Feedback-assisted Femtosecond Pulsed Laser Ablation of Non-Planar Metal Surfaces: Fabrication of Optical Apertures on Tapered Fibers for Optical Neural Interfaces

We propose a feedback-assisted direct laser writing method to perform laser ablation of fiber optics devices in which their light-collection signal is used to optimize their properties. A femtosecond-pulsed laser beam is used to ablate a metal coating deposited around a tapered optical fiber, employed to show the suitability of the approach to pattern devices with small radius of curvature. During processing, the same pulses generate two-photon fluorescence in the surrounding environment and the signal is monitored to identify different patterning regimes over time through spectral analysis. The employed fs beam mostly interacts with the metal coating, leaving almost intact the underlying silica and enabling fluorescence to couple with a specific subset of guided modes, as verified by far-field analysis. Although the method is described here for tapered optical fibers used to obtain efficient light collection in the field of optical neural interfaces, it can be easily extended to other waveguides-based devices and represents a general approach to support the implementation closed-loop laser ablation system of fiber optics.


Introduction
Ultrafast lasers opened new paradigms to material processing thanks to the high peak intensity achievable with focused pulses at relative low energies 1 . Femtosecond laser ablation and micromachining has been applied both on dielectric and on metallic samples [2][3][4] , and it has been used for the realization of laser-inscribed metallic surface structures such as micro-holes 4 , micro-grooves 5 , and micro-pillars 6 . Laser ablation has been also applied to optical fibers to obtain side emission of light by drilling micro discontinuities inside the fiber's core 7,8 . Although this process enables multi-site emission of light, it compromises the light propagation and collection properties of the waveguide itself, hence affecting its performances in sensing applications. This is a particularly relevant factor in applications where modal properties of the waveguide should be preserved while maximizing the light collection efficiency of the fiber.
Sensors based on Tapered Fibers (TFs) are an example 9 that has found a range of applications over the years, including sensing of biomolecules 10,11 , and, very recently, implantable systems to optically monitor neural activity 12 . In this latter case, the taper does not shrink at its central portion, as in canonical tapered fibers sensors based on transmission measurements, but it interrupts at its narrower section, creating a tip that enables for a smooth implantation in the brain and functional fluorescence collection 12 . The tip is then coated with reflective metal and a set of optical windows are opened along the narrowing waveguide, enabling guided modes of different order to interact with different regions of the brain. This approach enables micro-structured TFs (µTFs) to deliver and collect light with depth resolution along a single probe, a unique feature for optical neural interfaces [12][13][14][15][16] . Accordingly, tailoring the shape and size of the optical windows enables to engineer light delivery and collection volumes in order to collect functional fluorescence signal from groups of few neural cells 12 .
However, it is important that the structuring of the narrowing region preserves the effect of the waveguide on propagating modes, not to compromise the optical interaction between the environment and the modal content 17 .
The fabrication of µTFs with optical windows can be carried out with high-resolution fabrication techniques including Focused Ion Beam (FIB) Milling 18,19 , as well as fast-writing methods such as Ultraviolet Direct Laser Writing (UV-DLW) 20 . This latter approach allows for high-speed patterning of the metal coating at the cost of limited resolution and higher roughness of the window profile. However, both FIB and UV-DLW unavoidably undermine the underlying dielectric surface. Although this does not compromise the multi-site light delivery capability of the TFs, light collection efficiency would greatly benefit from an intact surface interacting with incoming fluorescence signal.
In this work, we propose a Feedback-Assisted Direct Laser Writing (FA-DLW) method to pattern the edge of a TF to optimize light collection properties for applications in the field of optical neural interfaces. The method takes advantage of laser ablation of the metal layer on the highly non-planar surface of the taper while it is submerged in a fluorescent bath. A scanning Near Infra-Red (NIR) femtosecond pulsed laser ablates the region of interest and simultaneously generates a two-photon fluorescent signal in the bath. Continuously monitoring this signal provides a spectral fingerprint that identifies when the metal is totally removed. The laser pulses interact mostly with the metal coating, leading to its ablation, while leaving the underlying glass surface intact. This enabled the production of optical windows that couple light in defined subset of guided modes, as verified by far-field analysis on the distal facet. The result is a fabrication approach that preserves the waveguides modal properties while optimizing the collection efficiency of the device.

