Chip-scale optical interconnection has emerged as an essential technology to realize high-density interfaces for next-generation data centers and high-performance computing systems [1]. In present Si photonic components, such as 100-Gbps-level transmission modules, fiber-to-chip interconnection is utilized, applying active alignment between the fiber block and the tapered waveguide edge [1]. However, achieving this active alignment results in high packaging costs, and this remains a big obstacle to mass-producing the optical components [2,3].

To avoid active alignment, interconnection technologies using polymer wires have been developed [4–11]. One of these technologies utilizes two-photon polymerization to form three-dimensional wires with nanoscale thickness by writing with focused laser beam [4–7]. This requires a lithography procedure that uses a photoresist on the sample. A non-lithography-based technology has also been developed: direct optical wire (DOW) bonding using open-to-air polymerization [8–11]. In this bonding, the polymer wire is drawn from a micropipette filled with polymer solution. This technology is easily adoptable for packaging processes, much like metal wiring. In a previous work [11], DOW bonding was applied to multimode (MM) devices: vertical-cavity surface-emitting laser and MM fiber.

In this Letter, we present the application of DOW bonding to silicon photonics chips that require a single-mode (SM) link. Wire bonding was applied between grating couplers for on-plane connection and also between a grating coupler and a single-mode fiber (SMF) for out-of-plane connection. Even though the DOW technology can potentially draw a thin SM wire, micrometer-scale MM wires were applied to ensure mechanical stability. A polymer stud bonded above the aperture was specifically designed for the SM link through the MM wire through simulation.

Figure 1 shows the architectural concept of the DOW bonding technology for chip-to-chip and chip-to-fiber interconnections. As a first trial for silicon photonic chips, we applied the wire bonding technology to a grating coupler designed for input and output coupling [12,13]. The wires were bonded with two schemes: grating-to-grating and grating-to-fiber, as illustrated in Fig. 1. The Si photonic chips were fabricated by the CMOS process using KrF photolithography and silicon on insulator wafers with 220-nm- and 2-µm-thick Si and buried oxide layers, respectively. The photonic chips consisted of full-etched SM waveguides with a width of 500 nm and a 10-µm-wide grating coupler. At the ends of the waveguide for input and output coupling, identical grating couplers were formed with a period of 620 nm and an etch depth of 70 nm, which were designed for SMF coupling in the C-band wavelength region. The length of the grating was 15.2 µm, corresponding to 24.5 periods of grating. External light from a tunable laser was coupled through an input grating coupler (not shown in Fig. 1 for simplicity) to generate a guided optical mode. The output couplers were linked with the wires to the input couplers of another chip or to an external SMF block.

 

Fig. 1. Architecture of chip-to-chip and chip-to-fiber interconnections using direct optical wire bonding technology. The wire is formed by the open-to-air polymerization of a polymer solution extruded through a micropipette. The wire bonding was examined for use in grating-to-grating and grating-to-fiber interconnects.

Download Full Size | PDF

The polymer wire was formed by controlling a glass micropipette filled with polymer solution, which was made by dissolving polystyrene powder in xylene solvent at a concentration of 0.5 wt.% [8–11]. As it is a polymer material, polystyrene offers high optical transparency, good surface adhesion, and a high refractive index n = 1.54 at 1550 nm [8]. A convex meniscus of the polymer solution was extruded from the micropipette (tip diameter: 0.5 µm) and contacted the grating region. As the liquid meniscus was stretched by pulling the micropipette, the polymer solution within the meniscus rapidly polymerized in the air and the volatile xylene evaporated, leaving the polystyrene solid. The wire shape was controlled by manipulating the pulling motion of the pipette. Through appropriate vertical and lateral movement, arch-shaped wires were formed in the grating-to-grating connection. For the grating-to-fiber connection, the micropipette was rotated by 90° while maintaining the pulling speed to reach a vertical facet of the fiber placed horizontally at the edge of the chip. The connected wire formed a half-arch shape, as illustrated in Fig. 1. Since the DOW bonding process is controlled by the physical actions of the micropipette, it provides a one-stop procedure for landing and wiring, analogous to electrical wire bonding.

Figure 2(a) shows microscope image of the optical wires linked between two separated chips on which grating couplers were formed. The stud shape is seen in the SEM image of the wires formed on one chip shown in Fig. 2(b). We established reproducible wiring procedures to control the radius of curvature within the range of ∼500–50 µm for a wire with a diameter of ∼12–4 µm, and a sample with a radius of approximately 150 µm and a diameter of 7 µm is presented in Fig. 2(a). The wire diameter was controlled by the pulling speed, which was in the range of ∼1–10 µm/s. After the wiring, thermal curing was performed at 200°C for thermal endurance.

 

Fig. 2. Optical wires directly bonded between grating couplers: (a) microscope image of the chip-to-chip connected wires; (b) SEM image, with an inset showing the stud shape formed at the end of the wire.

Download Full Size | PDF

The shapes of the ends of the polymer wire that contact the grating or fiber aperture play an important role in the guiding of light waves. The typical shape of a wire end formed by the contact of a liquid drop with the substrate has a wide contact area [10] and is very similar to the stud shape of metal wire ends. In the DOW bonding process, the contour of the stud can be controlled by adjusting the pulling speed of the micropipette. At a pulling speed of near to 1 µm/s, the stud could be stretched into a linearly narrowing cone shape, as shown in Fig. 2(b). Such a polymer stud can function as a tapered waveguide. To design the appropriate cone-shaped stud, we conducted a detailed study of the dimensions of the stud and wire using simulations of the light propagation in which the geometric parameters were varied.

Figure 3 shows a schematic of the structure of a wire connecting two grating couplers, indicating the geometric parameters investigated in the simulation. The wire is bent into an elliptical shape with horizontal radius a (half of the connecting distance), vertical radius b (the height of the wire), and diameter d. The stud is defined as having a bottom diameter D and a length L until it reaches the wire diameter d. The landing position x_l of the stud center along the x direction from the grating center is also defined. Some of the geometric parameters were determined using preliminary experiments and simulation. First, the connecting distance 2a was determined as 300 µm, considering the stability of the curved wire and the controllable range of our chips in the packaging work. For this distance, the wire height b was determined as 170 µm, which showed the lowest propagation loss. The wire diameter d was determined as 7 µm, considering the propagation loss and also the stability of the wire after connection.

 

Fig. 3. Schematic cross-sectional structure of a wire. The geometric parameters investigated in the simulation of light propagation are indicated. (a) Elliptic wire connected between grating couplers. (b) Tapered stud formed on the grating.

Download Full Size | PDF

The light propagation through the whole wire was simulated using a 2D-finite-difference time-domain (FDTD) simulator to analyze the coupling and propagation losses in the wire. We also used a 3D-FDTD simulator to accurately design the parameters (D, L, and x_l) of the stud that sensitively affect the coupling loss. However, this was only done for the outcoupling region. For the incoupling region, a 2D simulator was used, since the scale of the whole wire was too large to simulate it three-dimensionally. The coupling losses were calculated with P₂ and P₁ for outcoupling and with P₄ and P₃ for incoupling, where P₁ and P₄ are the powers that pass through the front of the grating and P₂ and P₃ are those that pass through the top of the stud, as denoted in Fig. 3(a). The propagation loss in the wire was calculated with P₂ and P₃. The insertion loss of the whole wire was calculated using the coupling and propagation losses.

Through the 3D simulations of the stud, we found that the output coupling loss reached and maintained a minimal value of 2.7 dB when D and L were larger than 11 µm and 20 µm, respectively. When D < 11 µm, a large amount of leakage took place at the grating–wire interface because the stud bottom could not sufficiently cover the grating area. When L < 20 µm, some scattering losses occurred in the tapered region of the stud since L was too short to satisfy the total internal reflection condition. The coupling performance also depended on the position of the stud. For the stud that was found to have the best geometry of D = 11 µm and L = 20 µm, the best landing position was found to be x_l = –2 µm, which was shifted a little to the front of the grating. This is due to the fact that a large portion of the light was scattered at the front side of our grating. Deviating from this position, the coupling loss dropped to 3.3, 2.9, 2.7, 2.8, and 2.9 dB at x_l = –1.0, –1.5, –2.0, –2.5, and –3.0 µm, respectively, at λ = 1550 nm. The 2D simulation confirmed that the position corresponding to minimum loss was almost the same in the outcoupling region.

Figure 4 shows the 2D simulation results for the coupling loss and the field distribution in the stud regions. The studs were set to the best geometry and position designed at the target wavelength of 1550 nm, as described above. To identify the wavelength tolerance, the wavelength of the incident light was varied from 1500 nm to 1620 nm for the same structure. The results indicate that most of the coupling loss occurred in the incoupling region, particularly in the wavelength range of ∼1500–1580 nm. Above this range, the incoupling loss became comparable to the outcoupling loss. However, below this range, the incoupling loss dropped significantly below the outcoupling loss level. This can be explained by considering that the diffraction angle diverges greatly from the normal direction in the short-wavelength region and the stud has a wider numerical aperture that is acceptable to inclined incident light compared to the grating. Notably, the loss spectra showed a comb-like oscillation, as presented in Fig. 4(a). This aspect indicates that the optical wire raises some resonance between the gratings.

 

Fig. 4. Simulation results for a directly bonded wire with the optimal geometry and position on the grating couplers. (a) Out- and incoupling losses in the wavelength range of ∼1500–1620 nm and (b) electric field distributions in the out- and incoupling studs for two different wavelengths, 1500 nm and 1600 nm.

Download Full Size | PDF

Figure 4(b) shows the distribution of the electric field of the light in the stud regions for two representative selected wavelengths, 1500 nm and 1600 nm. In the outcoupling region, most of the light is scattered at the front (near x = –5) of the grating for both wavelengths. However, above the grating surface (z > 0), the near field in the incoupling stud and the far field in the outcoupling stud exhibit complicated distributions that do not show the SM pattern. Also, the distributions are quite different for the two wavelengths. In the phase analysis, the phase was found to gradually vary along the grating surface in the outcoupling region. However, in the incoupling region, the phases impinging upon the grating surface showed a random distribution. This random phase distribution can cause increased loss in the incoupling region, together with severe mode conversion in the grated waveguide.

In the simulation of light propagation through an elliptically bent wire, the propagation loss calculated with P₂ and P₃ was negligible, with a value below 0.0001 dB. It should be noted that the polymer wire with d = 7 µm was far from the SM condition (d ∼ 1 µm), leading to many modes in the wire [14]. In a simple 3D simulation of the incoupling stud closely linked to the outcoupling stud, two modes, TE₀₀ and TE₁₀, entered the Si waveguide after passing the incoupling grating. Thus, serious mode conversion occurs in the incoupling, which can cause significant loss.

Figure 5 shows the insertion loss of the whole wire, calculated as the sum of the coupling and propagation losses obtained in the 2D simulation presented in Fig. 4. These values were compared with the experimental data (red line) measured for a DOW-bonded sample fabricated as designed. The measured data were normalized to those of a reference structure consisting only of a waveguide and input/output grating couplers [4]. They were also compared with a fiber link (blue line) packaged with an 8°-polished SMF block, representing a conventional coupling method for the grating. The measured insertion loss for the DOW coincided well with the simulated value, as shown in Fig. 5. The experimental and simulated free spectral ranges estimated from the interval of the oscillation curves were also nearly identical: 1.40 nm and 1.41 nm, respectively. This result indicates that the stud and wire were well formed (they were similar to the designed structures) and that the 2D simulation was able to predict the performance of the whole wire. In the fiber link, the insertion loss dropped markedly when the wavelength deviated from the central wavelength of 1560 nm. This proves that our wire link can provide superior wavelength tolerance performance. The best measured insertion loss was 5.8 dB for the DOW link at 1590 nm, which is considerably smaller than the best one for the fiber link, measured as 9.3 dB at 1560 nm. The deviation of loss values in bonded wires was within 2 dB in most cases. However, the insertion loss was much larger than that of an SM polymer wire link between inverse-tapered waveguides obtained through two-photon lithography [4,5]. In our DOW bonding, most of the loss can be attributed to the grating couplers, which have an inherently low coupling efficiency. Nevertheless, our DOW bonding can be simply implemented for chip-to-chip interconnection.

 

Fig. 5. Simulated and measured insertion losses of the directly bonded wire between the grating couplers. The measured data for a conventional SM fiber link between grating couplers is also provided as a reference.

Download Full Size | PDF

Figure 6 shows the simulated and experimental insertion loss results for the wire link between a grating coupler and a SMF. The polymer wire was bonded near the center of the polished aperture of a SMF. The bonded wire had a half-arch shape with a = 440 µm and b = 380 µm. The studs on the grating and fiber apertures were formed almost identically. The overall trend for the measured insertion loss is similar to the simulated one, with a large wavelength tolerance. However, there was some deviation between the measured and simulated loss level data. Also, their wavelength dependencies were different than the results for the grating-to-grating connection. In the simulated curve, oscillation due to resonance was lessened and the overall wavelength dependency was reduced compared to the grating-to-grating bonding. This result may be attributed to the fact that the light propagating through the polymer wire undergoes less reflection at the glass fiber surface than at the silicon surface and it is less wavelength sensitive compared to the grating surface. We note that the loss of light propagating from grating to fiber was similar to that of light propagating in the opposite direction; it showed a slightly lower value of about 0.4 dB on average, even though the MM wire was in contact with the SMF aperture.

 

Fig. 6. Simulated and measured insertion losses of the directly bonded wire link between the grating coupler and the fiber. The inset is a microscope image of a grating-to-fiber connected wire.

Download Full Size | PDF

The measured loss curve in Fig. 6 is lower than the simulated curve. We suggest that this discrepancy can be attributed to misalignment of the wire on the fiber aperture. In our system, the alignment accuracy of the micropipette was 0.25 µm on both planar and vertical surfaces, but the accuracy in chip (or optical bench) packaging was about 3 µm in the lateral direction (in the xy plane). For a grating-to-grating link, the wire can be formed into an arch shape without distortion in the xy plane. However, for the grating-to-fiber link, if the chip and fiber are misaligned with each other, the wire can be distorted to form vertical studs on both apertures. This distortion can be greatest during the last step of polymerization, when the pipette approaches the fiber aperture, and may result in some unexpected loss. There is thus the potential for further loss reduction by improving the wiring process, which could be achieved by anticipating the final destination of the bonding position, as well as chip alignment. With these improvements, the wiring technique of rotating the micropipette by 90° could be practically utilized for other devices such as a distributed feedback laser diode (DFB LD), which requires SM edge coupling. To attach the MM stud to a thin aperture of the DFB LD, it is necessary to widen the cross-sectional area above the aperture and to prepare anti-reflection layers for the polymer. The coupling loss could also be reduced by improving the grating structure to make it compatible with the polymer stud.

In conclusion, we have demonstrated, for the first time in our knowledge, an optical interconnection between SM devices achieved by direct optical wire bonding through open-to-air polymerization. The SM light was guided through a MM polymer wire which ended in a tapered stud shape that functioned like a mode converter on the SM aperture. The stud shape was designed by FDTD simulation and the wire bonding was examined for use in grating-to-grating and grating-to-fiber links. The experimental results for optical losses in the bonded wires were in good agreement with the simulation. The results of the polymer wire bonding showed there was much less wavelength dependency of its insertion loss compared to that of a conventional link using 8°-polished SMF on the grating coupler. We expect that this optical wire bonding technology, which has the benefit of versatile controllability as a micropipette is used to draw the polymer wire, can be practically utilized for various devices and chips with horizontal or vertical apertures. The wiring speed could be improved by using a highly volatile polymer solution and multiple pipettes for parallel processing.