High-spatial density snapshot imaging spectrometer enabled by 2-photon fabricated custom fiber bundles

Abstract

We report on a proof-of-concept snapshot imaging spectrometer developed using an array of optical fibers fabricated with 2-photon polymerization (2PP). The dense input array maps to an output array with engineered void spaces for spectral information. Previously, the development and fabrication of custom fiber arrays for imaging spectrometers have been a complex, time-consuming, and costly process, requiring a semi-manual assembly of commercial components. This work applies an automatic development process based on 2PP additive manufacturing with the Nanoscribe GmbH Quantum X system. The technique allows printing of arbitrary optical quality structures with submicron resolution with less than 5 nm roughness, enabling small core fibers/integrated arrays. Specifically, we developed an array prototype of 40×80 with 6-micron pitch at the input and 80-micron pitch at the output. The air-clad fibers had a core diameter of 5 μm. Fabricated optical fiber arrays were incorporated into a prism-based imaging spectrometer system with 48 spectral channels to demonstrate multi-spectral imaging. Imaging of a USAF target and color printed letter C as well as spectral comparisons to a commercial spectrometer were used to validate the performance of the system. These results clearly demonstrate the functionality and potential applications of the 3D-printed fiber-based snapshot imaging spectrometer.

A snapshot imaging spectrometer is a device that acquires spatial and spectral information in a single image acquisition. Advantages such as high optical throughput and no scanning make it ideal for low light levels or dynamic scenes [1-5]. As one class of snapshot spectral imagers, an integral field spectrograph (IFS) provides direct information with limited post-processing [6,7]. There are three major types of IFS: the image slicing/mapping mirror [8-10], the lenslet array [11,12], and the custom optical fiber bundle [13-16].

In fiber-based spectrometers, the object is imaged onto a fiber bundle’s spatially dense input and is transformed to a spatially sparse output. The output is imaged through a dispersive element where void spaces created by the bundle accommodate the object’s spectral information. Overall, the optical layout (re-imaging system with disperser) is simple and compact. A concept schematic of the principle is shown in Fig. 1.

Fig. 1.

Conceptual schematic of the fiber-based spectrometer with an example of possible fiber input/output.

The formatting of an optical fiber bundle is typically challenging. One common solution is to remap the output end as a single column [17-19], but this significantly limits the number of spatial samples. Also, in prior implementations, fiber-based spectrometers utilized commercial fibers assembled into custom bundles. Due to the limitations of the available components, assembling the fiber bundle for the imaging spectrometer was usually a semi-manual process involving the fiber assembly, cutting, stacking, gluing, and polishing. One example of a bundle allowing high spatial and spectral sampling resulted in a relatively large output area of 20 + mm field diameters [13,14]. This requires large/high-performance/custom optics to accommodate both the large field-of-view (FOV) and fiber numerical aperture (NA), usually greater than 0.25 [14]. As a consequence, the system was relatively expensive and large.

Recent advances in the 3D-printing technology have made it possible to solve many problems associated with fabricating fiber arrays by simply 3D-printing them. The two-photon polymerization technique, which uses a focused laser beam to polymerize a photosensitive material, creates a solid structure layer by layer, enabling submicron resolutions and optical quality components [20]. This simplifies the fiber bundle fabrication process, making it possible to dramatically scale up the number of fibers, while retaining a small form factor with excellent structural integrity. There is also more design freedom, which enables architectures that make more efficient use of the available sensor area. Numerous efforts have been made to 3Dprint fiber bundles with fewer than 150 fibers in applications aimed at light splitting and spectral sensing [21-23]. To obtain practical FOV and resolution, fiber structures need to provide thousands of fibers similar to commercially available imaging bundles. Here, we present a proof-of-concept 3D-printed fiber-based snapshot imaging spectrometer with 3,200 spatial samples (40×80 image format) and 48 spectral channels using the 2PP technique. This work pushes the boundary, that is, 3D-printing thousands of fibers with 2PP and achieving the highest reported spatial sampling to enable practical snapshot imaging spectrometry.

