1. Introduction

Micro- and nanofabrication techniques using light have demonstrated remarkable advancements in micro/nanoengineering [1]. The invention of multiphoton fabrication resulted in development of the three-dimensional nanofabrication technique [2], and direct lithography using extreme ultraviolet (EUV) light has pushed the limits of integration density in semiconductor manufacturing [3]. In the visible and near-infrared (NIR) wavelength regions, femtosecond laser processing of transparent materials in combination with chemical etching can be applied to manufacture microfluid-based devices [4]. Various techniques have been developed to meet the demand for nanoengineering in a wide range of fields, from fundamental science to industrial applications.

Recently, in the field of transmission electron microscopy (TEM), the use of transmission-type phase diffractive elements with sub-100-nm-thick materials has been attracting attention. For example, nanofabricated holographic diffraction gratings can be used to generate structured electron beams, such as electron vortex beams having orbital angular momentum, and electron Bessel beams demonstrating nondiffractive and self-healing properties [5–8]. Non-uniform electron beams are used to investigate the nature of electron wave functions and to develop novel measurement methods that cannot be achieved with conventional uniform electron beams. Furthermore, an elaborately fabricated multislit-shaped electron phase grating functions as a beam splitter instead of an electron biprism [9–11]. Such a device could be an essential component for atomic-resolution electron phase imaging in scanning TEM.

The electron phase elements reported to date were fabricated using the state-of-the-art focused ion beam (FIB) system and advanced manipulation techniques. However, because of the large momentum transfer to the material by the ion beam, which is typically composed of Ga$^{+}$ ions of 1–50 keV energy, determining the optimal processing parameters without breaking the membrane is difficult. In addition, the fabrication of each element requires more than 0.5 h for the need to individually fabricate each slit of the grating.

Instead of FIB milling, ultrafast laser ablation by femtosecond laser has been demonstrated as an effective method to treat nanomembranes because of the nonthermal nature of the processing. Recently, we demonstrated the fabrication of a holographic diffraction grating to generate electron vortex beams in TEM [12]. The grating was made of 35-nm-thick silicon and created by a two-beam interference process using an NIR femtosecond laser. This method allows for single-shot processing of two-dimensional periodic structures over an area of 30 $\mathrm {\mu }$m in diameter; this feature is an excellent advantage over FIB milling in terms of processing speed.

In this work, we demonstrate the application potential of the laser ablation technique to manufacture nm-thick submicrostructured elements. To the best of our knowledge, this is the first study to directly process a thin membrane (10 nm) using ultrafast laser ablation. We processed the membranes into a grating shape using two-beam interference processing technique and systematically measured the fabricated geometry by varying the input pulse energy near the ablation threshold. In addition, a femtosecond laser with a half-wavelength by the second-harmonic generation, instead of the NIR laser used in our previous work, was used to process the further increase in resolution. A grating spacing of 1.5 $\mathrm {\mu }$m was realized in our previous work, which provided a small electron diffraction angle of a few microradians in TEM, whereas a larger diffraction angle is preferable for practical applications. For instance, holographic gratings with about 650-nm grating spacing were employed to study the interaction behavior of electron vortex pairs [13]. Because of the halved laser wavelength, we obtained a grating spacing of $\sim$0.75 $\mathrm {\mu }$m.

2. Experimental setup

A schematic of the experimental setup is shown in Fig. 1. We used a laser source operating at a wavelength of 1040 nm with an energy of 40 $\mathrm {\mu }$J per pulse and a pulse duration of 310 fs full-width at half-maximum (Spirit One, Spectra-Physics). The wavelength of the output was shortened by the second-harmonic generation using a BiB$_{3}$O$_{6}$ crystal. The 520-nm wavelength beam was split into two paths with a certain power branching ratio by a combination of a half-wavelength plate and a polarized beam splitter. One of the beams was expanded to 12 mm in $1/e^{2}$ diameter, entered via a spatial light modulator (SLM; LCOS-SLM X10468-07 NIR, Hamamatsu Photonics K.K.), passed through two 4f systems (L1-L2 and L3-L4) with a total magnification power of 1/250, and focused on a sample at normal incidence. Here, a slight tilting phase mask with a diameter of 4 mm was created on the SLM to separate the truncated beam from the unmodulated light; the mask diameter defined a processing area of the sample. The other path equipped an optical delay line to synchronize irradiation timing, and it was focused on a sample at an incident angle of $\sim$45$^\circ$ by an off-axis parabolic mirror. The two laser pulses crossing on the sample created an interference intensity distribution and processed the membrane into a periodic grating shape by single-shot laser irradiation.