FA-DLW Setup
The particular advantage of the proposed method is the ability to monitor the ablation progress in real-time. In order to record the collection properties of the optical apertures during the fabrication, we designed a Two-Photon (2P) ablation system assisted with a feedback signal ( Fig. 1A). A fs pulsed NIR Laser (Coherent Chameleon Discovery, average output power at 800 nm of 1.8 W) was modulated by a Pockels Cell. A quarter wavelength plate was used to obtain circular polarization, and the beam was expanded to fit the size of a 6 mm galvanometric mirrors pair (GMs) employed to deflect the beam over the (x,y) plane. The beam was then expanded and collimated by a scan-tube lens assembly and sent on the back aperture of a 4x 0.28 NA objective lens (OBJ1). The setup was controlled by a DAQ system interfaced with the open source software Vidrio ScanImage.
The laser spot was raster scanned in the field of view of OBJ1 (2085x2085 µm 2 ) ( Fig.1.A) generating a fluorescence spot into a fluorescent bath (30µM C20H12O5 (Fluorescein) in Phosphate buffered saline (PBS)), in which the aluminum-coated tapered fiber was submerged.
The fluorescence collected by OBJ1 was reflected by a dichroic mirror (D1) and detected by a Photo-Multiplier Tube (µscope-PMT). We used the resulting 2P fluorescence image to position the beam over the area to be ablated. Light guided by the taper propagated in the patch-cord and entered the feedback branch by an objective lens (OBJ2). The signal was filtered by a band-pass (central wavelength 525 nm, bandwidth 39 nm) and a NIR block filters (cutoff wavelength 770nm, short pass), and was directed toward two different optical detectors. A Spectrometer offered a real time feedback during fabrication, while a second PMT (Fiber-PMT) allowed for measuring a collection intensity map from the window just after the ablation.

Optical Windows Fabrication
TFs were obtained from OptogeniX (www.optogenix.com), and were produced from NA=0.39 fibers (Thorlabs FT200UMT) with core/cladding diameter of 200µm/225µm through a heat and pull procedure, described in detail by Sileo et al. 19 . Thermal evaporation was then employed to conformally deposit a 400 nm-thick Aluminum layer around the taper, setting the deposition rate at about 1.5 Å/sec while the TF was continuously rotated with a stepper motor.
The size of the windows was set by restricting the scanning region of the GMs to a 40x40 µm 2 area in correspondence of a taper diameter of 60 µm. Fabrication parameters were chosen considering two factors. (i) The laser fluence has to be higher than the ~100 fs pulse ablation threshold of Aluminum, found to be Ft,Al ~ 0.1 J/cm 2 , 21,22 but lower than both the damage threshold and the ablation threshold for fused silica, found to be, respectively, Fdamage,SiO2 > 2 J/cm 2 and Ft,SiO2 > 3 J/cm 2 . 23 Hence, we set the laser fluence at F ~ 0.2 J/cm 2 . Working close to the threshold enables an ablation rate of ~2-6 nm/shot in air 22 , with the liquid environment further reducing this rate. 24 (ii) A wavelength which could efficiently excite 2P fluorescence in the feedback fluorescent solution had to be chosen, hence we set λ = 800 nm. We configured a dwell time on each raster pattern's point of tdwell=3.2 µs (about one frame per second with 512×512 pixels).
Representative spectra recorded during the process are reported in Fig. 2A as 'regime I' for0 < t < 75 s, 'regime II' for 75 s < t < 750 s, and 'regime III' for t > 750 s). During 'regime I', no light is coupled inside the waveguide because the aluminum layer has not been sufficiently ablated yet. This is followed by 'regime II' in which the integrated intensity progressively increases. This corresponds to progressing metal ablation from the region of interest. During 'regime III' the integrated intensity starts decreasing and oscillating, although a constant fluorescence signal could have been expected after the complete ablation of the metal. We attributed this behavior to the generation of cavitation bubbles in the liquid environment near the focal spot, due to local heating [26][27][28] . Since the laser is focused on the fiber surface, the bubbles grow in the vicinity of the collection region, hence reducing the fluorescence signal that can be collected by the feedback system.
Since the ablation takes place on a curved surface while the laser scans in a plane, the local effective fluence depends on the cosine of the incidence angle 29 . Thus, the radius of curvature of the taper imposes a variation of the ablation depth per pulse as a function of the surface profile versus the planar scan plane, following 29  Where α=4πκ/λ is the effective absorption index of Aluminum, and κ=7.05 is its extinction index at λ=800 nm taken from Ref. 30 .
We calculated these variations considering the TF profile in the x-direction (Fig. 2D) and the taper angle along the z axis (Fig. 2E). Ablation could take place only for positions at which the ablation depth per pulse is positive (Fig. 2F). From these calculations we expect that the window size is minimally affected by variations in the z-direction (expected z-size 40 µm) but that it changes considerably along the x-direction (expected x-size ~20 µm). A representative Scanning Electron Micrograph of a FA-DLW window is shown in Fig. 2G. From the SEM image it is clear that the shape of the window is similar to the expected one outlined by the isodepth lines in Fig.2F, with the final size differing by about 5 µm in each direction. This is because Eq. 1 does not consider the actual size of the laser spot, especially along the y direction, while our system has a PSF with lateral FWHM of 3 µm and axial FWHM of ~30 µm 31 .
Moreover, residual thermal energy deposition could affect the ablated region, enlarging its edges 32,33 .
The software MATLAB Image Segmentation tool was employed to calculate the effective area of the window and to compare it with the size of the ablation Field of View (Fig. 2H). The actual size was estimated in 1375 µm 2 , resulting in the 85% of the target 1600 µm 2 .

