Demonstration of a Single-Mode Expanded-Beam Connectorized Module for Photonic Integrated Circuits

—We present the optical and mechanical designs of an optically-pluggable package for photonic integrated circuits (PICs). Using off-the-shelf micro-optical components, low-loss ( ∼ 2 dB) expanded-beam connection between an edge-coupled SiN- based PIC and standard optical ﬁber is demonstrated. High reproducibility of coupling efﬁciency is demonstrated over up to 80 mating/de-mating cycles.

MFD). As a result, coupling of fiber to the PICs requires high-precision (<1 μm tolerance) and highly-stable mechanical alignment, and may require mode transformation optics to minimize mode mismatch loss.
At present, stable optical coupling between a PIC waveguide and an optical fiber is generally achieved by directly bonding the fiber (or array of multiple fibers) to the optical output facet of the PIC using a transparent optical adhesive. This is typically done using so-called "active alignment" where light is injected into the PIC, and the fiber position is adjusted to achieve maximum coupling as determined by a detector monitoring the output of the PIC, and then the adhesive is cured. The result of this process is a "pigtailed" PIC, that is, one having a permanently-attached optical fiber. Unfortunately, the presence of the pigtail is inconvenient for the subsequent handling of the PIC and its associated package/module. For example, the temperature sensitivity of the pigtail fiber cable makes conventional solder reflow processing to mount the module on a circuit board impossible. Additionally, the presence of pigtails of fixed lengths requires that either lengths are customized for a given application, or that there is a proliferation of part numbers and increased inventory. Furthermore, accommodation (e.g., in a coil) of any extra pigtail length after installation in the final application can increase system volume, and complicate cable management and replacement of failed transceivers. Thus, for multiple reasons, development of a PIC package with a convenient pluggable connector interface rather than a pigtail is very desirable.
While existing single-mode optical connectors designed for low-loss cable-to-cable connections are potentially adaptable for PIC interfaces [1], [2], they suffer a few drawbacks that are particularly problematic for PIC module applications. One issue is that they rely on physical contact between the fiber cores to effectively eliminate the refractive index discontinuity at the interface. This is achieved by polishing the fiber ends to cause slight protrusion of the core regions, then forcing the fibers together with adequate force to cause elastic deformation of the cores, thereby obtaining intimate optical contact. This approach, while suitable for connectors mating a small number of channels/fibers, becomes difficult for high-channel-count systems because the minimum connector mating force scales with the number of channels, and because achieving uniform physical contact across many fibers (e.g., >24 fibers) imposes unreasonable manufacturing tolerances. Another disadvantage of physical contact connectors is that they are extremely sensitive to particulate contamination. Particles trapped between the mated fiber ends can prevent intimate optical contact, and, if they happen to lie in the core region, can scatter the optical signal and contribute to loss. Furthermore, because of the tight lateral alignment tolerances imposed by the small mode sizes, the mechanical fit of connector mating parts must be very accurate. The resulting tight clearances for mating parts (e.g., the fit of pins in holes) increases sliding friction, thus contributing to high mating forces and increasing the potential for jamming in the presence of particulate contamination.
One approach that has been found to be promising for addressing many of the above issues is the use of expanded-beam optical interfaces in connections. To date, the approach has been most commonly applied to multi-mode fiber systems [3] but it is also being pursued for single-mode fiber-to-fiber connections [3], [4], [5]. The approach consists of using beam-expanding optical elements, typically lenses, on each side of the optical connector interface to couple the fiber modes to larger free-space beams. The lateral alignment tolerances that must be held between the expanded beams are then determined by the expanded beam dimensions rather than by the fiber core dimensions.
There are three main benefits to the use of expanded-beam coupling. First, the relaxation of lateral alignment tolerances means that the mechanical precision that must be achieved by the connector-to-connector alignment features is likewise relaxed. This has the dual benefit of easing manufacturing process control, as well as allowing looser fit in mating parts, thereby reducing the mating force. Second, the increased size of the optical mode means that the scattering loss produced by a particle in the optical path is proportionally reduced, making expanded-beam connectors more tolerant of contamination [3], [6], [7]. Finally, the expanded-beam connectors are generally designed such that there is no contact between the optical elements, thereby allowing low mating force and avoiding damage to optical surfaces due to debris being pressed onto/into the optical elements during mating [5].
Extraction of light from a PIC waveguide for subsequent coupling to a fiber via a connector may be performed in one of two ways: 1) edge coupling where the waveguide terminates at/near an edge of the PIC and emitted light propagates "horizontally", that is, approximately parallel to the plane of the major wafer surface [8]; 2) vertical coupling where light is turned at an oblique angle out of the major wafer surface by a deflecting structure such as a diffraction grating [9].
Both coupling approaches are used ubiquitously in photonics, and both have associated advantages and disadvantages.
