Femtosecond laser direct writing (FsLDW) technology [1–5] is currently used for high-quality micro–nano fabrication by scanning laser pulses with durations from tens to hundreds of femtoseconds [6]. Specifically, energy deposition at a time scale shorter than electron–phonon coupling processes in any materials due to the ultra-short pulse width can suppress the formation of heat-affected zone, which allows laser processing with high precision and resolution [7]. FsLDW can be divided into two categories: subtractive and additive manufacturing. As an established extension of conventional manufacturing techniques, additive manufacturing builds geometry in a voxel-by-voxel fashion [8]. Currently, multi-photon lithography [9–11] including the two photon polymerization (2PP) is the most versatile additive manufacturing technology that has the ability to form true 3D complex multi-element structures with sub-100-nm resolution, which is needed to access a wide field of micro-optics [12,13]. With femtosecond laser ablation (FLA) [3,14–16] or femtosecond laser-assisted dry/wet etching (FLAE) [17,18], femtosecond subtractive processing enables precise and thermal damage-free removal or modification of a variety of materials with a wide range of scientific [19,20], medical [21], and industrial applications [22]. As an alternative approach to the conventional photolithographic patterning, the FLA or FLAE process has attracted considerable attention because it is a non-photolithographic, non-vacuum, on-demand, and cost-effective metal patterning fabrication route that can be applied to various substrates [23,24]. Overall and broadly speaking, FsLDW is a digitized technology, distinguished from physical pattern masks; the term “digital” here indicates that the laser is controlled by digitized parameters and a digital image or computer-aided design (CAD) is utilized for material processing [23]. This technology has been successfully demonstrated in the fabrication of micro-optical devices [25,26], on-chip optical interconnects [27,28], biological architectures [29], etc.

Advanced devices with complex 3D geometries often require both additive and subtractive manufacturing support [30], such as multilayer micro-chips [31], photonic chip packages [32], 3D artificial structures [33], and glass-ceramic micro-optical components [12]. The advanced hybrid manufacturing strategy usually deploying both 2PP and FLA inherits the high-precision advantages of these two technologies. Although micro-/nano-fabrication methods based on both 2PP and FLA are well established, considerable resistance still exists against their cooperation. The major reason lies in the incompatibility of material diversity and laser processing parameters [34]. 2PP technology is inefficient due to the intrinsic low absorption cross section in materials and point-by-point direct-writing strategy [35]. And FLA-based material removal requires using active focusing systems [36] for wide-area manufacturing due to the limited focus size. Besides, hybrid manufacturing strategy in 3D direct laser writing usually requires switching between different equipment and/or adapting the laser processing setup. Therefore, the fabrication time consumption dramatically increases, resulting in existing works remaining at the laboratory research stage [30]. Overcoming the inherent shortcomings of these two techniques will greatly facilitate the cost-effective and multi-scale fabrication of complex integrated devices. On the other hand, adaptive optical technologies have shown great potential in material processing with lasers [37]. In these applications, the introduction of a spatial light modulator (SLM) as an adaptive optics device allows tailoring light features into different fabricating geometries for rapid laser processing. Among various adaptive structured beams, Bessel beams [38] can readily generate non-diffractive micrometer-scaled central lobes along a millimeter-scaled propagation distance outgoing the Rayleigh range of Gaussian beams [39,40].

Based on adaptive Bessel beams, we present in this paper the femtosecond adaptive optics-assisted additive–subtractive hybrid manufacturing as a novel fabrication strategy by merging direct laser writing techniques of 2PP and FLA. Spoof surface plasmon polariton (SPP) waveguide devices [41,42] with sub-wavelength confinement and strong field enhancement characteristics are fabricated as a validation of this manufacturing strategy. In the additive manufacturing stage, we use femtosecond 2PP to construct the waveguide frames in the device. Non-diffracting Bessel beams were adapted using a single phase-only SLM to realize parallel direct writing. With the aid of the adaptive optics, the overall exposure time of the structure is limited within 50 min. After fabricating the waveguides, gold films are deposited on the waveguide frame. Besides the application in the additive 2PP stage, Bessel beams are also deployed in the subtractive fabrication of the SPP excitation structure. In this stage, the FLA technique was used in carving the gold film into an SPP wave-excitation array. In comparison with the excitation array etched by the more expensive and complex focused ion beam technique, the device fabricated with the hybrid laser technique shows rival performance in terms of the excitation qualities. Besides, the fabricating efficiency is significantly enhanced by more than 16 folds with parallel material processing via adaptable Bessel beams. In addition, this strategy also shows great potential of FsLDW in wide-area manufacturing of integrated THz-wave devices. Unlike the conventional mask lithography, this strategy only takes 10 steps (including substrate pretreatment; see Appendix A Tables 1 and 2 for details) to fabricate SPP waveguide devices. To the best of our knowledge, this is the first-time that adaptive optics has been used to fabricate the SPPs devices by FsLDW.Table 1.