Collection properties characterization
The possibility to collect light with µTF is enabled by the dielectric openings, e.g. the optical windows, on the surface. Let us consider a dielectric TF (Fig. 3A)    The fluorescence signal collected by the window was recorder by the Fiber-PMT in terms of number of photons per pixel (Nf(x,y)), while the µscope-PMT builds an image of the same area, also in this case in terms of number of photons per pixel (Ns(x,y)) 31 . Considering the detection loss of the system, η was therefore determined by the following relation:

Collection efficiency maps
Where Q is a factor that takes into account the detection loss of the epifluorescence collection path and the term ( , ) represents the number of photons emitted by the excited fluorescence spot.
A representative collection efficiency map is displayed in Fig. 3D, showing iso-intensity lines in the x,y plane for different values of η. The maximum η achieved by the single window was estimated by summing all the pixel values with η > 90% of the maximum pixel value and dividing for the corresponding number of pixels, obtaining ηmax = (5.50 ± 0.13) × 10 -3 (mean ± standard deviation, n=3 fibers). We carried out a quantitative comparison between the FA-DLW window and a Focused Ion Beam milled window realized at the same diameter on a different metal coated TF (Fig. 3E) 18 , milling an area equal to the ablation FOV (1600 µm 2 ).
This resulted in the η maps displayed in Fig. 3F and a ηmax = (2.30 ± 0.06) × 10 -3 (mean ± std, n=3 fibers). The integrated intensity detected by FA-DLW spectrum (red) resulted to be ~2.6 times higher than the FIB spectrum (blue), as confirmed by the spectra in Fig. 3G.
To better compare the properties of the FA-DLW and FIB windows, we measured the surface area defined by the isolines at different percentages of ηmax (Fig. 3H). While the high relative efficiency areas (60% and 80% of ηmax) are similar, the low relative efficiency area (10% of ηmax) is 1.5 times wider for the FA-DLW one (3956 µm 2 versus 2676 µm 2 ). If the decay profiles of the normalized η are considered (Fig. 3G), two similar trends can be observed, with the efficiency that drops below the 20% of the maximum after the first 50 µm after the window.
This lets us suggest that FA-DLW windows are better suited to detect optical signals than FIB-milled windows, as the higher collection efficiency of the formers allows gathering a stronger signal from larger sample volumes.

