The integral field spectrograph (IFS) is an indispensable analytical tool for astronomical exploration. It enables the acquisition of data cubes in both spatial XY and wavelength directions, providing the spectrum of each spaxel within a two-dimensional (2D) field^[1,2]. Unlike conventional scanning spectrometers, an IFS captures the spectra of a three-dimensional (3D) data cube in a single integration by reassembling the telescope’s 2D image into a slit configuration^[3]. An IFS typically comprises two parts: the integral field unit (IFU) and the spectrograph. The IFU serves as an intermediary device, facilitating the optical connection between the telescope and spectrograph by transforming a rectangular field into a quasi-continuous pseudo-slit field positioned at the entrance of the spectrometer within its focal plane^[4]. Depending on the method used to reassemble the IFU’s image, there are three major IFU configurations, including the lenslet array, the fiber array, and the image slicer, as illustrated in Fig. 1^[5]. The image slicer method, compared to the other two technical approaches, enables the spectrometer’s detector to fully capture and utilize the spectrum by utilizing each pixel of the detector^[6]. Moreover, it ensures that the 2D spatial direction sampling and dispersion direction remain independent without any interference. Current instruments employing this approach include MUSE (VLT)^[7], UIST (UKIRT)^[8,9], NIRSpec (JWST)^[10–12], and GNIRS (Gemini Observatory)^[13].

The production of an image-slicer IFU involves two primary manufacturing methods. One method adopts the classical polishing approach on Zerodur to manufacture the slicers, which are subsequently assembled into a single image slicer through molecular adhesion of adjacent slices. The other method uses metals with monolithic or segmented optical elements, employing state-of-the-art diamond turning machines^[14]. The latter method typically yields inferior surface roughness quality or transition area smoothness between adjacent channels compared with the former, accounting for its predominant use in the field of infrared imaging, where the requirements are not as stringent^[15–18]. Image slicers used for the visible band typically require the classical polishing method. Conventional slice manufacture necessitates individual polishing of each slice. The current alternative approach involves employing the conventional polishing method to simultaneously polish all slices by aligning the displaced molecules in a stacked manner^[19]. Compared to single-piece production, step stacking offers enhanced uniformity. However, achieving parallelism in this technique requires exceptionally high optical and mechanical precision between slices^[20]. Achieving optical and mechanical alignment for ultra-thin glass slices presents a significant challenge.

The China Space Station Telescope (CSST) is equipped with five first-generation instruments, including a survey camera, a terahertz receiver, a multichannel imager, an IFS, and a cool planet imaging coronagraph. The IFS represents the first of its kind ever launched into space by China. The IFS incorporates a 32-channel image slicer as its IFU. The image slicer addressed in this paper was developed for the CSST-IFS by Tongji University to meet its specifications and environmental test requirements.

After analyzing the two methods used for bonding-based image slicer fabrication in this section, we propose an improved method for image slicer manufacturing. This novel method involves stacking ultra-thin slices in a staggered manner, followed by simultaneous polishing of all slices using traditional techniques. The polished slices are then reverse-stacked using molecular bonding techniques. The stacked blanks are selectively cut with a slicer to obtain the area with the least stress, which will serve as the working region. To assess the feasibility of this method, the Ritz method was used to calculate the energy changes during the molecular bonding process, providing technical guidance for the stacking procedure. The blade crack propagation method was employed to calculate the impurity tolerance during the stacking process. Besides, errors influencing the orientation of the slices were analyzed and allocated. The implementation of this fabrication method enhances the quality of molecular bonding, minimizes damage to the working area, ensures the precision of opto-mechanical assembly, and ultimately enables the successful assembly of a 32-segment image slicer. Consequently, the manufactured 32-segment image slicer effectively fulfills the spectral splitting requirements of the CSST-IFS. This image slicer has achieved the highest precision currently achievable in domestic manufacturing.

