Recent experiments on three-dimensional direct-laser-writing (DLW)1–5 optical lithography have been inspired by stimulated-emission-depletion (STED) optical microscopy.6–9 Using a novel photoresist composed of the photoinitiator 7-diethylamino-3- thenoylcoumarin in pentaerythritol tetraacrylate combined with tailored foci of light, the diffraction barrier could indeed be broken in the lateral as well as in the axial direction, bringing this form of lithography truly to the nanometer scale.10 However, the underlying depletion mechanism has been ambiguous. In this Communication, we perform lithography experiments with time-delayed excitation and depletion pulses of variable center wavelength. These data reveal a slow and a fast component with distinct spectral signatures. The fast component exhibits a time constant of about 1 ns and spectrally follows the anticipated gain spectrum. It can thus firmly be assigned to stimulated emission. The slow component is distinct from stimulated emission and lies in the range from 10 ns to 1 μs. These overall data allow for systematically optimizing the conditions in next-generation STED-DLW optical lithography.
Direct-laser-writing (DLW) optical lithography can be viewed as the three-dimensional (3D) counterpart of planar (2D) electron-beam lithography.1–5 It routinely allows for the fabrication of nearly arbitrarily complex 3D structures in a single processing step. Typically, photoinitiator molecules in a photoresist are excited via two-photon absorption by a tightly focused near-infrared laser beam, they generate radicals, and initiate a polymerization reaction only within the focal volume. The polymerized volume element (voxel) is the building block for more complex structures that are usually created by scanning either sample or focus. However, with typical minimum lateral (axial) feature sizes of 100 nm (250 nm), DLW is not yet a true nano-technology. Exploiting shrinkage11–12 can reduce feature sizes but is not generally applicable. However, linewidth and resolution in the sense of Abbe (i.e., minimum period of a grating) must not be confused.10 Governed by the diffraction limit, lateral (axial) periods below 210 nm (510 nm) were inaccessible in practice.10
Today, the corresponding limitation in fluorescence microscopy can be overcome in several ways, pioneered by S. W. Hell and his approach called stimulated-emission-depletion (STED) microscopy.6–9 By depleting the first excited singlet state of the molecules in the periphery of the focal spot, STED physically reduces the effective excitation volume, the extent of which was believed to be fundamentally limited. As proposed by Hell in 2000,7 this smaller excitation volume cannot only be used for fluorescence microscopy but, e.g., also to confine photo- chemical reactions to the nanoscale. Obviously, it would be highly desirable to improve the resolution of DLW towards that of electron-beam or deep-UV lithography while fully maintaining its 3D capability.
Recently, first experiments towards 3D optical lithography beyond the diffraction limit have been published.13–15 Translated to lithography the idea of STED means to initiate the polymerization reaction with a first laser (the excitation laser) and to quickly, reversibly, and locally stop (or inhibit) it with a second laser (the depletion laser). A spatially shaped depletion focus with points of zero intensity (like, e.g., a donut focus) can then be used to contract the effective reaction volume towards the zeros, ideally to spatial scales way below the diffraction limit. Stopping or inhibiting the polymerization reaction in DLW cannot only be accomplished by STED but has been realized in different ways including photo-activatable radical inhibitors14, 16 and the depletion of a long-lived yet unknown intermediate state of special photoinitiators in a scheme called RAPID.13, 17 We have found two photoinitiators that show a depletion behavior that is consistent with STED, namely isopropyl thioxanthone15 and 7-diethylamonio-3-thenoyl-coumarin10 (DETC). With our current DETC-based photoresist (Figure1a) we have demonstrated 2D and 3D resolution that exceeds the capabilities of regular DLW and beats the lateral and axial diffraction limit for the first time.10, 18 However, being restricted to continuous-wave (cw) depletion with a fixed wavelength of 532 nm in previous experiments, we could not determine the exact nature of the depletion mechanism so far and, hence, we have called our scheme STED-inspired DLW. We have recently performed femtosecond pump-probe experiments revealing that DETC favors stimulated emission in ethanol solution.19 However, the mechanism of the observed polymerization inhibition within our photoresist can yet be distinct, as the molecular transition rates can be different when dissolved in a monomer instead of ethanol. Furthermore, the states subsequent to the S1 depicted in Figure 1b have significantly longer lifetimes and are hence exposed to more depletion photons than the S1 itself. Thus, an effective polymerization inhibition caused by light absorption through such a long-lived intermediate state could possibly dominate over STED, even with small transient absorption cross-sections.