Process Flow of Fabricating SPP Waveguide by Mask Lithography (Excluding Mask Manufacturing)

The experimental setup of femtosecond adaptive optics-assisted additive–subtractive hybrid manufacturing is shown in Fig. 1(a). The laser source is a homemade femtosecond laser system, enabling 90-fs transform-limited laser pulses output at a 1-MHz repetition rate. A BBO crystal is inserted to generate the second harmonic (SH) wave centered at 525 nm for the 2PP fabricating process (not shown). After transmission through a beam expander (BE), the SH wave illuminates the phase-only SLM (Holoeye PLUTO, $1920\times 1080\mathrm{pixels}$). A half wave plate (HWP) is inserted to guarantee the maximum diffraction efficiency from the SLM. The L1 lens ($f=1\mathrm{m}$) and the microscope objective (MO, Olympus $20\times$, NA 0.4) constitute a telescope to shrink the beam, and a neutral density (ND) filter is used to control the exposure power. A CCD camera is located in the focal plane of the L1 lens. In our adaptive optics-assisted setup, an interference-based method allows sensor-less wave-front sensing and correcting the heavily distorted beams with the help of the SLM as demonstrated in our previous work [43]. Besides, the fabricating beam after the objective can also be adapted according to the specific fabricating structures by the programmable SLM.

To demonstrate the outstanding advantages of the novel strategy in parallel and wide-area manufacturing, terahertz (THz) functional devices with millimeter scales are fabricated and tested. THz photonics is a powerful and efficient tool for bio-sensing [44], nondestructive imaging [45], and next-generation communications [46]. In particular, SPPs are electromagnetic waves propagating along metal–dielectric interfaces, providing a promising way to enable terahertz device integration. Since most metals behave as perfect electrical conductors at terahertz frequency domains, the bound modes of terahertz surface waves are not supported on their surface. Researchers have proposed a corrugated metal structure to support the high confinement and propagation of SPPs at the terahertz frequency [47]. To distinguish them from optical SPPs, we call them spoof SPPs.

The geometrical features of the metal structure have a significant influence on the dispersion relation of the SPPs mode. The SPP waveguide devices in this paper are constructed by an array of periodic metal-filmed pillars similar to dominoes. The excitation region, coupling region, and straight waveguide region constitute the entire waveguide device. The pre-designed geometric features of the waveguide device are shown in Figs. 2(a) and 2(b). Coupling of free-space terahertz radiation to excite the SPPs was achieved by the periodic hole array in the excitation region, where the dimension of a single hole was $200\mathrm{\mu m}\times 40\mathrm{\mu m}$ and they were arrayed on the thin metal with a period of 485 μm along the $x$ direction and 250 μm along the $y$ direction. The coupling region is composed of columns with decreasing lengths, of which the maximum and minimum lengths are 1.58 mm and 0.28 mm, respectively. For operation in the terahertz frequency range, the parameters of the straight waveguide are chosen as $p=100\mathrm{\mu m}$, $w=200\mathrm{\mu m}$, $l=50\mathrm{\mu m}$, and $h=70\mathrm{\mu m}$. Numerical simulations of waveguide devices are presented in Appendix B (Figs. 8 and 9).

To accommodate the geometrical features of the SPP waveguide devices, non-diffractive Bessel beams generated by adaptive optics are adopted in the execution of the fast hybrid fabricating process. Bessel beams form a class of solutions to the Helmholtz equations that are propagation-invariant. Methods for generating non-diffractive Bessel beams rely on the conical intersection of wave-fronts generated by either a conical phase modulation and subsequent refraction or diffraction [39]. The phase synchronization and interference of the conical waves produce a highly localized central core (i.e., the main lobe) at an elongated distance far beyond the Rayleigh range of tightly focused Gaussian beams with similar transversal size, creating the so-called non-diffractive appearance. The non-diffracting behavior is a core advantage of the hybrid manufacturing strategy, avoiding non-critical sample positioning and thus allowing for fast parallel patterning over large non-flat surfaces [48].

The process flow diagram of SPP waveguide devices based on the hybrid additive and subtractive manufacturing is shown in Fig. 2(c). In the additive manufacturing stage, we complete the fabrication of the prototype SPP waveguide using the 2PP technique. By employing adaptive optics technology, Bessel beams with a controllable main lobe size are manipulated to achieve parallel exposure. After the magnetron sputtering deposition of the metal film, subtractive manufacturing is employed to fabricate the excitation region of the SPP wave by the scanning ablation with the Bessel beams. Finally, the functionality of the device containing the excitation region and waveguide structure is demonstrated in the terahertz near-field optical scanning test.

As the conventional 2PP method requires the layer-by-layer accumulation of tightly focused voxels with dimensions close to the Abbe limits, it usually takes hours or even days in our setup to fabricate millimeter-scaled devices even with simple microstructures as high as tens of micrometers. In fact, needle-like voxels can significantly accelerate the fabricating process by sparing the layer-by-layer scanning. Bessel beams are usually used to induce the needle-like voxels in 2PP as long as tens or hundreds of micrometers. And these beams can be easily generated with the help of the SLM by applying the phase below on the input plane wave as in Ref. [49]: $$\mathrm{Phase}{(\rho )}_{\mathrm{Bessel}}=-2\pi \rho /{\rho }_0,$$(1)where ${\rho }_0$ is the radial parameter and affects the features of the Bessel beams. A finer main lobe can straightforwardly produce finer feature size, but it will also reduce the throughput of laser direct writing and thus consume more exposure time. By balancing the feature size of the device and fabricating throughput, we choose the phase mask with ${\rho }_0=1.0\times {10}^3\mathrm{\mu m}$ on the SLM. To suppress the undesired interference in the Bessel beam introduced by the dead zone in between the pixels of the SLM, a blazed grating phase is superimposed in the phase. In this configuration, the phase information of the Bessel beam is encoded in the first order with other diffraction orders filtered out by an iris as shown in Figs. 1(c) and 1(d).

Figure 3 presents the intensity profiles of Bessel beams in direct space after the MO of the telescope shown in Fig. 1(a). Numerical results based on the angular spectral diffraction of the Bessel beam generated from the ideal Gaussian beam are presented in Fig. 3(a). Due to the presence of higher-order interference levels, the main lobe is surrounded by a series of concentric rings. Depending on the convergence angle and the lateral dimensions of the incident beams, the main lobe can propagate over significant distances of about 660 μm. Figure 3(b) shows the simulated transverse intensity profile at the position of maximum main lobe intensity [white dashed line in Fig. 3(a)] with full width at half maximum (FWHM) of $\sim 3.2\mathrm{\mu m}$. The measured intensity profiles of Bessel beams shown in Figs. 3(c) and 3(d) demonstrate the corresponding experimental results, showing good agreement with the numerical simulations. The experimental results of the intensity profile [Fig. 3(c)] show that the main lobe of Bessel beams occurs with a relatively even intensity profile along $\sim 242\mathrm{\mu m}$ propagation distance. Therefore, by fine-tuning the distance between the microscope objective MO and the wafer, this region within the nondiffractive Bessel zone is selected to participate in direct laser writing. Figures 3(e) and 3(f) are the SEM photos of the single voxel unit and crisscrossed microstructures directly written by the Bessel beams. During exposure, only the main lobe [marked by the yellow dashed box in Fig. 1(b)] can initiate 2PP by adjusting the laser attenuation via the ND filter. When the exposure parameters are configured as 120 μm/s (scanning speed) and 1.6 nJ (single pulse energy), the feature size of the Bessel beams for additive manufacturing is approximately 3.3 μm.

In the additive manufacturing stage of SPP waveguides, the negative photoresist (MicroChem SU–8 2035) is uniformly spin-coated on the 2-inch wafer at 1400 r/min. Soft baking was performed on a hot plate for 90 min at 65°C. The thickness of the photoresist after soft-baking just reaches the design height $h$ of the waveguide (calibrated by the step meter). The parallel exposure process only requires employing a high-precision two-dimensional ($X-Y$) translating stage. At a scanning speed of 120 μm/s, the overall consumption time is less than 50 min. The post-exposure bake was executed with the same steps as the soft-baking process, and, after that, the wafer was placed in the developer to rinse the soluble, unexposed negative photoresist. Besides, the heating and cooling rate of the post-baking needs to be carefully controlled at a low level to reduce the polymer’s internal stress after development. We note that the adaptive optics actually provides high flexibility in controlling the voxel features, thereby significantly reducing the total voxel number required in device forming. In fabricating the 2D patterned microstructures in high-aspect ratio similar to those in Fig. 3(e), the strategy with the diffraction-free Bessel beams can significantly simplify the 3D point-by-point scanning process in multiple layers into a single scan in the 2D pattern. As demonstrated in our preliminary work [50], the Bessel beam drastically reduces the time consumption in fabricating microtubes as high as 60 μm to 1/25 of the time consumption with the 3D point-by-point direct laser writing method.

After the 2PP process, a gold film with a thickness of 200 nm was deposited on the surface of the wafer by magnetron sputtering, including the sidewalls of the pillar array. The SEM images of waveguide topography obtained with an scanning electron microscope (SEM, TESCAN MIRA LMS) are shown in Fig. 4(a). The SEM results show that the waveguide micro-pillars meet the design specifications and have satisfying surface smoothness. Good morphology consistency is also maintained between different pillars in the straight waveguide region, indicating that additive manufacturing has good process stability. The height and upper surface profile information of a single waveguide structure obtained by using a step meter (Bruker DEKTAK XT) along the white dotted line path in Fig. 4(a) are shown in Fig. 4(b). The average height of the waveguide is 69.6 μm, only 0.4 μm lower than the design height ($h=70\mathrm{\mu m}$), and the root mean square (RMS) error of the upper surface roughness is 0.0628 μm. Further, we present the numerical results (Fig. 8 in Appendix B) of the dispersion curves of the pillars at these two heights, and the coincided curves show that the 0.4-μm discrepancy has trivial impact on the functionality of the waveguide device.

The usual technological approaches for metal patterning rely on various tools, among which are focused ion beam (FIB), photolithography, reactive ion etching, etc. [51] These fabrication methods are mature and highly performant. However, a series of procedures and extreme environmental assistance (e.g., vacuum or inert gas) are required and need to be optimized for each material used, which inevitably increases the overall costs and complexity of the whole process. In contrast, FLA offers attractive advantages as an alternative and complementary solution in terms of a mask-less and single-step process without the demand for a vacuum environment or chemicals. In addition, FLA can also yield individual geometrical features even below the laser wavelength due to the nonlinear optical processes involved.

Several properties of Bessel beams facilitate wide-area ablation processing of the material. Specifically, the intensity of the main lobe can be highly localized along a propagating distance several orders of magnitude beyond the Rayleigh range. This feature provides the possibility of surface processing on non-flat materials and eliminates the spatial constraints of sample positioning toward the laser focus [52]. In addition, the self-healing property of the Bessel beams effectively prevents the beams from being perturbed by obstacles or particles in their path.

Adaptive optics-assisted FLA completes millimeter-scale excitation region fabrication of SPP waveguide devices during the subtractive manufacturing stage of the hybrid manufacturing strategy. In this stage, the generated Bessel beams with a feature size of about 3 μm periodically ablate the metal region close to the waveguide region, making the ablated region transparent to terahertz waves and exciting SPPs. Considering the principles of material ablation, FLA requires a higher single pulse energy than the 2PP process in the additive manufacturing stage to reach the ablation threshold required for 200 nm gold film, in addition to ensuring that the Bessel beams can ablate steadily on the millimeter scale without being affected by the undulations of the sample. In our experiments, the minimum single pulse energy required to perform FLA is 0.6 μJ. According to the experimental results in Fig. 3(d), the main lobe of the Bessel beams only contains about 29% total energy, corresponding to $\sim 0.52\mathrm{J}/{\mathrm{cm}}^2$ laser fluence. This is comparable to the values ($\sim 0.5\mathrm{J}/{\mathrm{cm}}^2$) reported in the literature [53].

Figure 5 shows the optical microscope and SEM images of the entire waveguide structure after FLA. The overall fabrication time of the excitation region is about 30 min at an ablation rate of 200 μm/s. The current laser fluence configuration also reaches the melting threshold ($\sim 0.17\mathrm{J}/{\mathrm{cm}}^2$) [54] for silicon substrate, leading to the spatial redistribution of material via metal heat transfer, local melting, and rapid solidification. The morphology of the substrate surface after melting and reconsolidation is shown in Figs. 5(c) and 5(d), where disordered granular microstructures are induced under the scanning irradiation of the femtosecond laser. Although the anti-reflection effect of surface microstructures in the terahertz waveband has been demonstrated [55], its functional impact on SPP waveguide devices needs to be further verified, which will be explained in detail in the next section.

To verify the functionality of the on-chip SPP waveguide devices, the samples were characterized using a fiber scanning near-field terahertz microscopy (SNTM) system. The experimental setup of the SNTM system is shown in Fig. 10 (Appendix C). During the test, the THz wave (generated by a photoconductive antenna) was focused on the excitation region fabricated by the FLA, and its polarization direction was parallel to the waveguide direction. The terahertz near-field probe with a spatial resolution of 8 μm was located 70 μm above the waveguide and scanned the signals point by point with a step of 100 μm along the $Y$ direction and $X$ direction.

The collected near-field distribution of the waveguide surface is shown in Fig. 6. The normalized power ($|E_z{|}^2$) was imaged at 0.6 THz [Fig. 6(b)]. Tightly confined SPPs guiding along the waveguide structure were demonstrated by the near-field image. Figure 6(c) shows the terahertz spectrum comparison between the central position of the excitation region and the straight waveguide region. The major spectral peak of SPPs generated in the excitation region occurs at the frequency positions of 0.3 and 0.6 THz. However, the disappearance of 0.3 THz spectral peak in the straight waveguide indicates that the transmission on the waveguide cannot be well supported for the frequency deviating from the design wavelength, which means the devices have good mode selectivity. Figure 6(d) plots the measured intensity (the central wavelength 0.6 THz) as a function of distance with a polynomial fit (solid blue line), and the corresponding propagation loss was about 1.7 dB/mm. Figure 6(e) plots the time-domain signal curves of SPPs at the head and tail of the waveguide in the same temporal window. Thanks to the large range of the delay line in the SNTM, the delay arising from the traversing time of THz SPPs through the waveguide can also be determined as $\sim 15.4\mathrm{ps}$ in a single scanned window. This delay is mainly due to the change of the effective refractive index induced by the fabricated waveguide structure, and the corresponding numerical solution of the dispersion curve of the SPP mode is shown in Fig. 9. The attenuation of intensity along the $Z$ direction also demonstrates the ability of the waveguide to confine SPPs. To demonstrate this, the intensity at the start position of the straight waveguide ($X=0$, $Y=0$) is sampled as a function of distance from the surface [red scatter in Fig. 6(f)]. The spatial evanescent extension (distance where the field drops to $1/e$ of the surface value) in the air is 110 μm, obtained by an exponential fit (blue solid line). The spatial expansion of the textured surfaces with pillar array was reduced by several orders of magnitude compared to bare metal surfaces. The experimental results of the near-field characterization match the numerical simulation expectations, which means the on-chip SPP waveguide devices fabricated by the FsLDW process have both transmission mode selectivity and strong confinement of the spatial sub-wavelength field.

In the SEM photos in Fig. 5, the laser-induced disordered microstructures on the substrate surface are shown in the ablated pattern of the excitation region. To further evaluate the influence of the above laser-induced microstructures on the SPPs device performance, FIB [56] is used to refabricate the excitation region of another SPP device with the same feature parameters. In fabricating this device, the parameters in the additive manufacturing stage and the magnetron sputtering process are maintained. As shown in SEM photos of Fig. 7(a), each sub-wavelength hole in the excitation region of this device presents a clear and sharp boundary with a much smoother etched area compared with those ablated by the laser in Fig. 5(d). Note that it takes more than 8 h to etch the whole excitation region with FIB, while the FLA processing takes only $\sim 30\min$. And both devices show nearly identical performance on SPP excitation despite the rugged ablated excitation region by FLA (as will be shown in the following paragraphs). Furthermore, the common equipment of FLA with 2PP techniques and no requirement of vacuum environment makes FLA more cost-effective and environmentally friendly.

Irradiated by the same THz source with identical power and polarization state, we perform the functional test of the device in the SNTM system (the distance between the probe and the upper surface of the waveguide is still accurately controlled at 70 μm). Figures 7(b)–7(d) compare the performance of the SPP devices with excitation arrays fabricated by the FIB and the femtosecond Bessel beams. Both devices present spectral peaks in the excitation region close to the predesigned 0.6 THz as shown in Fig. 7(b) except a tiny shift to 0.62 Hz for the device with an FIB-etched excitation region. This probably comes from a slight deviation from the predefined positioning of hole arrays in FIB processing. As for the device with a laser-ablated excitation region, measured spectra show a bit enhancement in higher frequency range (0.64–0.7 THz). We attribute this to the potential broadband anti-reflective effects in the ablation-related microstructures as shown in Fig. 5(d). As shown in Fig. 7(d) and in Fig. 6(b), the normalized mode distributions [$|E_z(x,y){|}^2$] near the waveguide surfaces in both devices are highly consistent with each other. The excited SPP fields are gradually shrunken in the funnel region and propagated along the following straight waveguide. In addition, the two waveguide devices exhibit different intensity decay curves [Fig. 7(c)]. Devices with the excitation region fabricated by the FIB have a slightly higher transmission loss ($\sim 2.0\mathrm{dB}/\mathrm{mm}$), which may be caused by the fluctuation of the femtosecond laser source during the additive manufacturing stage and the immature cooperation between various processes. Despite the slight performance differences between the two devices, almost the same intensity of SPP radiations is excited in the micro-arrays fabricated by both techniques at the starting point of the waveguide ($X=0–1\mathrm{mm}$). Therefore, we conclude that the disordered granular microstructures show insignificant adverse effects on the excitation of the SPPs. Overall, the adaptive optics-assisted subtractive manufacturing strategy enables fabricating the excitation regions in a much shorter time (30 min with FLA versus 8 h with FIB) and at much lower cost without degraded performance in comparison with the commercial FIB scheme.

In summary, a convenient, eco-friendly, and cost-effective 3D manufacturing method is developed by integrating additive manufacturing (2PP) and subtractive manufacturing (FLA) into a single framework using FsLDW. Moreover, adaptive optics technology with SLM as the core device provides a variety of increased capabilities for laser processing. By generating non-diffractive Bessel beams capable of overcoming Rayleigh range limitations, this strategy significantly compensates the inherent inefficiencies of traditional 2PP 3D forming technologies and effectively avoids non-critical sample positioning, thus allowing fast patterning over large non-flat surfaces without the aid of any focusing feedback device. As a validation of the hybrid manufacturing strategy, SPP waveguide devices with millimeter dimensions are successfully fabricated. Additive manufacturing with 2PP as the implementation solution simultaneously completes the parallel direct writing of the coupling region and waveguide with a device height of 70 μm in less than 50 min. The patterning of the metal films in the excitation region of the device is accomplished using a non-vacuum and chemical-free FLA technique (subtractive manufacturing) with a fabricating time $\sim 30\min$. The performance of spoof SPP devices was evaluated quantitatively based on the SNTM measurements. Tightly confined SPPs with a 110 μm spatial evanescent extension were demonstrated to propagate along the sub-wavelength straight waveguide, which is in good agreement with the numerical results. In addition, comparative results with FIB etching confirmed that the laser ablation subtractive manufacturing technique shows rival device performance even with granular microstructures in the excitation region. We expect that this adaptive optics-assisted hybrid manufacturing strategy can be applied in more fields such as complex terahertz on-chip integrated devices, flexible photonics, and holographic metasurfaces.

Acknowledgment. The authors acknowledge Professor François Courvoisier for the helpful discussions and suggestions on polishing the manuscript. Particular acknowledgements should be given to Professors Lin Zhang, Chao Jin, and Bowen Liu for the assistance in FIB etching, metal coating, and FLA. Also, we thank the Shiyanjia Lab for supporting the SEM tests in this study [58].