 

Fig. 1. Schematic of the experimental setup. SHG, crystal for second-harmonic generation; filter, harmonic separation filter; HWP, half-wave plate; PBS, polarized beam splitter; SLM, spatial light modulator; delay, optical delay line; OAP mirror, off-axis parabolic mirror. L2 is a concave lens, which formed a 4f system together with a convex lens L1. L4 is an objective lens with a numerical aperture of 0.4 and working distance of 20.2 mm, which formed an f4 system together with a convex lens L3.

Download Full Size | PDF

We used self-supporting membranes made of silicon nitride of 10 and 30 nm thicknesses (Norcada), which are originally designed for use in TEM analysis. The size of membrane was 100 $\mathrm {\mu }$m $\times$ 100 $\mathrm {\mu }$m. Thus we could create around 10 grating masks on the same membrane.

3. Processing results and analysis

The membranes were treated while changing the output power of the laser source with a fixed power branching ratio. The results of the processed 10-nm membranes obtained via scanning electron microscopy (SEM) are presented in Fig. 2. The images in the lower row are enlarged versions of the corresponding upper images. The input pulse energies were 3.1, 3.9, 4.8, 5.6, and 6.4 $\mathrm {\mu }$J from the left to the right, respectively. Since the laser source had a spatial mode of $\mathrm {M}^{2}<1.2$, we could estimate corresponding peak laser fluences by assuming the fundamental Gaussian profile, as were 0.67, 0.89, 1.1, 1.3, 1.6, and 1.8 J/cm$^{2}$, respectively. The pulse energy and fluence for the ablation threshold were estimated as 2.4 $\mathrm {\mu }$J and 0.67 J/cm$^{2}$, respectively. We found that the membrane was processed into the grating shape with a diameter of $\sim$16 $\mathrm {\mu }$m, which reflected the set aperture on the SLM, and had the grating spacing of $\sim$735 nm, which was determined by the laser wavelength and the crossing angle of the two pulses. The obtained width of the grating crossbeams systematically reduced as the input pulse energy increased. The lower right image shows that part of the grating was distorted, indicating that the upper limit of the input energy was reached.

 

Fig. 2. Two-beam interference processing of 10-nm membranes under an input pulse energy of 3.1, 3.9, 4.8, 5.6, and6.4 $\mathrm {\mu }$J (from the left to the right). The dark part of the image indicates through-slits. The right images are details of the center of the left images. The scale bars are 3 $\mathrm {\mu }$m (upper) and 500 nm (lower).

Download Full Size | PDF

The measured crossbeam width, that is the width of unprocessed bar-shaped structure of the fabricated gratings, are plotted in Fig. 3. There are three different results for the 10-nm membrane, as is shown in Fig. 2 (open circles), the 30-nm membrane (triangles), and the 10-nm membrane with another power branching ratio (squares). The errors could be attributed to the measurement error and the nonuniformity of the crossbeams. The horizontal axis represents the normalized fluence $F/F_{\textrm {th}}\geq 1$, where $F$ is the input fluence and $F_{\textrm {th}}$ is the experimentally obtained threshold fluence. In each case, the crossbeam width systematically narrowed with increasing pulse energy, and the grating was broken at a certain input pulse energy.

 

Fig. 3. Variation in the measured widths of the crossbeams with normalized input laser fluence. The thickness of the processed membranes is 10 nm for the open circle and triangle markers, and 30 nm for the square markers. The set power branching ratio was different for the circles and triangles. The curves are the model functions for the process. See the main text for the detail about the dagger and the asterisk markers.

Download Full Size | PDF

To explain those results, we considered a simple model following the assumption in [14,15]; namely, part of the membrane is removed when exceeding the threshold fluence. Two plane light waves with the normal incidence and the 45$^{\circ }$ incident angle were assumed to create the interference distribution on the sample as is shown in Fig. 4(a). For our experimental condition, the width of the laser overlapping region on the sample is given by $w=c\Delta t/ \sin \theta =132$ $\mathrm {\mu }$m [16], where $c$ is the speed of light, $\Delta t=310$ fs is the pulse duration, and $\theta =45^{\circ }$ is the crossing angle. This is wider than the processing area of interest, which is $\sim 16$ $\mathrm {\mu }$m. The one-dimensional distribution of the laser fluence lies on the $x$ axis that is perpendicular to the interference fringe can be written as

 

Fig. 4. Schematics of the assumed model for ablation processing. a) Conceptual drawing of input light waves, the created intensity distribution on a sample, and the resulting structure. b) Relationship between the crossbeam width $w_{\textrm {cb}}$ and the distribution of the fluence $F$. $F_{\textrm {th}}$ is the ablation threshold fluence; $x_{\textrm {th}}$ is the position at the threshold.

Download Full Size | PDF

With a certain threshold fluence, $F_{\textrm {th}}$, we obtained the equation, $F_{\textrm {th}} = f(x_{\textrm {th}})$, as is shown in Fig. 4(b). Here, $x_{\textrm {th}}$ is the threshold boundary position. The realized crossbeam width in the processing is derived from the grating period, $\sqrt {2}\lambda$, minus the ablated width, that is the width of the region where the fluence is above the threshold: $w_{\textrm {cb}}=\sqrt {2}\lambda -2x_{\textrm {th}}$. Therefore, by solving $F_{\textrm {th}} = f(x_{\textrm {th}})$ for $x_{\textrm {th}}$ and substituting it into the above expression of $w_{\textrm {cb}}$, the following model function was obtained:

The function with the fit parameter of $\alpha$ was fitted to the experimental data, as is shown in Fig. 3. The solid, dotted, and dashed lines are the fitted curves with $\alpha =0.6$, 0.76, and 0.84, respectively. The curves are in good agreement with the data, suggesting that the fabricated geometry depends on the intensity distribution of the superposed laser pulses. In particular, it was found that the dependence of the crossbeam width could be controlled by changing the visibility. However, designed values of $\alpha =0.52$, 0.52, and 0.83 for the circles, triangles, and squares were somewhat different from the fit parameters, respectively. In the experiment, data for the same conditions were acquired within 10 minutes, whereas data for other conditions were acquired more than a day apart. The mismatch in the timing of femtosecond pulses due to the long-term drift of the optical path length may have caused the decrease in the interference.

4. Realization of nanostructures

We found that the realized minimum geometries reached nearly 100 nm for the through-slits of the grating and sub-100 nm for the crossbeams. Figure 5(a) shows an SEM image of the observed narrowest through-slit (the dark part on the image) of the sample indicated by a dagger ($\dagger$) in Fig. 3. The measured width was $113\pm 8$ nm, which corresponds to $\sqrt {2}\lambda -w_{\textrm {cb}}$. The processed sample also revealed that a part of non-through groove structures (the medium-gray line-shaped parts on the image) had a narrower width than that of the through-slit structures, suggesting that a through-slit with a width of sub-100 nm can be obtained with a thinner sample. Furthermore, Fig. 5(b) presents an SEM image of a fabricated grating indicated by an asterisk (*) in Fig. 3. The grating had a crossbeam with the narrowest width of $43\pm 8$ nm, which is less than one-tenth of the laser wavelength.

 

Fig. 5. SEM images of the realized nanostructures. a) Realized narrowest through-slit in the experiment. The measured width indicated by arrows was $113\pm 8$ nm. The medium-gray parts of the images show non-through groove structures. The thickness of the processed membrane was 30 nm. b) The narrowest part of the crossbeam of the fabricated grating. The measured width indicated by arrows was $43\pm 8$ nm. The thickness of the processed membrane was 10 nm. The scale bars are 500 nm.

Download Full Size | PDF

Both nanostructures had processing challenges in terms of uniformity because the realized widths of the slit and the crossbeam were overly sensitive to the variation in input fluence according to the plot in Fig. 3. The uniform intensity profile of the incident laser and flatness of the membrane are important for the grating fabrication, whereas the most critical criterion for high-quality fabrication is to accordingly design laser intensity distribution. If the gradients around the peaks and valleys of the distribution are steeper, it is easier to fabricate the nanostructures. In addition, accurately synchronizing and maintaining pulse timing is necessary to establish this approach, because mismatched timing results in reduced interference visibility. An optical system using a common objective lens with a large numerical aperture [17,18] would be more appropriate than the two-beam interference setup presented in this article, in which long-term visibility tends to change due to the thermal expansion and contraction of the optical table.

5. Conclusion

In conclusion, we demonstrated the fabrication of gratings with a diameter of 16 $\mathrm {\mu }$m using self-supporting membranes with thicknesses of 10 and 30 nm using by single-shot laser irradiation. We found that the laser fluence, normalized to the inherent threshold of the target material, is a critical design parameter in processing of the nanomembranes. We reduced the scale of periodic spacing to the submicrometer scale, which was half of our previously reported result. If we employ shorter wavelength light, e.g. ultrafast pulses in the XUV wavelength region [19], the spacing will be further reduced. Furthermore, the appropriate input energy provided the 113-nm through-slit and the 43-nm crossbeam width. Importantly, the structures, which were much smaller than the laser wavelength, were demonstrated by direct processing by light. Ease of designing the processing parameters and speed of processing are the significant advantages of the ablation technique compared to conventional FIB milling.

The fabricated gratings will be in demand in manufacturing electron-optical phase elements and membrane-based micro-electro-mechanical systems [20,21]. In addition, the presented method can be applied to verify unexplained laser processing dynamics, such as the formation of the laser-induced periodic surface structures (LIPPS) [22,23] and polarization dependence in laser micro/nanoprocessing [24], because it is easy to confirm whether the material is processed in a local area of interest.