One motivation for the exploration and utilization of stimulated emission as depletion mechanism is its fast and reversible nature. While a potential STED depletion is performed within the S1 lifetime of the initiator (typically between τ = 0.1 ns13 and τ = 4 ns), the characteristic timescales of alternative depletion mechanisms can be much longer. For example, in case of a photo-activatable inhibitor, the relevant timescale is the lifetime of the activated inhibitors (τ = 2 μs20 to τ ≤ 200 ms16). For the intermediate state in RAPID lithography13, 17 the lifetime is probably in the ms range. (The effect described in ref. 17 leads to a self-depletion by the excitation laser itself at a scan velocity v = 30 μm/s. As the self-depletion is considerably decreased for slightly higher scan velocities, the depletion timescale must be on the order of the exposure time of a single voxel or even longer. Hence, τ ≈ d/v = 6 ms with voxel diameter d = 180 nm.) On the one hand, such long interaction times are attractive, because they allow for the use of robust and inexpensive cw lasers and can also prevent diffusion-blurring of the excitation pattern, as they can still catch active species diffusing out of the region of interest. On the other hand, long interaction times also directly limit the maximum attainable scan speed. Suppose a method could create voxels with diameter d = 20 nm. After a single voxel's exposure, the next adjacent voxel could not be created before the characteristic timespan τ had passed at least once. This limits the maximum scan velocity v for, e.g., 20 nm wide lines to between v = d/τ = 1 μm/s and v = 10 mm/s for an inhibitor scheme and v = 3.5 μm/s for RAPID, whereas the maximum scan velocity for true STED would be as large as v = 20 nm/4 ns = 5 m/s. One should keep in mind that the resolution improvement in such STED-inspired dual-beam approaches13–15 is only valid for sequential exposure. Indeed, in a single-shot parallel exposure (like, e.g., using interference gratings for both excitation and depletion), the width of the exposed lines can become arbitrarily small. However, the attainable spatial period (i.e., the resolution in the sense of Abbe) remains diffraction-limited because the periods of the excitation and the depletion pattern are both diffraction- limited.21 Thus, scanning or multiple exposures are inherently connected to such schemes and minimum exposure times are critical. Consequently, we consider using STED – though necessitating higher laser powers – as very promising for next-generation 3D lithography and perhaps even for nanoscale 2D lithography with decent throughput.
In this Communication, in order to study the temporal decay and the spectral sensitivity of the depletion pathway(s) of DETC, we perform pump-probe-like lithography experiments. We use 100 fs pulses with 820 nm center wavelength at 80 MHz repetition rate for two-photon excitation. Synchronuous pulses with an estimated duration of 250 ps and tunable center-wavelengths between 500 nm and 600 nm are used for depletion (see Figure 1c). Details on setup, resist and exposure conditions are available as supporting information.
To obtain a quantitative measure of the depletion capability, we determine the polymerization-threshold power Pth. We define the latter as the smallest excitation laser power that still yields clear polymer lines that survive the development process. The corresponding lines are selected manually in the optical dark-field micrographs. We start by writing a series of lines with increasing excitation power (horizontal axis) and zero depletion power directly onto the substrate-resist interface (lowest row in Figure2a) at a constant scan velocity of 100 μm/s. In this series, we determine the undisturbed polymerization-threshold power Pth,0. Next, we switch on the depletion laser tuned to 532 nm wavelength. Increasing the depletion laser power (vertical axis) shifts the threshold to higher powers Pth,shifted, because part of the excited photoinitiator molecules are depleted and cannot contribute to the polymerization reaction. This means that for a given depletion power the polymerization can be effectively inhibited for writing powers between Pth,0 and Pth,shifted. The relative shift in threshold power (Pth,shifted–Pth,0)/Pth,0 in% is used as a measure of the depletion capability. It can be seen from Figure 2a that the depletion capability increases up to an optimum average depletion power of 10 mW. Beyond that power, the depletion effect gets weaker and finally, for yet higher depletion powers, the threshold even decreases. Here, the depletion laser leads to enhanced excitation that dominates over the depletion effect. On the basis of additional fluorescence experiments (not depicted), we conclude that excitation by the depletion beam is mainly via single-photon absorption.
Next, we change the time delay between excitation and depletion pulses in order to investigate the temporal decay of the involved intermediate species. Figure 2a and b show the resulting patterns for two characteristic pulse delays Δt. In Figure 2a the depletion pulses arrive before the excitation pulses (Δt = -1 ns). In Figure 2b the depletion pulses arrive shortly after the excitation pulses (Δt = +0.2 ns). The corresponding relative shifts in polymerization-threshold power extracted from these test patterns are plotted in Figure 2c and show different behavior. We repeat the test pattern for various further time delays (not depicted). The resulting Δt-dependent relative threshold shifts are plotted in Figure 2d for selected depletion powers. Solid lines are fits to the data (see supporting information). Interestingly, even for Δt < 0 (i.e., the depletion pulses arrive shortly before the excitation pulses) we determine a positive threshold shift, which means that polymerization can still be suppressed, although 12 ns have passed since the last pulse of the 80 MHz excitation pulse train. Around zero time delay, the attainable depletion rises within the pulse duration of the depletion pulses. The decay time of roughly 1 ns can likely be assigned to the S1 state's lifetime τ1 of DETC, while another intermediate state with a considerably longer lifetime (e.g., the T1 state) is likely responsible for the depletion at negative (or large positive) time delays. So far, the fast effect (1 ns) originating from the S1 could be either stimulated emission or be induced by excited-state absorption into yet higher singlet energy levels. The unexpected slow effect (>12.5 ns), however, is definitely distinct from STED.
In previous experiments with DETC in ethanol solution, a different S1 lifetime of τ1 = 0.1 ns has been found.19 This seeming discrepancy can be explained by solvent-dependent non-radiative decay rates. To justify this assumption, we measure the fluorescence quantum efficiencies of DETC dissolved either in ethanol or in our monomer pentaerythritol tetraacrylate (PETTA) and obtain values of 2.3% and 29%, respectively (see supporting information). Assuming a solvent-independent radiative decay rate, we find that the S1 lifetime is also increased 12.5-fold to τ1 = 1.25 ns. We speculate that this increase in the more viscous monomer is due to steric hindering of conformational changes necessary for inter-system crossing, which is the process competing with radiative decay for S1 depopulation. This finding is in good agreement with the value of τ1 = 1.0 ns determined by fits to the above polymerization-experiment data.
To further clarify the nature of the fast process, we repeat the above experiments for different depletion wavelengths. For wavelengths ≤516 nm we observe little to no depletion at all (gray region in Figure3e). When approaching the fundamental absorption of the photoinitiator molecule, the overall behavior is dominated by single-photon absorption (SPA) of the depletion beam. For each depletion wavelength between 524 nm and 600 nm, we observe the same general temporal behavior as for 532 nm. However, the relative strengths of the observed slow and fast components change with wavelength. As depicted in Figure 3a, the fast effect decreases towards longer wavelengths while the slow effect gets even stronger. Circles are raw data, solid lines are fits like in Figure 2d. These different dependencies are another indication that two distinct depletion mechanisms are at work in our photoresist. To illustrate the different behavior of the two effects, we select two corresponding time intervals in each curve and average over 3 data points (indicated by filled circles in Figure 3a). One curve like in Figure 3a then yields only two values of the threshold shift, one value for the timing situation #1, where the depletion pulse arrives before the excitation pulse, and one value for timing situation #2, where the depletion pulse arrives “just in time” for optimal depletion. For timing #1 only the slow effect contributes, whereas for timing #2, both the fast effect and the slow effect contribute.