The 3D-printed optical fiber array is designed as a repeated structure that utilizes a single printing FOV. Figure 2 shows the general design strategy. Fibers are made of three segments: two straight segments and one 90° turning segment. This is one of the simplest designs that can achieve a dense input and a sparse output for a large fiber array. One layer of fiber bundles (40×1) is generated in Mathematica [Fig. 2(a)]. Then the layer is duplicated 80 times in DescribeX (native software of Nanoscribe) with the array function [Fig. 2(b)]. A zoomed-in section of the bundle output is shown in Fig. 2(c). Each fiber is designed to be in contact with the walls only at the input and close to the output. Hence, only the length of the fibers differs. There are two advantages of the design in general: reduced background noise and enhanced mechanical stability. The turn reduces the background signal from the bundle input. Figure 2(d) is a section of the bundle output obtained with a bright-field microscope. On the other hand, utilizing a single printing FOV avoids stitching artifacts, which can cause misalignment between layers. Note that stitching errors can reduce the fiber throughput and compromise the mechanical stability of the array. In the future, to increase this area and expand spatial sampling, a custom system calibration process needs to be developed to compensate for stitching. The fiber diameter is set at 5 μm (this value allows 80 fibers in one FOV) and bending radius to be 150 μm, which exceeds the critical radius to ensure no radiation loss [24]. The fibers have a symmetric 6 μm pitch (5 μm core + 1 μm gap) on the input side and 6 μm ($x$) by 80 μm (z, 5 μm core + 75 μm void space) pitch on the output. The output pitch in the $z$ direction determines how much void space can be used for spectral channels. The entire structure has dimensions of 480 μm ($x$)×424 μm ($y$)×3456 μm(z), which falls within the FOV of 495 μm ($x$) ×495 μm ($y$) of the 25X objective (NA = 0.8) used in the Nanoscribe GmbH Quantum X system. The system also utilizes femtosecond light pulses at 780 nm. Both hatching (lateral) and slicing (axial) distances are 0.3 μm with a laser power set to 60mW and a scanning speed set to 120,000 μm/s. The roughness/form of the surface is determined by the hatching and slicing distances, with the trade-off that using smaller values extends the fabrication time. The combination of the laser power and the scanning speed determines the exposure dose. A lower dose results in reduced mechanical strength of the fiber array and raising the risk of structural collapse during post-processing. A parameter sweep for these two parameters is then applied to avoid this issue, which is similar to the methodology presented in [25], also considering the surface quality (roughness/form). Consequently, the printing process took approximately 24 h to complete. The fiber structure is fabricated using proprietary IP-S photoresist by Nanoscribe. Post-processing of the structure consisted of immersion in the SU-8 developer for 20 min followed by 2 min in IPA to wash away the unpolymerized resin. As shown in Figs. 3(a) and 3(b), the miniature size of the fabricated fiber structure is demonstrated by its comparison to a one-cent coin. The SEM image from Fig. 3(c) reveals the output end aligns with the design, but the side view image from Fig. 3(f) presents selected wider layers toward the output area. This becomes more pronounced from the SEM image with a 45° view [Fig. 3(g)] showing the fabrication artifacts along the possible post-processing effects of capillary force, where portions of some fibers are bent/merged with its neighbors and may result in crosstalk effects. Analysis of the surface quality using Zygo NewView 5000 white light interferometer demonstrates the discernible staircase pattern of the straight segment of a single fiber as a result of layer-by-layer printing, which is shown in Fig. 3(d) and 3(e). Each step is measured to be approximately 0.3 μm, which corresponds to the 0.3 μm slicing and hatching distance. The result is leakage of light and potentially may affect the image quality.

Fig. 2.

Design of the 3D-printed fiber array: (a) single -layer STL file generated in Mathematica; (b) whole fiber structure created by DescribeX with the array function; (c) view of a section of the output end of fibers from DescribeX; (d) image of a section of the output end from a bright-field microscope, with illumination into the input end.

Fig. 3.