The IFS of CSST is designed in a conventional three-stage configuration, consisting of a fore-optical system, an IFU, and a spectrometer. The optical design is depicted in Fig. 2. Light is initially directed into a two-reflection relay optical path with a magnification factor of 9.26, originating from the focal plane of the optical telescope. Subsequently, it undergoes further magnification of 9.26 at the image plane before being incident on the image slicer. Currently, the image slicer’s width measures 0.25 mm, corresponding to a spatial resolution of 0.2″. The installation of reflective mirrors at corresponding exit pupil positions functions as a slicer that projects 2D spatial images into different spatial directions. The slicer is composed of individual mirrors, which, in conjunction with the rear optical path, serve to reduce the image size. Ultimately, a resolution of 0.1″, corresponding to 15 µm pixels, is achieved on the detector. A long slit is formed by the reflection of light through different lens filters, which is then split into two paths using color filters and directed separately into two spectrometers to generate spectral images, enabling comprehensive imaging of each point on the 2D plane. The key specifications of the image slicer are summarized in Table 1.

The Laboratoire d’Astrophysique de Marseille (LAM) in France proposed a staggered stacking method in 2006 to reduce the manufacturing cost of image slicers, as shown in Fig. 3^[19].

The proposed approach involves arranging N slices within a framework, with the reference point positioned on the left side. Each slice is assigned a specific length corresponding to the desired offset at its end. Once the stacking process is completed (which typically takes only a few hours), the entire stack undergoes polishing (as shown in the middle of Fig. 3). Subsequently, each slice is pushed towards another reference point, accounting for their varying lengths and resulting in distinct offsets between adjacent slices.

The proposed method presents several inherent issues: I.determining the bonding mechanism between slices during the stacking process: adhesive versus intermolecular bonding;II.mitigating the deformation caused by cumulative stacking stress during slicing, which leads to alterations in the shape of the polished surface;III.establishing synchronization with the polishing frame upon completion of the slice stacking.

The LAM in France employs a stack of three metal cylinders as alignment aids during the final assembly of the slicer, employing edge pattern monitoring to ensure precise alignment^[20]. Furthermore, surface shape measurements are conducted after each molecular bonding process to promptly address any issues related to excessive stress. The advantages of the two methods have been summarized in this paper. To enhance accuracy while reducing costs, we employed the following approach to construct a slicer comprising 32 slice surfaces. I.The slice material is composed of Schott D263T glass, selected for its low surface roughness, exceptional flexibility, high thermal expansion coefficient, and cost-effectiveness. The ultra-thin glass sheet is precisely cut into nearly flawless cubes of varying lengths through laser cutting, as illustrated in Fig. 4(a).II.The glass sheets of varying lengths are stacked using the fixture depicted in Fig. 4(b) and bonded through molecular adhesion. The stacking process results in the formation of two groups, each consisting of 16 sheets.III.The step stacking blocks are filled with rectangular ultra-thin glass sheets of the same model but varying lengths, as illustrated in Fig. 4(c). Besides, hemispherical protection blocks are adhered to both the upper and lower surfaces. This approach effectively prevents uneven pressure distribution during grinding and polishing, thereby minimizing the risk of bonding failure. Furthermore, the center of a sphere can be determined through its outer circle, enabling accurate positioning.IV.The stacked assembly is ground and polished on its side to enhance surface form accuracy and roughness using conventional spherical manufacturing techniques, as illustrated in Fig. 4(d).V.The pre-fabricated spherical stacking blocks undergo disassembly using heat. The stacked layers, held together by molecular bonding, are systematically dismantled via stress-induced deformation and sequentially labeled.VI.As shown in Fig. 4(e), the reverse stacking of 16 layers of sub-cutting surfaces above and below is achieved through molecular bonding.VII.The two groups of stacked sub-cutting surfaces are reconfigured in accordance with the specifications outlined in the blueprint.VIII.The two groups of sub-cutting surfaces, as depicted in Fig. 4(f), are meticulously assembled and secured using UV glue.IX.A surface coating is applied to achieve the desired reflectance across the ultraviolet to infrared wavelengths, in accordance with the specified requirements.

During the step stacking of glass sheets, microscopic impurities may be introduced at the edges of the slices due to wiping. Furthermore, nanoscale marks may result from the wiping solvent. Imperfect bonding quality may occur during the molecular bonding process. To address the aforementioned issues, wiping and rebonding can be performed after disassembly. However, repeated disassembly and bonding may cause inevitable edge damage. To balance the risk and bonding quality, it is necessary to accept bonding imperfections in higher-risk processes. To evaluate the tolerance for imperfections, it is necessary to analyze the molecular bonding conditions affected by these imperfections.

Based on the wafer direct bonding conditions proposed by Yu and Suo^[21], in order for two slices of the image slicer to bond at room temperature, the following conditions must be met: U−Γ·S<0,(1)where Γ=γ1+γ2−γ12, γ1 and γ2 denote the surface energies of different slices, and γ12 represents the interfacial energy after slice bonding. Γ represents the adsorption energy, which denotes the energy required per unit bonded area during the slice bonding process. The value of Γ can be measured using the blade crack propagation method^[22]. U represents the accumulated elastic strain energy after slice bonding. S denotes the bonded area of the slices. For optimal bonding, the total elastic strain energy after bonding must be less than the reduction in interfacial energy between the two slices. The elastic strain energy is not only related to the material properties of the bonded slices but also to their surface shape. Due to the ultra-thin nature of the slices, the primary surface shape error is related to power. The primary change before and after bonding arises from this power difference. The elastic strain energy can be calculated based on the change in power before and after bonding.

For a slice, the strain energy U generated by bending during the bonding process can be expressed as U=12∫AD[(∇2w)2]dA,(2)where D=Eh312(1−ν2) denotes the bending stiffness of the thin plate, E represents the Young’s modulus, ν is the Poisson’s ratio, and h denotes the thickness of the thin slice. A represents the area of the thin plate. w(x,y) is the deflection function that represents the displacement of the slice due to bending. ∇2w is the Laplace operator applied to the deflection function w: U1=D12∫A(1R1−1Rt)2dA,(3)U2=D22∫A(1R2−1Rt)2dA,(4)where D1 and D2 are the bending stiffnesses of slices 1 and 2, respectively. The optimal fitted radii of power for slice 1 and slice 2 prior to molecular bonding are denoted as R1 and R2, respectively. After molecular bonding, the radius of curvature for the two bonded slices is denoted as Rt: Utotal=U1+U2,(5)Utotal=EA24(1−ν2)[h13(1Rt−1R1)2+h23(1Rt−1R2)2].(6)

To minimize the total elastic energy Utotal, we apply the Ritz method by determining the value of Rt that minimizes the total energy, achieved by solving the following equation: ∂Utotal∂Rt=0.(7)

The simplified result is 1Rt=D1R1+D2R2D1+D2.(8)

Substituting Eq. (8) into Eq. (6) and simplifying yields Utotal=EA24(1−ν2)·h13h23h13+h23(1R1−1R2)2.(9)

Let h1+h2=1 and h1h2=t. Substituting these into the equation yields Utotal=EA24(1−ν2)·t3(1+t)3(t3+1)(1R1−1R2)2.(10)

Since the surface shapes of the upper and lower surfaces of the slice surface remain constant, their corresponding functions are plotted in Fig. 5.

The curve in Fig. 5 illustrates that during the molecular bonding process, an increased thickness difference between the two slices lowers surface elastic energy, thereby promoting bonding. Therefore, sequentially stacking the 32 slices one at a time would facilitate the molecular bonding process. The figure further confirms that a larger thickness difference between two slices reduces surface elastic energy, further facilitating bonding. Thus, a stepwise assembly approach is optimal for bonding the 32 slices^[23].

The expression for calculating the adsorption energy using the blade crack propagation method is Γ=3tb2E1tw13E2tw2316L4(E1tw13+E2tw23).(11)

As shown in Fig. 6, tb is the thickness of the blade, W denotes the width of the bonded wafer, tw1 represents the thickness of bonded slice 1, tw2 denotes the thickness of bonded slice 2, E1 represents the Young’s modulus of slice 1, and E2 represents the Young’s modulus of slice 2. L represents the length of the crack after molecular bonding. The parameters required for calculating the adsorption energy are shown in Table 2.

Using the parameters in Table 2 and Eq. (11), the adsorption energy between the two slices is given by Γ=0.06396J.(12)

The peak to valley (PV) surface shape precision of the upper and lower surfaces of the slices manufactured by traditional double-sided polishing is 1/8λ. The binding energy is calculated using the parameters in Table 2 and Eq. (11). The binding energy for stacking on substrates with different thicknesses is shown by the black line in Fig. 7. By applying Eqs. (1) and (6), the maximum PV value for molecular bonding on substrates of different thicknesses is obtained. The corresponding maximum impurity particle size calculated from the PV value is shown by the red line in Fig. 7. Notably, the binding energy for molecular bonding between two slices, and the tolerance for impurity particle size are the lowest. The maximum impurity particle size is 0.1035 mm, which is essentially consistent with the limit of human eye resolution. This suggests that no impurities should be visible to the human eye on the slices before bonding. At the same time, due to the continuous stacking process, PV errors caused by impurities accumulate, which imposes higher requirements on the wiping process before molecular bonding to ensure minimal contamination.

The adoption of a stacking-based approach for image slicer production offers the potential to optimize cost-efficiency, streamline production cycles, and enhance iteration efficiency. However, the protracted manufacturing process still presents formidable challenges. Achieving high-precision manufacturing necessitates error allocation throughout the production process, primarily encompassing fabrication errors in ultra-thin slices, stacking errors, and surface shape errors.

For thin slices in a cubic configuration, the primary error source stems from the flatness of its six faces and the angles between each face pair. However, due to the reduced thickness of sub-cutting units, they exhibit reduced sensitivity towards errors in the stacking and assembly process along the thickness direction. Consequently, we simplify this error by distributing it between standard 2D rectangular ultra-thin slices and errors in surface shape on both sides. As depicted in Fig. 8, precise control of angles A1 and A2 is essential in the production of a 2D standard rectangle to ensure the perpendicularity of sides a, b, and c while maintaining their straightness. The discrepancy between angles A1 and A2 directly influences the relative positioning of the final focal points. The impact values can be calculated using the following formulas^[24]: ΔXA1=R×tanΔA1(13)and ΔXA2=R×tanΔA2.(14)

The influence value of the straightness error on side b(ΔSb) is given by ΔXb=R×ΔSbLBC.(15)

The length influence on side b remains unchanged as ΔLb=ΔLb. During the cutting process, length errors can be maintained below 1 µm, and stacking effects are negligible and will not be further discussed. It is essential to allocate errors for angle deviations (ΔA1, ΔA2), as well as for straightness deviation (ΔSb).

The stacking error consists primarily of two components: fixture error and slices stacking placement error. As shown in Fig. 9, the fixture is aligned using a three-point method. Ideally, the angle between columns A, B, C, and the base plate should be 90°, with column A in contact with edge a of the ultra-thin sheet while columns B and C are in contact with edge b of the ultra-thin sheet. The angular deviation ΔA3 in the X direction between the A-pillar and the base affects the focal point position, which is given by ΔXA3=h×tanΔA3.(16)

In Eq. (15), LBC represents the distance between columns B and C, while the angular deviations ΔA4 and ΔA5 of columns B and C with respect to the base in the Z direction have an impact on the focal point, which is given by ΔXA4=R×tan(ΔA4−ΔA5)LBC.(17)In Eq. (16), h denotes the thickness of the ultra-thin slices, which is 0.25 mm. The angular error ΔA3 has a relatively minor impact on the stacking process. Besides, the diameter errors of columns A, B, and C influence the stacking process. Specifically, the diameter error ΔDA of column A affects the focal point as ΔXDA=ΔDA. In contrast, the combined effects of diameter errors ΔDB and ΔDC in columns B and C, respectively, contribute to changes in focal point position according to ΔXDBC=R×ΔDB−ΔDCLBC.(18)

Notably, compared to other factors, the diameter error of column A exerts a relatively smaller impact on the focal point. Furthermore, errors in parameters such as ΔA4, ΔA5, ΔDB, and ΔDC are distributed between columns B and C.

To ensure precise stacking of the slices, it is imperative that the ultra-thin slices and columns A, B, and C exhibit no stress contact and are meticulously aligned during the stacking process. The placement of the slices introduces three attitude angle errors: pitch, yaw, and roll. However, pitch and roll errors can be disregarded due to the implementation of molecular bonding. The yaw angular error ΔA6 significantly impacts the focal point as denoted by ΔXA6=R×tanΔA6.(19)

The surface shape error can be categorized into low-frequency surface shape accuracy error and high-frequency roughness error. The low-frequency surface shape accuracy error is quantified by PV and root mean square (RMS). The high-frequency roughness is characterized by root mean square deviation (Rq). During the fabrication process of the image slicer, there are two stacking procedures involved. First, rectangular glass sheets are stacked in a step-like manner, and a spherical tool is used to create a ball-shaped surface. Then, the finished sub-slicing surfaces are stacked to form the image slicer. These two stacking processes induce varying levels of stress, resulting in changes to low-frequency surface shape accuracy errors before and after the second stacking procedure. High-frequency roughness is less influenced by macroscopic micro-deformation factors.