Figure 3b illustrates the corresponding values for timing #1 plotted versus depletion power and depletion wavelength. The optimum depletion powers are indicated by black circles. Beyond those powers, the behavior is likely dominated by SPA of the depletion laser. For short depletion wavelengths, the optimum depletion power is lower than for larger wavelengths, consistent with the higher SPA probability at shorter wavelengths. The maximum threshold shift of 45% is found for the longest depletion wavelength. Missing data for high depletion powers in Figure 3b–e are due to dominating SPA for short wavelengths and due to limited depletion power available in our setup at longer wavelengths. Figure 3c shows the same plot for timing #2 (where both effects contribute). Clearly, the attainable maximum relative threshold shift of 59% is larger. Furthermore, the optimum depletion powers (black circles) shift towards larger powers, indicating that the fast effect can further limit the effect of parasitic SPA.
We assume that each depletion component increases the initial threshold by a certain factor. If both effects contribute (like in timing #2), the increased polymerization-threshold power is the undisturbed value multiplied by the factors of both the slow and the fast effect. We can now separate the threshold increase caused by the fast effect by dividing the increased threshold values from timing #2 (where both effects have contributed) by the corresponding threshold values of timing #1 (where only the slow effect has contributed) for each wavelength and each depletion power. Using the resulting fictitious shifted threshold value (where only the fast effect would have contributed) we can calculate a relative threshold shift. The result is shown in Figure 3d. We find that the fast effect gets less pronounced for longer wavelengths. One should keep in mind, that by dividing through the values of timing #1, we probably do not only remove the contributions of the slow effect, but also parts of the limiting contribution of SPA. Hence, the fact that Figure 3d shows very little saturation towards high depletion powers should be taken with caution.
For a more quantitative evaluation we calculate effective cross-sections for the involved intermediate states according to (see supporting information). The corresponding data is depicted in Figure 3e, into which data from Figure 3b–d at 10 mW depletion power have entered. The data is normalized to the maximum value of the green curve. For comparison, we plot the absorption spectrum and the spectrum of the stimulated-emission cross-section22σsm (λ) ∝λ4F(λ) of DETC in arbitrary units along the vertical, where F(λ) is the previously measured fluorescence spectrum of the compound. These spectra were taken in PETTA solution. Again, the slow effect (red) gets more pronounced towards longer wavelengths. The combined effective cross-section (green) has its maximum around 532 nm wavelength. The retrieved effective cross- section for the fast component (blue) nicely follows the shape of the stimulated-emission cross-section. We consider this finding a strong indication that the fast depletion effect is actually due to stimulated emission.22
In a different set of experiments we find that the time constant of the slow component is below 1 μs (see supporting information).
In conclusion, we have shown that the photoinitiator DETC dissolved in the monomer PETTA can be efficiently deactivated via stimulated emission. This is the first unambiguous demonstration of a photoresist suitable for true stimulated emission depletion, raising hopes for significant future resolution improvements in 3D STED-DLW. The time constant of this fast process is about 1 ns. Furthermore, we identify a second slower depletion mechanism with a time constant in the range 10 ns–1 μs. The nature of this mechanism is yet to be determined. The data presented in this paper enable optimal choice of the appropriate depletion wavelength, pulse length and repetition rate for next-generation nanolithography setups. In particular, repetition rates of both excitation and depletion laser lower than 1 MHz would eliminate the unknown slow component. Depletion pulse durations of 100 ps to 1 ns would optimally exploit the remaining stimulated emission. Furthermore, the presented analysis allows for quantitative testing and comparison of the depletion performance of possible future photoinitiators for 3D STED-DLW.