola L, Parlati S, Peralta L, Perl J, Pfeiffer A, Pia M, Ribon A, Rodrigues P, Russo G, Sadilov S, Santin G, Sasaki T, Smith D, Starkov N, Tcherniaev E, Tome B, Trindade A, Truscott P, Urban L, Verderi M, Walkden A, Wellisch J, Williams J, Wright D and Yoshida H 2006 Geant4 Developments and Applications IEEE Trans. Nucl. Sci. 53 270–278
  62. [62]Allison J, Amako K, Apostolakis J, Arce P, Asai M, Aso T, Bagli E, Bagulya A, Banerjee S, Barrand G, Beck B, Bogdanov A, Brandt D, Brown J, Burkhardt H, Canal P, Cano-Ott D, Chauvie S, Cho K, Cirrone G, Cooperman G, Cortés-Giraldo M, Cosmo G, Cuttone G, Depaola G, Desorgher L, Dong X, Dotti A, Elvira V, Folger G, Francis Z, Galoyan A, Garnier L, Gayer M, Genser K, Grichine V, Guatelli S, Gueye P, Gumplinger P, Howard A, Hrivnácová I, Hwang S, Incerti S, Ivanchenko A, Ivanchenko V, Johnson V, Jones F, Jun S, Kaitaniemi P, Karakatsanis N, Karamitros M, Kelsey M, Kimura A, Koi T, Kurashige H, Lechner A, Lee S, Longo F, Maire M, Mancusi D, Mantero A, Mendoza E, Morgan B, Murakami K, Nikitina T, Pandola L, Paprocki P, Perl J, Petrovic J, Pia M, Pokorski W, Quesada J, Raine M, Reis M, Ribon A, Ristić Fira A, Romano F, Russo G, Santin G, Sasaki T, Sawkey D, Shin J, Strakovsky I, Taborda A, Tanaka S, Tomé B, Toshito T, Tran H, Truscott P, Urban L, Ushizinsky V, Verbeke J, Verderi M, Wendt B, Wenzel H, Wright D, Wright G, Yamashita T, Yarba J and Yoshida H 2016 Recent Developments in Geant4 Nucl. Instr. Meth. Phys. Res. A 835 186–225
  63. [63]Frase C G, Gniesser D and Bosse H 2009 Model-Based SEM for Dimensional Metrology Tasks in Semiconductor and Mask Industry J. Phys. D: Appl. Phys. 42 183001
  64. [64]Perona P and Malik J 1990 Scale-Space and Edge Detection Using Anisotropic Diffusion IEEE Trans. Pattern Anal. Mach. Intell. 12 629–639DOI

Paper details

vlnc:W_k348fyyn
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents