Micromirror fabrication for co-packaged optics using 3D nanoimprint technology

These days, demand for cloud computing applications is rapidly expanding and can cause increased data center traffic. Global data center IP traffic is assumed to be growing at a rate of 25% per year and traffic inside the data center is assumed to account for 71.5% of total data center traffic according to Cisco's forecast.¹ The bandwidth of application-specific integrated circuits, which process electrical signals in data center networks, is also increasing yearly,² consuming 25.6 Tb/s in 2019.³ The optical connection in a data center is based on a pluggable module, where electrical signals from the ethernet switch are transferred and converted to optical signals on the edge of a printed circuit board by a pluggable transceiver. The switching bandwidth is expected to increase to 51.2 Tb/s in the near future, and the limitation of transmission capacity due to power consumption of the electrical path is concerning.⁴ 

To solve power bottlenecks in beyond 51.2 Tb/s ethernet switch, co-packaged optics (CPO) architecture has been proposed.⁵ In the CPO module, optical components such as photonic chips and electronic elements such as ethernet switches are integrated on a board, and signals are converted from electronics to optics using a silicon transceiver near the ASIC. By replacing a part of the power-consumptive electrical wires with optical links, CPO utilizes low power consumption. In 2019, CPO collaboration was established by hyper-scale data center companies,⁶ and related technologies are being extensively researched.^7–9 

Recently, an active optical package (AOP) was proposed^10–13 as one of the CPO configurations. The schematics of the AOP substrate are shown in Fig. 1. Silicon photonics are embedded on an organic package substrate and polymer waveguides are integrated as optical links, as shown in Fig. 1(a). Optical connection between a silicon waveguide and a polymer waveguide is required. Some coupling technologies for silicon photonics have been developed, such as spot size converter,¹⁴ grating coupler,¹⁵ evanescent coupler,^16,17 photonic wire bonding,¹⁸ elephant coupler,¹⁹ and optical pin.^20,21 The coupling solution should be selected considering various aspects of application, manufacturing, wavelength/polarization characteristics, and coupling efficiency. For the AOP substrate, we developed a coupling technology with a pair of 3D micromirrors, as shown in Fig. 1(b), for broadband, wavelength/polarization independent optical connection in compact coupler size. The bottom-side mirror has a biconic surface to convert the beam size between a silicon waveguide and a polymer waveguide. The top-side mirror has an angled flat slope. Using fabricated optical connections, including a micromirror-based optical coupler, the low-loss, broadband optical transmission¹¹ and 112 Gb/s PAM4 transmissions at 25  and 85 °C were demonstrated.¹³ 

3D micromirrors for the AOP substrate were fabricated using grayscale photo lithography. The bottom-side mirror was made of photosensitive polyimide, and the top-side mirror was made of a polymer for cladding. A material for the bottom-side mirror can be selected without considering material optical properties such as reflective index and transparency since light does not propagate inside the mirror. Photosensitive polyimide used for the bottom-side mirror is suited to being formed by grayscale photo lithography, and its fabrication is stable. The top-side mirror has a role as an overclad of polymer waveguides, and the material selection is restricted. The grayscale photo lithography with polymer resists is sensitive to the process conditions and requires high fabrication accuracy.¹¹ 

In this study, instead of grayscale technology, UV nanoimprint lithography technology,^22,23 which can fabricate structures by transferring the pattern of a mold, is introduced to top-side mirror fabrication for obtaining a stable mirror shape.²⁴ The micromirror fabrication is demonstrated on a silicon wafer by a step-and-repeat imprint process and evaluates the shape and misalignment of the patterned mirror to confirm the availability of a top-side mirror process by nanoimprint technology. In Sec. II, a fabrication process for micromirrors with imprint technology is considered by a hyperelastic analysis of the imprint process model using the finite element method. As the top-side micromirror for the AOP substrate functions to propagate the light form bottom-side mirror to a polymer waveguide and its height misalignment affects coupling efficiency, it is necessary to select an imprint process that can obtain height accuracy in terms of fabrication height. Then, the micromirror fabrication is demonstrated on a silicon wafer by the selected imprint process. In Sec. III, the shapes of the micromirror patterned by imprint technology are discussed in terms of the range of mirror shape and misalignment. The study is summarized and concluded in Sec. IV.

The fabrication process of the optical connector in the AOP substrate is explained in the literature.¹¹ The bottom-side mirror is formed in the cavity of a Si photonics chip by the grayscale UV lithography and is encapsulated with a transparent polymer. Then, polymer underclad and waveguides are integrated by the laser lithography, and finally, the top-side mirror is fabricated. In this study, the imprint process is introduced to top-side mirror fabrication, instead of grayscale lithography. Here, the top-side mirror is desired to be more than 20 μm in height and angled by 45° since it is required to reflect beams collimated for coupling to a polymer waveguide with a mode field diameter of around 10 μm. In this section, to verify the ability of an over 20 μm-height micromirror fabrication by photo nanoimprint technology, patterned micromirrors are demonstrated on a 4-in. silicon wafer. Polydimethylsiloxane (PDMS) is selected as a mold material because it has replicability and high gas permeability. Since the top mirror plays the role of guiding vertical light to the polymer waveguide and the height alignment is required in mirror fabrication, the imprint process with a guide structure, as shown in Fig. 2, is considered.

For deciding the height guide in micromirror fabrication using soft imprint lithography, the deformations of the hyperelastic material in the imprint process model were analyzed using the finite element method and commercial software ANSYS Mechanical, as shown in Fig. 3. The schematics of analysis are shown in Figs. 3(a) and 3(b). They are simplified models of the imprint process where the PDMS mold, having a 30 μm-height mirror, is pressed onto a sample. The mirror length on the y-axis is set to 500 μm. There are 50 μm-height and 150 μm-width guide structures on both sides of the mirror. The guide on the right is PDMS in case A, as shown in Fig. 3(a); in case B, it is a cured polymer, which is approximated to steel, as shown in Fig. 3(b). The polymer underclad and waveguides are assumed to be integrated on the other side, so the right guide is a polymer guide in both cases. In this analysis, the fluid polymer resist between the mold and the sample was not considered. The thickness of the mold holder, the sample, and the PDMS mold is 500 μm. The 30 μm-height mirror structure has 45° slopes on both sides, and its width is 70 μm. As shown in the 3D analysis model of case A in Fig. 3(c), the analysis area is 600 μm in the x-axis direction and 500 μm in the y-axis direction, and the shape is uniform in the y direction. The force in the range of 1–36 kPa was added to the top surface of the model, of which size is 500 × 600 μm², and the bottom surface of the sample was fixed as boundary conditions to analyze the amount of deformation in both cases. Here, the materials of the sample and the mold holder are approximated by steel since they are much harder than PDMS. As mechanical properties of the hyperelastic material PDMS, the second-order Ogden model is used, where the material constants μ₁ and μ₂ and α₁, α₂, and incompressibility parameter D₁ are 0.000 342 8 MPa and 0.131 62 MPa and 7.7991, 3.6718, and 0.001 646 3 MPa⁻¹, respectively.²⁵ 

Figures 3(d) and 3(e) show the total deformation around the mirror when the applied force is 20 kPa in cases A and B, respectively. The total deformation around the PDMS mirror in case A is larger than that in case B. For further evaluation, Fig. 3(f) shows the maximum value of z-axis directional deformation in the mirror part, and Fig. 3(g) shows the maximum value of the x- and y-axis directional deformation. The horizontal axis represents the applied pressure obtained by dividing the applied force by the area of the mold surface, 500 × 600 μm². The deformation mainly occurs in the z-axis direction, as shown in Fig. 3(f). The z-axial deformation of the mirror part is linearly increased in proportion to the applied pressure, and the increase is larger in case A than that in case B. For example, at 20 kPa, z-axis directional deformation is 9.6 μm in case A and 8.1 μm in case B; thus, the difference is 1.5 μm. As shown in Fig. 3(g), the mirror part is slightly deformed in x- and y-axis directions. By comparing the two cases, x-axial deformation is larger in case A than that in case B, as in z-axial deformation. The y-axial deformation is almost the same in two cases. When the applied pressure is 20 kPa, the x-axial deformation in case A is 0.61 μm larger than that in the other.

According to the result, in case B, where guides on both sides are steel, indicating a harder material, the z-and x-axis directional deformation is smaller than that in case A, where the left guide is an elastic material PDMS. This means that by adopting a UV-cured polymer as a guide material on both sides, fabrication height fluctuations due to pressure variation can be suppressed than by adopting a PDMS guide. From the result of these fabrications and calculations, imprinting the mirror to prepared polymer guide is suited to top-side mirror fabrication, which requires height alignment.

Master molds and PDMS replica molds were prepared before fabrication demonstration of micromirrors, as shown in Fig. 4. For quick and flexible fabrication of mold prototypes, master molds were formed by grayscale photo lithography with photosensitive polyimide, which is used in the bottom-side mirror. Since the master mold is repeatedly used as a template for the replica mold and photosensitive polyimide is suited to the grayscale process, the mirror shape instability dependent on the process condition can be avoided compared with the directly forming polymer clad mirror by grayscale photolithography. The process overview of a master mold and a replica mold is shown in Fig. 4(a). The polyimide is spin-coated onto the silicon wafer so that its thickness is around 30 μm and is exposed to UV, where laser intensity is controlled by each pixel. After development, it is coated with gold for use as a master mold. By applying the feedback of the difference between a fabricated and a design shape and correcting the map of laser intensity, a fabricated master mold becomes closer to the ideal mirror. Then, PDMS is poured into the master mold and cured by UV. Finally, PDMS is released from the master mold as a replica mold. The master mold is designed to have two slopes on both sides, of which the angle is 45° and the height is 30 μm. There is a 10 μm-width flat surface between the slopes.

The surface profiles of fabricated master and replica molds, which were measured by a laser scanning microscope, are shown in Fig. 4(b). The profile of the master mold is shown with a dotted line and that of the replica mold is shown with a green solid line. The z-axis is defined to be vertical to the silicon wafer and its positive direction is downward. Table I shows values of heights and left and right slopes' angles of the fabricated molds' mirrors. The angles were obtained by linearly fitted using the least squares method, where the fitting error was set to be less than 300 nm. There is slight difference in mirror shapes between the master mold and the replica mold. The replica mirror is 0.5 μm higher and its slope angle is 0.9° larger than the master mirror. This difference could be caused by the cure shrinkage of PDMS. The obtained replica mold was used in mirror fabrication in Sec. II C.

The result of the hyperelastic analysis in Sec. II A shows that in the micromirror imprint, higher press tolerance can be obtained by the guide structure of a hard material, not PDMS. The mirror fabrication by the selected imprint process is demonstrated on a silicon wafer without other optical elements, such as bottom-side mirror and polymer waveguide, to evaluate the shape of mirror without the effects of other elements' processes, as shown in Fig. 5. The demonstration is performed by step-and-repeat imprinting on the wafer, where nine mirrors are fabricated in one step. Four times imprints were performed with the mold pressure of 8.6–9.7 kPa. To make the guide area as large as possible, a groove structure for the mirrors was adopted as a height guide. Figure 5(a) shows the 3D view of mirror fabrication by imprinting into a groove. The z–x cross view of the schematic is shown in Fig. 5(b). The average guide groove depth was 36.7 μm and the width was 278.0 μm. The average mirror height in the replica mold was 31.1 μm. The replica mirror length and the groove length in the y-direction is designed to be 500 and 700 μm, respectively. The patterns with grooves were prepared on a 4-in. silicon wafer by laser lithography. Since imprinting on the guide groove is prone to cause bubble errors, the guide groove has air bubble traps to prevent remaining air near the mirror, as shown in Fig. 5(a). The size of the PDMS mold was 11 × 34 mm².

The microscope image of the fabricated layout including nine mirrors by imprinting one step is shown in Fig. 5(c). As the layout has been designed in accordance with the silicon chip, the mirrors are not uniformly located. Figure 5(d) shows the surface profile of mirror C fabricated in four imprints, which is the mirror in the lower left of the microscope image. The shapes of the fabricated mirrors are shown with solid lines and that of the replica molds are shown with a dotted line. The average height of the fabricated mirror C is 30.5 μm in four imprints. The minimum value in the z-axis of the surfaces is set to zero to be easily compared. Based on the surface profile, the stable mirror shapes were obtained for four times imprinting, even though there was some deformation from the mirror shape replica mold. The shape of the master mold's mirror C used for obtaining the replica mold has been shown in Fig. 5(b) in Sec. II B.

Table II shows the values of mold-pressure, prepared guide groove height, mirror height, polymer thickness after the imprint process, and polymer thickness over the guide, of mirrors A, B, and C in four imprints. The polymer thickness after mirror fabrication was measured by the laser scanning microscope with a film thickness measurement mode, and the guide groove height was the measured value of the groove for mirror B in each layout. Figure 6 shows the schematic of micromirror fabrication after the imprint process with explanation of thicknesses used in the tables. The polymer thickness over the guide was obtained by subtracting the prepared guide height from the polymer thickness after the imprint process. The average polymer thickness over the guide of three mirrors A, B, and C in four imprints was 9.6 μm. Here, the minimum value is 6.9 μm and the maximum is 11.9 μm, as shown in Table II. It was determined that the maximum difference in the fabricated mirrors' height position was 5.0 μm for three mirrors in four times imprint, except for the difference in the prepared guide height. Table III shows the difference in polymer thickness over the guide at mirrors A–C for the four times imprint process. According to the table, the difference in the fabricated height position in four imprints at mirrors A, B, and C was 2.6, 2.7, and 2.7 μm, respectively. The maximum mirror-dependent difference was 2.3 μm. The fabrication position difference depends not only on fabrication but also on the mirror position in the mold. The mirror-dependent height misalignment can be improved by adding a tilt correction to the PDMS mold. The fabricated shape of the mirrors and the effect of height misalignment are discussed in Sec. III.

In this section, the mirror shapes and height alignment are discussed. Over 30 μm-height mirrors were obtained by imprinting into the guide groove. The surfaces of mirror C were linearly fitted using the least squares method, where the fitting error was set to less than 300 nm. Figure 7(a) shows the shape of mirror C in imprint #3 with linear slopes, angle, and length of flat slope obtained by the fitting. Table IV shows values of mirror height, angles of left and right slope, and lengths of flat slope on left and the right sides of mirror C in four imprints. The angles and the slope length were obtained by linear fitting. The maximum difference of height and angle is 0.2 μm and 0.2°, respectively. Thus, the mirrors were patterned with almost same angle and height in four times imprinting. The difference of flat slope length is 0.8 μm on the left side and 4.0 μm on the right side. The flatness of slopes is not stable on the right. The length difference of the flat slope mainly occurred at the lower bottom of the mirror, where the mold mirror has flat slope defects. The flatness of the imprint mirror can be improved by using a replica mold with flat slopes.

There are shape differences between mirrors of the replica mold and fabricated micromirror, as shown in Fig. 5(d). According to Tables I and IV, the imprinted mirror is up to 0.6 μm lower than the replica mold. The maximum difference of angle is 0.7° in the left slope and 1.3° in the right slope. Since the height of the imprinted mirror is lower than the PDMS mold, this difference is supposed to be caused by deformation of the PDMS convex mirror structure in the imprint process due to force back of residual polymer resist in the groove. The thickness of the residual polymer under the mirror in the groove is the difference between polymer thickness and mirror height as shown in Fig. 6. The average thickness of the residual polymer is obtained as 15.5 μm by subtracting the mirror height from the polymer in Table II. The deformation of PDMS's mirror could be reduced by widening the groove. Even if there are shape difference between the mold and the imprinted mirror, the mirror shape can approach the ideal design by properly feeding back to the shape of the master mold considering the mirror shape fabricated by the imprint process.

The evaluation of the fabricated mirror by imprinting into the groove is shown in Fig. 7. Figure 7(a) shows the shape of the mirror C in imprint No. 3 and (b) shows the shape of mirror fabricated by grayscale photolithography reported in Ref. 11, with the result of linear fitting using the least squares method, where the fitting error was set to less than 300 nm. The slope length of the imprinted mirror is more than 23.5 μm according to Table IV, and the value is 16.7 μm in the case of grayscale lithography as shown in Fig. 7(b). Thus, the fabricated mirror shapes are improved by imprint technology compared to grayscale lithography.

By using the fitting result of the fabricated mirror C in imprint No. 3, the coupling efficiency of the optical coupler from a silicon waveguide to a polymer waveguide was calculated using commercial software, Zemax OpticsStudio, to evaluate the height alignment in mirror fabrication. The schematic of the calculation is shown in Fig. 7(c). The diameter of the top-side mirror was set to 23.5 μm, and the angle was set to 45.9°. The bottom-side mirror was defined as a biconic mirror, rotated 45° around the y-axis, of which the curvature was set to 26.84 μm at the y-axis and 52.18 μm at the z-axis. The positions of the mirrors were optimized before calculation. To consider the fabrication height error, the optical coupling efficiency was calculated by verifying the position of the top-side mirror.