Each slice has a tolerance of ±1′, corresponding to an error of ±0.099μm between adjacent focal points. Consequently, error allocation is necessary for sensitive errors, as shown in Table 3. The impact of individual errors is limited to specific channels, whereas cumulative errors have a widespread effect across all channels.

The tilt of each sub-cutting surface is measured with parallel light imaging, as shown in Fig. 10. The parallel light passes through a beam splitter and enters the image slicer, where the reflected light is focused onto the CCD surface after being reflected by the beam splitter. Each slice’s low-frequency surface accuracy is measured using a four-dimensional (4D) interferometer. Furthermore, a ZYGO white-light interferometer is used to inspect the working surface profile of the image slicer for edge collapse and defects.

The focus position obtained by the CCD is illustrated in Fig. 11. The pixel-based positional data are then converted and used to calculate the inter-pixel distance. All design requirements, except for the X-direction spacing between channel B1 and channel B2 exceeding the error threshold, are met satisfactorily. The angular deviation between channel B1 and channel B2 was measured to be 1.34′, exceeding the designed tolerance of 1′. Upon inspecting the sub-cutting surface of B1, damage is detected at its contact point with column C, caused by misalignment during the initial stacking disassembly.

The slice is detected using a 4D interferometer, as shown in Fig. 12. The interferometer’s sampling CCD has a resolution of 2 K. Besides, surface shape tests are conducted before and after dynamic thermal testing to verify the stability of the molecular assembly. All subsections’ surface shapes meet the design requirements, with most low-frequency surface shapes achieving accuracy better than 1/200λ(λ=632.8nm).

The surface profile is next examined using a Zygo white-light interferometer, as shown in Fig. 13, to verify the presence of defects such as edge collapse and damage. Besides, the surface quality is assessed both before and after conducting the force-thermal experiment. Numerous defects are present on both sides of the image slicer due to fragmentation during cutting with a cutting machine. However, no significant defects are observed within the central 8 mm working area, with no gap between slice surfaces. After the force-thermal experiment, no increase in edge defects is found; however, shadow areas are noted between the upper group of 16 channels and the lower group of 16 channels. Due to the sharp height variation at the junction of the two channel groups, the coating layer at the junction is suspended and detached after the vibration test. Further improvements to the coating process are required to prevent the coating layer at the junction from detaching under the influence of vibrations.

In this paper, an advanced methodology is presented for manufacturing image slicers, summarizing existing fabrication techniques. The proposed approach utilizes a push-and-stack method for image slicer production while mitigating stress through cutting. Using elastic thin-plate theory and the Ritz method, the relationship between slice PV and elastic surface energy after molecular bonding is calculated, leading to the theoretical conclusion that layer-by-layer stacking is more efficient than multi-layer bonding. By inversely calculating the maximum impurity tolerance during the stacking process based on molecular bonding criteria, the molecular bonding process for 32 slices is theoretically optimized. An analysis of the production process was conducted to determine error distribution.

Using this methodology, a miniature IFU image slicer with 32 slices is successfully produced for the CSST-IFS prototype. A parallel light incidence detection technique is employed to measure subtangent plane attitude, and all design requirements are met, except for out-of-tolerance issues caused by disassembly damage in the B1 channel. The surface roughness accuracy of high-frequency shape falls below the required Rq value of 0.8 nm, and the low-frequency shape accuracy RMS is significantly less than 1/40λ(λ=632.8nm). Furthermore, no defects or new defects are observed before or after dynamic thermal testing within the working area, as verified by Zygo white-light profilometry. However, partial film removal occurs at the junction of the upper and lower groups due to membrane suspension during the mechanothermal test. To address this issue, future disassembly processes will include strengthening procedures such as separation using feelers and chamfering on the three sides in contact with the support body to prevent further damage.

The successful fabrication and testing of the image slicer prototype highlight the reliability of the proposed methodology and its potential to advance next-generation spectroscopic systems, such as the CSST-IFS. Future work will focus on enhancing the robustness of molecular bonding, further optimizing stress tolerance under extreme conditions, and extending this method to more complex image slicer designs. This research provides a solid foundation for improving IFU manufacturing precision, addressing key challenges in integral field spectroscopy, and paving the way for broader scientific and industrial applications.