3D-printed fiber structure with different views and analyses: (a) size comparison of the proof-of-concept 3D-printed fiber array with a one-cent coin using a handheld microscope (Dino Lite); (b) output ends with details; (c) SEM image of a section of output area; (d) surface profile of a single fiber with Zygo interferometer, measurement in (e) shows staircase step to be 0.3 μm; (f) side view of the structure obtained using Primo Star Zeiss Microscope with 2.5X objective, taken over two FOVs to acquire the entire structure; (g) SEM image of side view of a few layers with a 45° tilt.

The proof-of-concept snapshot imaging spectrometer is shown in Fig. 4. It consists of the imaging lens, 2PP fabricated structure, and re-imaging (including disperser) system. First, the target is imaged to the input end of the fiber array (FOV = 240 μm×480 μm) through the microscopic slide by the Nikon 4X finite-conjugated microscopic objective MSB50040 (NA = 0.1, WD = 25 mm). The output is then imaged by a re-imaging system onto a camera. The signal from the output goes through a collimating lens MVPLAPO 1X (Olympus, NA = 0.25, f = 90 mm); a single-band bandpass filter, FF01-535/150-25 (Semrock); a dispersive prism, P-WRCO43 (Ross Optical, BK7, 6-degree deviation angle); and a focusing lens, MVX-TLU (Olympus, f = 180 mm). A bandpass filter is used to select the 460–610 nm band with an average transmission of >93%. The output is reimaged onto the camera with a magnification of 2. The dispersed image is acquired on a PCO edge 5.5 sensor.

Fig. 4.

Prototype of the snapshot imaging spectrometer on the optical bench.

Acquired images need to be remapped to a spectral datacube. A look-up table is created in the calibration process to map spectral/spatial locations of the object. The raw images are spatially and spectrally calibrated, flat-field corrected, and background subtracted to generate multi-spectral images. An advantage of the 3D-printed fiber structure is its regularity. This simplifies the calibration process compared to the semi-manually fabricated bundles reported previously [15]. MATLAB’s image region property functions are applied with a proper bounding box and a threshold setting to find the centroid of the bright pixels on the image with a narrowband filter. Spectral calibration is used to locate all 48 spectral channels by repeating the same steps for three narrowband filters, which is similar to our previous work [11]. The dispersion angle on the sensor is designed to be 64° with respect to the horizontal axis in order to reduce the output pitch and the total height of the structure.

A flat-field image ($\mathrm{F}$) is necessary to compensate for the intensity difference of individual fibers/fiber rows. It is taken by replacing the target with white paper under the same illumination conditions [Fig. 5(a)]. A dark-field image ($\mathrm{D}$), which is captured when the illumination source is covered, is also required for background subtraction. The flat-field corrected image ($\mathrm{C}$) for the scene image ($\mathrm{S}$) is then obtained from the following equation:

$$C=\frac{\mathrm{S}-\mathrm{D}}{\mathrm{F}-\mathrm{D}}.$$ (1)

Fig. 5.

Characterization of the fiber array: (a) flat-field image; (b) fiber input end with unpolymerized IP-S left; (c) scatterplot of relative intensity versus rows of fibers. Error bars represent the standard deviation.

Based on the design of the fiber structure, the length of the fibers increases by 80 μm for each row from the left to right in Fig. 5(a). This corresponds to the intensity decrease of different rows of fibers. Figure 5(c) shows gradual signal loss through a total length of 3767 μm. The overall loss between the shortest and longest fibers is approximately 60%, with the absolute attenuation coefficient of 1.14 dB/mm. These losses are most likely attributed to combinations of roughness, absorption, and dissipation of higher-order modes in these short structures. In addition, the flat-field image is not entirely uniform. One reasonable explanation is that there is residue of un-polymerized IP-S left around the bottom of the structure causing the leakage of light and nonuniform bundle areas for uniform illumination. Figure 5(b) shows the area of non-polymerized resin which could not be washed out in the post-processing from the input end after multiple washes. It is likely due to the limited spacing between fibers, which reduces the liquid flow and prevents the developer from getting into the center of the compact input end. This results in dimmer area on its transition to air.