Figure 7(d) shows the relationship between the height misalignment of the top-side mirror and coupling efficiency from a silicon spot size converter (SSC) to a polymer waveguide with a wavelength of 1310 nm. The beam radius emitted from SSC is set to 1.14 μm at the y-axis and 1.13 μm at the z-axis; the mode field diameter of the polymer waveguide is set to 9.20 μm. In mirror fabrication in Sec. II C, the difference in the fabricated height for three mirrors in four imprints was 5.0 μm, so the coupling efficiency was more than 70% (−1.55 dB) when the height alignment was assumed to be ±2.50 μm. In addition, the difference of mirror C in four imprints was 2.7 μm, and when the height alignment was ±1.35 μm, the coupling efficiency was more than 88% (−0.56 dB). Thus, it is possible to improve the coupling efficiency by correcting the tilt of the mold and reducing misalignment in the imprint plane.

According to the evaluation of the mirror fabricated by the proposed imprint process, for top-side mirror fabrication, by preparing grooves for mirrors with the polymer clad, core layers, and imprinting the mirror into the groove, a mirror shape with enough length-flat slopes can be stably obtained. The height alignment can achieve up to ±2.50 μm, where the calculated coupling efficiency is more than 70%, by verifying the median value of the fabricated height position in advance. The coupling efficiency of 70% (−1.55 dB) is acceptable, since the reported fiber coupling efficiency of the high-efficiency grating coupler for O-band is −1.9 dB.²⁶ A grating coupler is a vertical coupler in silicon photonics, and one of generally selected optical interfaces in co-packaged optics. Considering that there are additional losses, such as propagation loss at polymer waveguides and coupling loss between a polymer waveguide and an optical connector in the AOP substrate, it is desirable to improve the coupling efficiency by correcting the tilt of the mold.

In this study, UV imprint technology was adopted for stable and high-throughput micromirror fabrication of a polymer mirror of an optical coupler in CPO, instead of grayscale UV lithography technology, by which the shape of fabricated mirror depends on the process condition. The imprint process with the PDMS mold was determined as imprinting onto a guide structure in the cured polymer clad layer for height alignment by the result of the hyperelastic analysis. The fabrication of over 30 μm-height micromirrors was demonstrated by imprinting onto a guide groove prepared with the polymer clad on a silicon wafer. The stable mirror shapes, where the maximum difference of height and angle was 0.2 μm and 0.2°, respectively, were obtained for four imprinting. The height position difference of three mirrors in four imprints is 5.0 μm. The height misalignment is assumed to be ±2.50 μm, where the calculated coupling efficiency of an optical coupler is more than 70% (−1.55 dB). The height misalignment could be improved by adding tilt correction to the replica mold. These results imply that the shape of the top-side mirror for CPO can be stably fabricated by adapting imprint technology and enough coupling efficiency could be obtained by the imprinted mirror even if the fabrication height misalignment is considered. In future, the top-side mirror can be fabricated by imprint technology with other optical elements of the AOP substrate such as a bottom-side mirror and optical connection can be experimentally verified.

This study was based on results obtained from a project (No. JPNP13004) commissioned by the New Energy and Industrial Technology Development Organization (NEDO).

The authors have no conflicts to disclose.

Fumi Nakamura: Data curation (lead); Formal analysis (lead); Investigation (lead); Methodology (equal); Writing – original draft (lead); Writing – review & editing (lead). Kenta Suzuki: Formal analysis (supporting); Investigation (supporting); Methodology (lead); Writing – original draft (supporting); Writing – review & editing (supporting). Akihiro Noriki: Formal analysis (supporting); Investigation (equal); Methodology (equal); Writing – original draft (supporting); Writing – review & editing (supporting). Takeru Amano: Formal analysis (supporting); Funding acquisition (lead); Investigation (supporting); Project administration (lead); Writing – original draft (supporting); Writing – review & editing (supporting).

The data that support the findings of this study are available from the corresponding author upon reasonable request.