Edge coupling is typically less polarization sensitive than grating coupling, and it is not subject to the fundamental tradeoff between coupling efficiency and operating wavelength range [10]. This tradeoff is a serious limitation in devices meant for application in wavelength-multiplexed systems. Edge coupling also facilitates a configuration in which the fiber cable exits the module approximately parallel to the printed circuit board (PCB) on which the module is mounted. This geometry simplifies low-profile routing of the cable to the backplane or front plane of the chassis holding the PCB. The principal drawback of edge coupling is that the PIC chips cannot easily be tested until they have been sawn and removed from the wafer, and perhaps even packaged.
Vertical coupling using diffraction gratings has the advantage that the grating can be designed to improve mode matching to the fiber, thus lowering loss. Also, devices can be tested before the wafer is sawn into individual PIC chips, thus allowing defective chips to be eliminated before further processing is wasted. Additionally, it is possible to fabricate fiducial marks on the chip surface to aid in alignment of connector elements to gratings. However, mounting a connector structure on the surface of a chip can consume valuable "real estate", potentially resulting in the need for a larger chip and fewer chips per wafer. Furthermore, the desire for horizontal routing of cables requires that some optical "turning element", e.g., based on a bent fiber, be mounted on the chip surface. This element can increase the height of the PIC package, and thereby increase board spacing and lower system density. Such solutions entered the market recently, e.g., PhotonicPlug by Teramount [11], however it requires wafer post-processing to integrate the optics.
Driven primarily by the desire to maximize the wavelength range available for wavelength multiplexing while minimizing optical loss, the work reported below focuses on edge coupling. A demonstration prototype package for a realistic reference PIC was designed and assembled with the following objectives: r Pluggable SM expanded-beam interface. r Optical coupling at the edge of the PIC. r Less than 2 dB waveguide to (standard) fiber cable coupling loss per facet.
r Horizontal mating of connector to the PIC module. r Fiber exit from module parallel to the PCB surface. r 250-micron pitch standard 12 fiber ribbon cable. r 8 channels of optical signal.
This publication expands on the results presented at ECOC 2022 [12]. Challenges in the assembly and performance of the prototype package are reported below. The results demonstrate that the approach can address the critical need for a manufacturable, low-loss, rugged, single-mode expanded-beam connector interface that is suitable for edge coupling to PIC modules.

II. REFERENCE PIC
Using fully-realized PICs in the development of materials, techniques, tooling and processes for attachment of the fiber array to a PIC is not a cost-effective strategy, due to high cost of fabricating and processing fully-functional chips. A new approach in photonic packaging is to develop general-purpose reference PICs [13]. These relatively-simple reference PICs mimic the essential properties of the fully-functional chips with respect to generic requirements for packaging interfaces: either optical or electronic. That is, the reference PICs duplicate features such as optical mode profile, port pitch, etc. which are relevant to package development. But since they are designed to be as generic as possible, a single reference PIC design can be used in the development of packaging for many different fully-functional chips. In this way, the NRE (non-recurring engineering) cost of the reference chip can be spread over the cost of many fully-functional chip designs. The result is that the cost per-chip for the reference PIC is approximately 20x lower and cost per mm 2 is correspondingly 10x smaller compared to a PIC from a dedicated MPW run. Additionally, the reference PICs can be supplied pre-characterized, thereby simplifying interpretation of packaging data obtained with them. Thus they can be realistic yet inexpensive substitutes to be used in development, benchmarking and optimization of photonic and electronic packaging processes, such as fiber or micro-optic attachment, wire bonding, flip-chip bonding and reflow.
The reference PIC used for development of the pluggable package prototype reported here was designed by the PIXAPP Photonics Pilot Line and fabricated by Lionix on the SiN platform. The design is shown in Fig. 1(a). The chip is very small (3.5 x 3.5mm 2 ), which allows many copies to be fabricated on a single wafer. It contains only features needed for development of optical packaging: a series of straight-through waveguides as well as two series of 4 loopback waveguides -one on each side of the chip. There are 3 short loopbacks (labeled S1-S3), where adjacent inputs/outputs are connected to one another, and 1 long loopback (L1), which connects the outermost inputs/outputs. The bend radius is 150 μm for S-and 130 μm for L-loopbacks. The total length of waveguide is around 3500 μm and 5000 μm respectively. The designed mode-field diameter (MFD) of the input/output edge couplers is 10 μm at 1550nm to match the mode size of standard single-mode fiber (SMF) thereby minimizing mode mismatch loss.
An important feature of the SiN platform, besides being transparent at visible wavelength (which expands the range of possible applications of SiN photonic chips), is smaller refractive index contrast between the waveguide and air as compared to silicon on insulator (SOI) waveguide devices (∼2:1 contrast as opposed to 3.4:1) [14]. Since the majority of optical adhesives have refractive indices in the range of 1.45-1.6, the lower contrast between the waveguide effective index and the adhesive index reduces the reflections at the interface and prevents optical cavity effects. SOI-based PICs typically require processing of the facet -either through polishing or very slow etching to create an optically-smooth surface to limit scattering. For the case of SiN, the PIC can be singulated using a standard dicing saw with fine grit (of the order of 2 μm), without the need for any additional polishing of the facet. A section of sawn wafer is shown in Fig. 1(b). While the residual roughness from this procedure introduces additional coupling losses in air, once the gap is filled with epoxy, the scattering is reduced and coupling efficiency improved.