To evaluate the spectral response of the imaging spectrometer prototype, Roscolux color filters were used as a target. The results were quantitatively compared to the spectrum obtained with an Ocean Optics USB4000 visible light spectrometer. Figure 6(a) presents comparison of these spectral distributions for 465 to 600 nm wavelength range. The majority of Ocean optics spectrometer measurements fall within or close to the shaded regions, where the shaded regions represent the standard deviation across multiple fibers within the imaged area. All graphs were proportionally scaled at the center of spectral range and then normalized across measurements. From the measurement of 1 nm narrowband filters, the FWHM of the 3D-printed imaging spectrometer is significantly larger [Fig. 6(b)] (resulting from lower dispersion and sampling – 48 spectral samples) while matching expected filter wavelengths (488 and 514 nm). Lower spectral resolution/wider FWHM results in less apparent (smoothed out) spectral features in comparison to Ocean Optics measurements (658 spectral channels).

Fig. 6.

Comparison of spectra of Ocean optics and 3D-printed fiber-based imaging spectrometer: (a) measurements of Roscolux 44, 92, and 3202 filters with the shaded region representing the error (standard deviation); (b) measurements of 488 and 514 nm narrowband filters.

Imaging results for a negative 1951 USAF Hi-Resolution Target and the colored letter C are shown in Fig. 7. Both targets are illuminated by a fiber optic illuminator (Leeds 8300) with halogen light bulb of 3300 K (Pro lights, EKE 21 V, 150 W, MR16, Gx5.3 Base) and a diffuser. Figure 7(a) shows the images of USAF group 5 digits as well as element 6, which are imaged with a 4X magnification onto the input end of the fiber array. Compared to the reassembled images, raw images are stretched horizontally by the design of the fiber array. Figure 7(b) shows images of colored letter C, which is imaged with a magnification of 0.25X onto the input. This was done by reversing the imaging system to enable capture of the entire letter in the FOV of a fiber bundle. In spite of the resulting low spatial sampling, this allows us to demonstrate the spectral performance of the system by imaging the letter C, which is composed of a mixture of colors, including purple, blue, green, yellow, and red. The reference color camera image and composite color image obtained with the imaging spectrometer are presented for comparison. The bottom portion of letter C is printed purple, which is difficult to represent accurately in the prototype due to the limitations in the wavelength range of the spectrometer. Nevertheless, by selecting different spectral channels from 465 to 600 nm, the letter intensity clearly varies from bottom to top, providing a clear representation of a color transition.

Fig. 7.

Imaging results of snapshot imaging spectrometer: (a) USAF target showing raw images and reassembled single-channel images of 560nm, with top images showing bars in group 5 element 6 and subsequent images showing digits in group 5; (b) color letter C showing the ground-truth image taken with Dino Lite microscope under the same illumination condition as the imaging spectrometer, the color image after reassembly and single-channel images of the imaging spectrometer showing 24 out of 48 channels with pseudocolor.

In summary, a proof-of-concept 3D-printed fiber-based snapshot imaging spectrometer, with 40×80 spatial samples and 48 spectral channels, has been demonstrated. The results unequivocally showcase the potential of this approach for increasing spectral/spatial sampling (larger datacubes). While not demonstrated, the introduction of the cladding material to control the fiber NA will provide an additional level of control. This will require new structure design strategies, for example, adding external optical epoxy for cladding. The transmission and uniformity of the fiber array also need to be carefully characterized and optimized. The unpolymerized resin left over at the bottom is not desired. We will explore both the design and post-processing methods such as the critical point drying to ensure complete removal of the unpolymerized resin. Efforts should be pursued to enable high spatial and spectral sampling as well as an integrated, compact format, aiming to further decrease the fiber diameter while maintaining the mechanical stability. The gap between the fibers will be adjusted to minimize crosstalk, for the purpose of maintaining spatial resolution. Beam modulation will also be applied to mitigate the staircase pattern and provide a better surface quality. We envision this imaging spectrometer to become platform technology and encompass applications spanning from biomedical imaging through environmental imaging and remote sensing. Since the fiber bundle is on the millimeter scale, we recognize the potential to miniaturize the entire system and incorporate it into portable handheld devices or small unmanned aerial vehicles (UAV).

Data availability.

Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.