Angle Selective Light Coupling setup and Far-Field Imaging
To identify the transversal propagation constant of modal content excited by light coupled through the window, we have implemented the optical path in Fig. 4 A, B. The patterned TF was placed in a PBS:Fluorescein bath, and a Continuous Wave laser at 473 nm was injected into the fiber with a specific angle to maximize output power from the window. Generated fluorescence was collected by the same window and it was directed through a far-field imaging path by a dichroic mirror.
Light emitted by the fiber facet can be expressed as a weighted sum of plane wave components w(x,y) which propagate at manifold (θxz, θyz) angles (in the scheme in Fig. 4.C, only one plane wave is shown). Passing through lens L3, those components are separated and focused at different points 33 ( , ) ( tan( t , an( ) ) ) xz yz where R(u, v) is the magnitude of the distance vector of the pixel from the center of the resized (u, v) plane, f3, f4 and f5 are respectively the focal lengths of L3, L4 and L5 in Fig.3.B.
Because of the cylindrical symmetry of the modes' propagation into the waveguide, we observed a ring-shaped intensity distribution pattern on the detection sensor.
Representative far-field patterns of light detected from a window placed at a diameter d = 55 µm is displayed in Fig. 5A, after background subtraction to avoid the influence of patch fiber auto-fluorescence on the measurement (the background image was obtained in a nonfluorescent PBS bath). The image was numerically segmented by: (i) a low-pass intensity threshold to remove the residual excitation laser and outliers, and (ii) a high-pass intensity filter to remove residual noise (placed at two times the mean value of the background pixels). A representative segmented image is shown in Fig. 5B. Subsequently we plot the histogram of the kt values related to non-zero pixels from the centroid of the segmented image (Fig. 5C).
The kt interval of the modal subpopulation outcoupled from the patch cord is then extracted from the histogram as the values greater than the 90% of the maximum (red dot in Fig. 5C).
As expected, the collection diameter determines the modal subset at which the collected light propagates in the fiber. From the comparison of far field images, we observed that FIBmilled and FA-DLW milled windows collect modal subsets with compatible kt. Therefore, the FA-DLW window shows higher collection efficiency, while preserving unchanged the modal coupling properties of the waveguide, an important feature for depth-selective fluorescence collection 12 .

Conclusions
We propose an ultrafast Direct Laser Writing setup that takes advantage of real-time spectral feedback to optimize light collection properties of microstructures realized on the edge of a fiber optic-based device. The system was applied to ablate a conformal metal coating on the edge of a tapered fiber, while monitoring the two-photon fluorescence signal simultaneously generated in the environment by the same fs laser pulses. When compared with canonical FIBmilled patterns, a quantitative analysis of the optical properties of collected light in both nearfield (collection efficiency maps) and in far-field (outcoupled modes dispersion analysis) show ~60% improvement in the collected signal, while the uncoupled modal subset is comparable between the fabrication processes.
We attributed these findings to the lower alteration of glass induced from the laser ablation process, in terms of both roughness and dielectric properties. From the SEM micrographs (Fig.   3C, E), we observe no notable texture in the active area of the FA-DLW window. On the other hand, FIB processing shows no textures in the central part, for an extension of ~20 µm in the transversal direction with respect to the TF axis, but a rough surface in the lateral part, where the curvature of the fiber is higher. In addition, the FIB milling on glass is known to induce a variation of its optical properties, with a refractive index rise from n=1.46 to n~1.75 in the visible range, and therefore a transmissivity decreases from T=90% to T=80-85%, due to Ga + ions implantation 37 . Moreover, the feedback offered by the developed system, as well as the possibility to characterize optical apertures' properties using a single setup, sensibly reduce the overall fabrication time with respect to FIB.
As µTFs have proven their potential in depth-resolved photometry experiments 12 , we envision that the presented FA-DLW fabrication of optical apertures on the edge of the TF surface could be employed to engineer the collection sites along the TF itself. Indeed, spatialselective photometry in high-scattering media, such as the brain, could benefit from this type of processing, as it requires high collection efficiency. Although here we showed FA-DLW in the field of optical neural interfaces, the same concepts can be extended to a range of fiberbased biosensors, which would greatly benefit of optimized and selective interactions of guided modes with the environment.

Disclosures
LS, MDV, BS and F. Pisanello are founders and hold private equity in Optogenix, a company that develops, produces and sells technologies to deliver light into the brain. Tapered fibers commercially available from Optogenix were used as tools in the research.