Snapshot hyperspectral imaging microscope enabled by cladded waveguide array fabricated with 2-photon additive manufacturing

Abstract

Snapshot spectrometers capture spatial and spectral data in real-time, offering transformative potential in biomedical imaging and environmental monitoring. However, conventional designs struggle to reconcile high spatial-spectral sampling density with device miniaturization. To address this, we present a compact snapshot imaging spectrometer leveraging a densely packed 3D waveguide array fabricated via two-photon polymerization (2PP), an additive manufacturing technique that achieves submicron precision in complex geometries. The design features 26,000 straight waveguides in an angled end-face configuration, with a 4 µm pitch and 2.5 µm core size. By introducing a vertical layer height increment of 32 µm, the array facilitates spectral data distribution across 40 pixels. The compact structure, measuring 852 µm x 552 µm x 4093 µm, offers new opportunities for integrating snapshot spectroscopy into portable devices. The system's performance is evaluated through measurements of spectral resolution, crosstalk, and throughput. Validation using USAF resolution target imaging and the biological microscopic samples demonstrates its potential to deliver accurate and high-efficiency spectral data.

1. Introduction

Snapshot imaging spectroscopy is an analytical technique that has been increasingly recognized across fields such as biomedical imaging [1,2], remote sensing [3] and autonomous driving [4] for its capability in chemical identification using tens to hundreds of spectral channels in a single exposure. Among various direct [5,6] and indirect [7–9] snapshot imaging methods, the integral field spectrograph stands out for directly capturing spatial and spectral data, minimizing computational artifacts, enabling high-throughput data collection, and supporting real-time data acquisition. Over the years, three primary types of integral field spectrographs have been developed: those employing an imaging mapping mirror [5,10], optical fiber bundle [11,12], or lenslet array [13,14]. Of these, the fiber bundle method offers the advantages of a simpler optical system and enhanced compactness [6]. Additionally, with the flexibility to rearrange the fiber input-to-output configuration, it is anticipated to allow tunable adjustments between spectral and spatial sampling in the foreseeable future.

With the high design flexibility offered by optical fiber bundles, extensive efforts have focused on developing fiber-based imaging spectrometers. In these systems, optical fibers enable spatial reformatting to create void space at the output end for spectral information collected at the densely packed input end. Typically, this reformatting involves arranging the output fibers in a single column, with the other dimension dedicated to spectral sampling [15,16]. However, this approach limits spatial sampling. To address this, an advanced technique using single fiber ribbons was developed, achieving approximately 30,000 spatial sampling points [11]. Despite this advancement, several challenges remain, including a laborious fabrication process, low fill factor, reduced throughput, and high optical system requirements. Fabrication is a semi-manual process involving diamond saw cutting, stacking, gluing, and end-face polishing. The repetitive, semi-manual nature of this work introduces single-point failure risks, where any error can necessitate a restart, substantially increasing fabrication time for larger arrays. Additionally, the fill factor at the input end is limited due to a photomask-based core selection process to avoid overlap of dispersion. For larger fiber arrays with outputs reaching 20 + mm in diameter, high-performance, large-diameter optics are required to re-image the fiber bundle onto the camera, supporting a large field of view (FOV) and a high numerical aperture (NA ≈ 0.25).

Given these limitations, two-photon polymerization (2PP) is emerging as a key technique for the next generation of waveguide-based imaging spectrometers. Widely used in micro-optics for applications such as lenses and waveguides with sub-micron resolution, 2PP has also enabled various efforts to 3D-print waveguide arrays for light splitting and spectral sensing [17,18]. However, existing waveguide arrays are typically limited to fewer than 150 waveguides, which restricts their practical utility. While the printing dimensions pose constraints, research has demonstrated that scaling these arrays to hundreds or thousands of waveguides could substantially enhance their applicability in photonic imaging systems [19,20]. Recent advancements have further enhanced the capability of 2PP to create larger, higher-resolution waveguide arrays. A specific mode, two-photon grayscale lithography (2GL), is designed for fabricating 2.5D and 3D structures with varied shapes and heights by modulating laser power during exposure [21]. This modulation produces a gradient in polymerization along the beam's propagation, allowing the voxel size to vary smoothly within a single layer, which enables fine detail even with coarse slicing settings.

In this paper, we present a 3D-printed cladded waveguide array with 26,000 spatial and 40 spectral sampling points for snapshot imaging spectroscopy. Our previous work demonstrated a proof-of-concept array with 3,200 air-clad waveguides [22]. Here, we introduce several advancements: a refined waveguide design utilizing a modular approach as well as incorporating angled end-faces instead of bending radii to avoid optical loss by approximately threefold, enhanced fabrication techniques, reduced production time (from 24 hours to 8 hours), the addition of cladding material, a more uniform output profile, and an eight-fold increase in spatial sampling, enabling high-quality imaging with an expanded FOV. To our knowledge, this array achieves the highest spatial sampling reported to date for 3D-printed waveguide arrays. Tests using USAF resolution targets and biological microscopic samples demonstrate the spectrometer's accuracy and efficiency and its potential in biomedical studies.

2. Design

2.1. Principle of waveguide-based snapshot imaging spectrometer

The waveguide-based snapshot imaging spectrometer captures both spectral and spatial information in a single shot. As illustrated in Fig. 1, the sample’s image is relayed to the densely packed input end of the 3D-printed waveguide structure as an approximately 2:3 aspect ratio rectangle. The waveguides then expand in the z-direction and realign into slanted rows, creating void space for dispersion in the horizontal direction at the output end. This output is re-imaged and then dispersed using an optical system consisting of two lenses configured as an afocal system and a dispersion prism. Filters are also used to define the spectral range. Finally, the camera captures an image that contains both spatial and spectral information. Following spatial and spectral calibration, as well as flat-field correction, a reconstructed image can be obtained, yielding a datacube of (x, y, λ).

Fig. 1.

Schematic of the optical system, illustrating image formation at various points: from the sample through the waveguide input and output ends, to the final dispersed image on the camera. Note that the waveguide structure is depicted out of scale relative to the other optical elements for visibility. Each waveguide has a square cross-section with 2.5 µm sides and the full array consists of 26,000 waveguides. The schematic is not representative of the actual number of waveguide cores, which is significantly higher than shown.

2.2. Design considerations/design space

Although the theoretical upper limit of the datacube size is constrained by the camera’s pixel count, the actual size largely depends on the design and fabrication of the waveguide structure. Key parameters such as waveguide pitch (core-to-core distance), waveguide input-end FOV, and the structure’s height have pivotal roles in defining the datacube dimensions. The pitch is crucial for balancing spatial and spectral content. A smaller pitch is generally advantageous, enabling more cores to be integrated within a given area, thereby reducing the overall structure size and eliminating the need for bulky optics. However, the resolution of the re-imaging system sets a practical lower limit on how small this distance can be. With a fixed core size, increasing the pitch enhances spectral sampling but compromises spatial sampling. Meanwhile, input-end FOV determines the sample area that can be imaged onto the waveguide structure, with magnification adjusted based on the fore-optics configuration. On the other hand, the size of the input-end waveguide structure is constrained by the fabrication limitations of the QuantumX 3D printer, specifically the maximum printable area within a single FOV. Additionally, the height of the waveguide structure imposes another critical constraint. Due to the modular design of the waveguide, which creates void space mainly through increased z-height, the height must fully utilize the longer dimension of the detector. By optimizing the magnification of the re-imaging optics to ensure the pitch is resolvable, the height should be maximized to achieve higher spatial sampling. These design considerations collectively influence the final performance and capabilities of the system.

As shown in Fig. 2, several examples are provided to illustrate the relationship between datacube size, input-end FOV, and pitch, assuming the same waveguide structure parameters as in this study. A smaller pitch and a larger input-end FOV are optimal for achieving a larger datacube. However, for our 3D-printed waveguide structure, the input-end FOV is limited to 0.47 mm² due to fabrication constraints, and the pitch is set to 4 µm due to the resolution limits of the reimaging optics. These parameters yield a final datacube of 1.04 MP (125, 208, 40).

Fig. 2.

Design space for datacube (x, y, λ) distribution as a function of input-end FOV and pitch. The datacube dimensions (125, 208, 40) correspond to the spatial sampling (125 × 208) and spectral sampling (40) achieved with the 3D-printed waveguide structure, resulting in a total data volume of 1.04 megapixels (125 × 208 × 40).

2.3. Waveguide structure design

Typical fiber-based imaging spectrometers utilize a 2D-to-1D dimension reduction fiber bundle, often using a square-to-line geometry [23,24]. However, this approach is limited in spatial sampling, as it often uses fiber bundles with a few thousand fibers. To overcome this limitation, we propose an alternative design based on straight waveguides with angled end-faces, which is commonly used to enhance light coupling efficiency and minimize back reflections, like angle-cut fibers [25,26]. In this design, the angled end-face primarily functions to redirect light toward the re-imaging optical system. Consequently, the overall structure may resemble a triangular form, with straight waveguides and a rectangular input layout extending vertically to create void spaces essential for spectral sampling. Furthermore, the end-face angle provides several advantages compared to traditional bending turns: it enables a straight waveguide design, requires no additional support, reduces power losses from bends and saves space. A summary of the waveguide structure parameters is provided in Table 1. Each waveguide features a square cross-section with 2.5 µm sides. The tilt angle of each waveguide’s end-face is set to 48.6 degrees to align with the optical axis of the subsequent optical system, while accounting for the 7.2-degree structural angle resulting from the triangular form.

Table 1. Waveguide Structure Parameters.

Waveguide size	2.5 µm
Pitch	4 µm
XY Dimension	852 µm × 522 µm
Height	4093 µm
End-face angle	48.6 degree
Structure angle	7.2 degree
Height increment factor	0.8
Datacube	125 × 208 × 40

Spatial sampling was a key consideration during design. Efforts focused on minimizing waveguide size to fit more waveguides within a given printing area while ensuring they could still be adequately resolved by the optical system. Additionally, the printing FOV (852 µm × 522 µm) was expanded beyond the calibrated FOV (495 µm × 495 µm) of the QuantumX 25× objective. Notably, no significant performance loss was observed within the extended printing area. Both of these aspects will be discussed in the later sections. As a result, a spatial sampling array of 125 × 208 waveguides, totaling 26,000 waveguides with 2.5 µm core size and 4 µm pitch, was achieved using a modular approach that facilitates rapid file generation and scalability. To demonstrate the design process, Fig. 3(a) shows a smaller, conceptual array of 20 × 20 = 400 waveguides. First, a set of waveguide layers of 1 × 20 was created. These layers were designed with varying output heights to produce a ramp-like output sequence, as depicted in Fig. 1 and Fig. 3(a). In the conceptual array, this is shown as a 4 × 20 set, consisting of 4 layers of 1 × 20 waveguides, each with slightly different output heights. The full array was then assembled by repeating these sets to complete the waveguide structure (5 repetitions of the 4 × 20 set resulted in the 20 × 20 array). In the actual full-scale array design, the repeatable blocks consisted of 125 × 16 waveguides, which were designed and exported as an STL file using Mathematica. The number of waveguides per row, set at 16, was directly tied to the spectral sampling capabilities. Increasing this number would increase the structure's height, while the spectral sampling itself is also correlated with the ratio of waveguide pitch to core size. To maintain a straight dispersion line on the camera and simplify data processing, the 16 waveguides within each row were progressively increased in height. To create a more compact structure, the height increment factor was fine-tuned to 0.8 of the waveguide size (2.5 µm), resulting in a 2 µm step increase per layer. This adjustment leverages the angled view of the re-imaging optical system, which is oriented perpendicularly to the output end. Since each waveguide appears smaller than its actual height due to the viewing angle, a slight reduction in actual height is achievable. Consequently, the varying output heights of the waveguides—forming a regular pattern in the waveguide array (Fig. 3(a))—correspond to the extruded waveguide segments beyond the slanted wall support. In the actual fabricated array, each row consists of 16 waveguides within a repeatable block, with extruded heights ranging from approximately 4 µm (minimum) to 34 µm (maximum).

Fig. 3.

Modular design of waveguide structure: (a) A demo array of 20 × 20 waveguides illustrating the design process; (b) The final array of 208 × 125 waveguides, showing the input and output ends.

The STL file of a 16 × 125 block was imported into NanoPrintX—Nanoscribe’s software that enables 2GL with 3D design capabilities. Using the array function in NanoPrintX, the block was duplicated 13 times to form the complete array. An STL file of the sidewall, which is used to enclose the liquid-phase IP-S as cladding, was also combined with this array STL to generate the final printing file. The actual fabricated array, shown in Fig. 3(b), presents the input and output ends of the 208 × 125 waveguide structure.

2.4. Spectral imaging system with implemented waveguide array

As shown in Fig. 4, the sample is imaged to the input end of the waveguide structure using a microscope objective. A 4× finite-conjugate objective (MSB50040, NA = 0.1) magnifies the sample by a factor of 4. Based on the waveguide structure angle, the relay imaging system is tilted by 7.2° to ensure alignment of the output end with the subsequent optical system. The collimating lens is selected to fully encompass the field of view (FOV) of the output end (>4093 µm) while maintaining NA matching with the waveguide output. A Fujifilm XF 35 mm f/1.4 photography lens is chosen for its sufficient FOV and ability to preserve the waveguide output’s NA (0.34 > 0.32). An achromatic doublet (AC508-150-A, f = 150 mm, Ø2”) was used as the focusing lens along with the photography lens, achieving a magnification of 4.28. Resolution testing using a USAF target demonstrated that the combination is capable of resolving Group 8 Element 1 (3.9 µm). With the PCO 5.5 Edge CMOS camera, the 16.6 mm image of the waveguide output end precisely fits within the camera's 16.64 mm sensor length. A P-WRC043 disperser (nominal deviation 6°) is used, with a spectral range of 500–650 nm defined by one long-pass (FEL0500, Ø1”) and one short-pass (FES0650, Ø1”) filter. Considering the waveguide structure parameters, spectral range, and disperser, the camera achieves a spectral sampling of 40 spectral channels. The system is validated on a bench setup, with future plans to integrate it into a more compact and automated platform.

Fig. 4.

Waveguide imaging system setup.

3. Fabrication

The waveguide structure was fabricated using the QuantumX two-photon lithography system (Nanoscribe GmbH). The system employs ultra-short 780 nm light pulses focused through a 25×, 0.8 NA objective to polymerize a voxel of photoresin in immersion dip-in laser lithography (DiLL) mode. A list of printing parameters is provided in Table 2. By dynamically modulating the laser exposure dose, 2GL technology enhances 2PP, enabling precise control over voxel size, continuous height variations, and high surface quality, while permitting coarser slicing spacing (1.2 µm) and maintaining fine hatching spacing (0.25 µm). The scanning speed (200 mm/s) and laser power (78 mW), which together determine the exposure dose, are adjusted based on the observed printing results to address issues of under- or over-exposure of the structure. Parameters such as grayscale power exponent, minimum power, and mesh Z offset were set to the preset values optimized for 2GL, primarily to ensure surface quality and shape accuracy. The structure is designed to fit within a single printing FOV, meaning the stage was only moving in the z-direction. This was done to avoid misalignment issues related to stitching, though this approach limits the printable area. While the calibrated FOV for the 25× objective was 495 µm by 495 µm, it has been demonstrated that the actual printing area can be extended beyond this limit. Through careful monitoring of the printing process and validation using SEM images, a final structure size of 852 µm by 522 µm was reliably achieved.

Table 2. Printing Parameters.

Hatching Spacing	0.25 µm
Slicing Spacing	1.2 µm
Scanning speed	200 mm/s
Laser power	78 mW
Grayscale Power Exponent	2
Grayscale Minimum Power	8 mW
Mesh Z offset	2.5 µm

Since the DiLL mode supports printing with only one material, an alternative method was required for the cladding. To address this, we printed the waveguide core using cured IP-S and encapsulated the liquid-phase IP-S to serve as cladding, which also eliminates the need for its removal, particularly in the densely pack input area of the waveguide array. This process involves printing walls around the entire waveguide array. At the output end, as the waveguide's end-face angle diverges from the wall, the final section of the waveguide extends by an average of about 20 µm. Note that no noticeable array performance change was observed over up to 4 months of use and testing.

This configuration affects the NA of each waveguide. In the visible range, the NA was approximately 0.32, determined by the refractive index difference between the core material (1.515, 2-photon polymerized IP-S) and the cladding material (1.48, liquid-phase IP-S). Previously, with air cladding, the NA was 1, leading to challenges in light collection efficiency. This new cladding method produced waveguides with NA similar to commonly achieved in commercial fiber and waveguide structures and allowed better compatibility with optical coupling and collections systems.

To further improve the mechanical strength and adhesion of the waveguide structure, a 10 µm base was added to the STL design and printed first, between the substrate and the waveguide. This base provided additional stability to the lower sections of the structure.

The waveguide structure, approximately 4 mm in height, presented fabrication challenges in maintaining a stable meniscus of the IP-S resin between the objective and the substrate, particularly towards the end of the printing process. In early attempts, the meniscus would break when the objective was about 3 mm from the substrate. To address this issue, a cone-shaped reservoir, illustrated in Fig. 5(a), was designed in SolidWorks and fabricated using polylactic acid (PLA) on an Ultimaker S5 printer. This reservoir prevented the IP-S resin from spilling over the sides while maintaining an opening at the bottom. It ensured consistent contact between the objective and the IP-S resin, even as the objective moved further from the substrate to fabricate the taller sections of the waveguide, as shown in Fig. 5(b).

Fig. 5.

3D printing of the waveguide structure with a cone-shaped reservoir: (a) Design and dimensions of the reservoir in SolidWorks; (b) Image demonstrating the maintained meniscus during printing, captured using the in-chamber camera of the QuantumX system; (c) Completed waveguide structure after printing.

The entire printing process was completed in approximately 8 hours and 30 minutes, with the waveguide structure shown in Fig. 5(c) prior to post-processing. Printing with 2GL significantly reduces fabrication time—for instance, while 3,200 waveguides with 5 µm cores previously required 24 hours to print using 0.3 µm slicing, 2GL technology can fabricate 26,000 waveguides (1.2 µm slicing) in a fraction of the time without compromising precision or quality. Post-processing of the waveguide structure involved immersing the printed sample in SU-8 developer for 20 minutes, followed by a 2-minute rinse in 99.9% isopropyl alcohol (IPA). Figure 6(a) provides a size comparison of the waveguide structure alongside a one-cent coin for scale. A microscopic image captured using a Keyence VHX digital microscope (Fig. 6(b)) confirms the successful fabrication and structural integrity of the waveguide. Scanning Electron Microscope (SEM) images of the waveguide's output end, taken at 530× magnification (Fig. 6(c)) and a further magnified view at 2300× magnification (Fig. 6(d)) using the BMC Phenom Desktop SEM, reveal a highly uniform and precise structure that closely adheres to the design specifications. This method demonstrates a significant advancement in regularity and precision compared to the semi-manual bundling process, highlighting its potential for scalable and efficient fabrication.

Fig. 6.

Fabrication results of the waveguide structure: (a) Image of the structure standing in front of a one-cent coin, comparing the size; (b) Microscopic image of the whole structure captured with a Keyence VHX digital microscope; (c) SEM image of the waveguide output end at 530× magnification; (d) Further magnified SEM view (2300×) of the output end.

4. Characterization

4.1. Crosstalk

To quantify crosstalk between adjacent cores of the waveguide, light was focused onto a 5-μm slit and then the slit was de-magnified using a 10×, NA = 0.25, finite-conjugate objective to the input end of the waveguide. The width of the imaged slit at the waveguide input was determined by diffraction limit (NA of the objective) rather than a result of simple geometric demagnification, with the PSF size similar to that of the waveguide core diameter. In order to distinguish the cores clearly, a reimaging system with 11× magnification was implemented (f = 35 mm photography lens and f = 400 mm achromatic doublet). Crosstalk is assessed using the formula:

$$Crosstalk=\frac{S_{n\ne 0}}{{\sum }_{-n}^{+n}{S}_n}\times 100\mathrm{\%},$$ (1)

where S_n denotes the intensity of the n^th waveguide recorded by the camera, with n = 0 corresponding to the illuminated waveguide. In contrast, $S_{n\ne 0}$ represents the intensity value from adjacent waveguides, where positive values of n refer to the cores positioned above the illuminated one (n = 0), while the negative n values correspond to those positioned below. Due to the slit’s geometry, measurements of crosstalk are confined to one direction, with an expected similar response in the other direction. Given the waveguide's length of approximately 4 mm, cores at three different positions—0.5 mm, 2 mm, and 3.5 mm—were selected to evaluate crosstalk performance for a 4 μm pitch structure, as shown in the images captured by the camera (Fig. 7(a)–(c)). Additionally, cores at the middle length (2 mm) were analyzed to compare structures with identical core sizes but varying pitch (3 μm, 4 μm, and 5 μm), presented in Fig. 7(d)–(f). Successful alignment of the optical system is indicated by illumination of the central core, confirming accurate targeting within one core. For crosstalk analysis, measurements focus on the neighboring cores—specifically, the two upper and two lower cores adjacent to the illuminated core. Cores located further away exhibit light levels below the detection threshold and are therefore not included in the crosstalk evaluation.

Fig. 7.

Images of the output end captured during the crosstalk measurement setup for the 4 μm pitch waveguide with different lengths: (a) 0.5 mm, (b) 2 mm, and (c) 3.5 mm. Additionally, images of the middle section for structures with varying pitches are shown: (d) 3 μm, (e) 4 μm, and (f) 5 μm.

The results, calculated using Eq. (1) and summarized in Fig. 8(a) and 8(b), with numerical values provided in Fig. 8(c) (average values and standard deviations are calculated from three measurements each), reveal two key trends. First, Fig. 8(a) demonstrates that increasing the waveguide height results in higher crosstalk levels. The increased length of 3 mm (from short to long sections) leads to longer light propagation distances, resulting in a crosstalk increase of approximately 5% for the 1st adjacent cores and 2% for the 2nd adjacent cores. Conversely, Fig. 8(b) shows that adopting a larger pitch while maintaining a constant core size effectively reduces crosstalk, as the increased distance between waveguides lowers coupling and reduces mode overlap. For the 1st adjacent cores, the 5-µm-pitch structure exhibits approximately 7% lower crosstalk than the 3-µm-pitch structure. For the 2nd adjacent cores, the crosstalk values for the 5-µm-pitch structure are negligible and thus excluded from the analysis.

Fig. 8.

Bar charts and table illustrating crosstalk measurements for waveguide structures under two distinct configurations: (a) waveguides with a fixed pitch but varying heights (0.5 mm, 2 mm, and 3.5 mm); (b) waveguides with a fixed height but different pitches (3 μm, 4 μm, and 5 μm). Additionally, a table (c) is provided, summarizing the average values and standard deviations (each calculated from three measurements) corresponding to the bar charts.

Note, that larger core-core distance introduces trade-offs, such as a lower fill factor (e.g., 25% with a 5-μm pitch) and reduced overall spatial sampling. For the 4-μm pitch configuration, all crosstalk measurements indicate less than 10% leakage to adjacent cores. Taking these findings into account, the waveguide structure with a 2.5-μm core and 4-μm pitch was identified as the reasonable design compromise for imaging applications.

4.2. Optical attenuation measurement

Using the same reimaging optical system as in the crosstalk measurement setup, but with the slit removed, flat-field images were captured. These images are presented in Fig. 9(a), (b), and (c) for the waveguide structures with a 2.5 μm core and 4 μm pitch at slicing distances of 1.2 μm (a) and 0.4 μm (b), respectively, and for the structure with a 4 μm core and 6 μm pitch (c). Due to the limited FOV of the camera (of the testing system), which captures only a portion of the flat-field image, three images were combined to analyze the optical attenuation across the entire structure’s length.

Fig. 9.

Flat-field images of waveguide structures: (a) and (b) show the structure with a 2.5 μm core and 4 μm pitch at slicing distances of 1.2 μm and 0.4 μm, respectively, while (c) depicts the structure with a 4 μm core and 6 μm pitch. The corresponding normalized intensity profiles along the waveguide length are plotted for (d) and (e) the 2.5 μm core and 4 μm pitch structure at different slicing distances, and (f) the 4 μm core and 6 μm pitch structure.

For the 2.5 μm core structure, the total number of rows is 125, with each row corresponding to a progressively longer waveguide segment. The normalized intensity as a function of length is plotted in Fig. 9(d) and (e). Similarly, Fig. 9(f) shows the normalized intensity vs. length for the 4 μm core structure.

The optical loss of the 2.5 μm core structure, measured over a total length of 4.05 mm (125 rows), is approximately -0.352 dB/mm, which is primarily attributed to the inherent absorption of the IP-S photoresin. To evaluate the impact of slicing distance under 2GL mode on the waveguide performance, two slicing distances—0.4 μm (fine) and 1.2 μm (coarse)—were used during fabrication. These results demonstrate that the coarse slicing distance leads to minimal increases in optical loss, with only a difference of about 3% compared to the fine slicing distance. However, the coarse slicing approach reduces fabrication time by two-thirds, making it a more efficient option. In comparison, the 4 μm core structure shows an optical loss of about -0.369 dB/mm over a length of 4.12 mm (81 rows). This suggests that core size has not significant affected optical loss. On the other hand, our proof-of-concept design with a 5 μm air-cladded core demonstrated a significantly higher optical loss of -1.14 dB/mm. The reduced optical loss in cladded waveguides demonstrates the crucial role of liquid-phase IP-S cladding, which minimizes interfacial scattering compared to air-clad structures and enhances light confinement stability, despite the cladding material's inherent absorption.

4.3. Spectral resolution

Three narrowband filters with a bandwidth of 1 nm (centered at 514.5 nm, 532 nm, and 632.8 nm) were employed to evaluate the spectral resolution of the 3D-printed waveguide-based imaging spectrometer (Fig. 10(a)). The full width at half maximum (FWHM) was determined across the field of view (FOV) of the waveguide structure by fitting the measured data to a Gaussian profile. For comparison, a commercial spectrometer (Ocean Optics USB4000) was used to benchmark the resolution. Our measurements revealed that the spectral resolution increases at longer wavelengths, a result attributed to the non-linear dispersion introduced by the BK7 prism. Despite this wavelength-dependent variation, the measured resolution values align closely with the theoretical average resolution of 7.5 nm (derived from the Nyquist criterion: 150 nm spectral range divided by 20 resolvable features), validating the system’s performance.

Fig. 10.

Measurements of spectral resolution and spatial resolution of the 3D-printed waveguide-based imaging spectrometer: (a) Spectral resolution test conducted using narrowband filters at 514.5 nm, 532 nm, and 632.8 nm, with results compared to those obtained from an Ocean Optics visible light spectrometer; (b) Spatial resolution test performed using a USAF resolution target, demonstrating resolution capabilities up to group 7 element 6 (4.38 μm).

4.4. Spatial resolution

The theoretical resolution at the input end, determined by a pitch of 4 μm, corresponds to a spatial resolution of 125 LP/mm. In practice, the resolution is influenced by the front optics. For instance, when a sample is imaged onto the waveguide input using a 4×, NA = 0.1 finite conjugate objective, the spatial resolution of the optical system is determined by the Rayleigh criterion (defined as 0.61 λ / NA). With a wavelength of 600 nm and NA of 0.1, the theoretical spatial resolution is calculated as 3.66 µm. The waveguide structure, with a 4 μm core-to-core spacing and 4x magnification, imposes a spatial resolution limit of 1 μm in the object plane. Therefore, the overall spatial resolution of the system is ultimately determined by the objective, resulting in a final resolution of 3.66 μm. This is supported by the experimental results shown in Fig. 10(b), where the system resolves group 7 element 6 of the USAF target, corresponding to a measured resolution of 4.38 μm. The theoretical resolution of 3.66 μm corresponds to Group 8 Element 1, which is one element beyond the measured value. The slight discrepancy between the measured resolution (4.38 µm) and the theoretical value (3.66 µm) may be attributed to a combination of practical factors inherent to the experimental system. First, waveguide crosstalk, though measured at less than 10% (Section 4.1), introduces localized intensity leakage between adjacent cores, broadening the point spread function (PSF) and reducing spatial contrast. This effect is more pronounced in longer waveguides due to increased light propagation distance from previous crosstalk measurement, leading to worse spatial contrast on the right side of the image compared to the left. Additionally, alignment tolerances in the re-imaging optics or detector placement can introduce sub-micron positional errors, further degrading resolution. While the measured resolution exceeds the theoretical limit by approximately 20%, this discrepancy aligns with expectations for a system integrating multiple subsystems with inherent tolerances.

4.5. Imaging result

The waveguide-based imaging spectrometer was employed for imaging of microscopic slides, capturing a datacube with dimensions of 125 × 208 × 40 (x, y, λ). Following established protocols from previous work [22], raw images are processed via spatial / spectral calibration and flat-field correction to reconstruct the datacube. A look-up table is created during calibration to map spectral and spatial locations of the object. For spatial calibration, waveguide centroids are identified using MATLAB's regionprops function on thresholded narrowband-filtered images, generating a precise grid of (x,y) coordinates. Spectral calibration involves applying centroid detection to three reference wavelengths (514.5 nm, 532 nm, and 632.8 nm), followed by fitting a second-order polynomial—matching the second-order Sellmeier dispersion behavior of the BK7 glass prism—to map 40 spectral channels across the 500–650 nm range. The 3D-printed waveguide array’s regularity simplifies this process compared to manual fiber bundles [11], eliminating positional variability. Flat-field correction (using uniform white-light images) and background subtraction normalize intensity variations, ensuring datacube fidelity.

To demonstrate the system's capabilities, imaging results of a pine stem are presented (Fig. 11). The spectrometer achieves microscopic-level spatial resolution while simultaneously delivering detailed spectral information across 40 spectral channels, 18 of which are displayed here. Specific spectral features are clearly resolved, such as the upper-right region (marked with blue arrow) of the ground truth image, which exhibits a blue hue. In the single-channel images obtained from the spectrometer, this region progressively darkens as the wavelength increases, underscoring the system's ability to accurately capture and resolve spectral variations. Quantitative analysis of 10 pixels in this region reveals a consistent decrease in normalized intensity from 0.7 to 0.35 (a 50% reduction) across this wavelength range. Conversely, the red-highlighted region in Fig. 11 exhibits the opposite trend, with normalized intensity increasing from 0.42 to 0.74 (a 76% increase) as wavelength increases.

Fig. 11.

Hyperspectral imaging of a pine stem microscopic slide, showing the ground-truth RGB image and single-channel images (18 of 40 spectral channels). Intensity plots compare two small regions with distinct colors, displaying mean values ± standard deviation (taken from 10 pixels).

To further demonstrate the biological imaging capabilities of our system, we applied it to hematoxylin and eosin (H&E)-stained human skin and hair follicle sections. H&E's characteristic spectral signatures create an ideal scenario for hyperspectral system evaluation. As shown in Fig. 12, our system spectrally resolves these spatially co-localized stains, with the three distinct components (nuclei, stromal regions, and illumination reference areas) exhibiting characteristic spectral profiles that enable precise separation beyond the capabilities of conventional RGB imaging.

Fig. 12.

Representative spectral channels (18 of 40 shown) with corresponding characteristic spectra of three identified components: hematoxylin-stained nuclei, low-signal stromal regions, and illumination reference zones. Spectral profiles display mean intensity values ± standard deviation (error bars), derived from averaging over 10-pixel regions per component. An RGB reference image is provided for comparison, which captures a representative FOV relative to the single-channel images.

5. Discussion and conclusion

We have successfully developed a high-density waveguide-based snapshot imaging spectrometer, achieving spatial sampling of 26,000 and spectral sampling of 40. This advanced system integrates a modular design approach, angled end-face waveguides with 2.5 μm core size, and cladded waveguide structures, to overcome key limitations of traditional fiber bundle-based spectrometers. Utilizing 2PP combined with 2GL, we fabricated a 26,000-waveguide structure (to the best of our knowledge the highest demonstrated spatial sampling of 3D-printed waveguide array) and within a printing time of approximately 8 hours. The angled end-face configuration eliminates the need for bending radii, simplifying design and printing process, and minimizing optical loss while allowing a compact design.

A critical advancement in this work is the implementation of a cladded structure using liquid-phase IP-S encapsulation. This approach reduces optical loss compared to air-clad waveguides, and lowering the NA from 1.0 to 0.32. This enhancement results in easier/more efficient light coupling (into the waveguide array) and collection by reimaging optical system (from array output). Future work will explore varying exposure doses or laser powers during printing to achieve different refractive indices for the cladding, enabling precise control over the NA and fully polymerized structure.

This work demonstrated a bench-top microscopic imaging system, exemplified by microscopic spectral imaging results. To further advance the presented modality, a design with a more uniform aspect ratio of the output is needed to efficiently increase spatial and spectral sampling. This will require a different design and fabrication approach, and exceeding limits of structure height and print area. For biological applications like cell signaling, blood oxygenation monitoring or wound healing assessment, these improvements would enable more precise spatial mapping of spectral features in dynamic systems. In addition, a more compact and integrated optical system needs to be implemented to allow light efficient, integrated imaging spectrometers, which will enable practical and user-friendly devices, and expanding applications in microscopy, biomedical diagnostics, including real-time tissue oxygenation mapping and wound healing progression monitoring, but also remote sensing, and general-purpose portable spectroscopy. We believe that with ongoing improvements, this technology has the potential to produce compact, relatively low-cost snapshot spectrometers and open new possibilities for real-time, high-resolution spectral analysis across diverse fields.

Funding

National Institutes of Health 10.13039/100000002 ( R21 EB033160, R21 EY035108).

Disclosures

Dr. Tomasz Tkaczyk has financial interests in Attoris LLC focusing on applications and commercialization of hyperspectral imaging technologies.

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.