The effect of index matching is shown in Fig. 2(a), which shows wavelength-dependent coupling efficiency (CE) averaged over the two input/ouput (I/O) facets of a loopback waveguide. The data presented take into account the 0.3dB average loss of an additional FC/APC connector that is present in the sample arm. Due to polarization sensitivity of the PIC, polarization control paddles were used at the input. At optimum polarization (TE) and in air, the losses are around 3-4dB, depending on the facet, with a small (0.5dB) wavelength dependence over the 200nm span. This relatively high loss value is due to scattering from the rough (not polished) as-diced surface of the PIC facet, as well as imperfect mode field matching between the PIC waveguide and fiber. With an index-matched liquid at the interface (Cargille Liquids Series A; n D = 1.552) the scattering losses are reduced, so the coupling loss decreases to 1-1.5dB and there is smaller variation between different channels.
We define alignment tolerance as the displacement (lateral or angular) at which the signal drops to 80% of maximum, i.e., a 1dB loss penalty. Fig. 2(b) compares the alignment tolerances of a standard fiber array to the reference PIC both in air and with the index-matching liquid (n D = 1.552) present. Here the X-direction is along the edge of the PIC while Y is normal to its major surface. The tolerances are in range of 2.1-2.7 μm, depending on the displacement direction, with tolerance along the X axis being more relaxed.

III. OPTICAL DESIGN AND SIMULATION
The tolerances shown in Fig. 2(b) are not sufficiently large for pluggable operation [5]. The losses resulting from attempted mating in this case would likely exceed 10dB, clearly illustrating the motivation for expanding the mode between the pluggable elements. Fig. 3 shows a Zemax simulation of edge coupling between a PIC and a fiber using micro-lenses. The sub-assembly consisting of the dedicated micro-lens attached to the fiber will be called a connector. The connector, along with the PIC microlenses (which are attached to the PIC using an optical epoxy), expand and collimate the modes from both sides of the optical interface (see Fig. 3(a)). As a result, the simulated displacement tolerance increases as a function of the beam size (taken at 1/e 2 ), as can be seen by the black curve in Fig. 3(b). The extent of mode expansion has a practical limit imposed by the pitch of the lenses. Since collimated Gaussian beams undergo diffractive expansion, in order to avoid interchannel cross-talk, the size of the expanded beam should not be greater than ∼55% of the lens/channel pitch. Thus the expansion limit for 127 μm tight-pitch arrays is ∼70 μm, while for 250 μm standard-pitch it is ∼140 μm. For such a wide beam, the lateral tolerances can exceed 30 μm, while at the lower practical end of expansion, such as for a 50 μm-wide beam, the lateral tolerance is 12 μm. The relaxation of lateral tolerances comes at the cost of tightened angular tolerance, as shown by the red curve in Fig. 3(b), therefore there is a trade-off in the mechanical design of the package: angular deviations will need to be better controlled as the beam gets wider.
Another effect of beam expansion can be seen in Fig. 3(c), which shows coupling efficiency as a function of axial distance between the lenses, i.e., the air gap. Wider beams allow the elements to be placed further apart (even several millimeters) and axial positioning tolerances are greatly expanded. Efficiency at a given separation can be optimized by proper choice of the conic parameter of the lens surface.
In principle, with proper engineering, the pluggable optical coupler can be made agnostic to the MFD at the PIC waveguide facet. This is achieved when both sides of the link couple to the same expanded mode. This is shown in Fig. 3(d), where axial displacement is simulated for different symmetrical mode sizes of the PIC waveguide output and for a 130 micron wide beam. The fiber and its micro-lens are kept the same and the only element that is being varied is the PIC lens to properly expand the PIC mode. The result is that the curves are all the same, therefore the performance of the coupler is the same regardless of the PIC. This can simplify the logistical aspects of photonic packaging, since the connector can be standardized across the whole industry and only the design of the PIC lens need to vary according to the waveguide characteristics of the PIC in the package.
In designing the edge-coupled pluggable prototype, we have assumed the MFD of the PIC to be 10 μm and circularly symmetrical. The optical design of the MTP connector (Molex VersaBeam [15]), which composed the other side of the optical link, was given by the part manufacturer, and it expanded the mode from a single mode fiber to around 50 μm. With these parameters, Zemax simulations allowed for selection of an offthe-shelf micro-lens from SUSS Micro-Optics (Cat. #18-00997) [16]. The lens is made of fused silica, and has 600 μm thickness and 192 μm radius of curvature. It has a broadband AR coating on the lensed side, which allows direct attachment to the PIC using UV adhesive of index n = 1.55 without forming a resonant cavity. Of several lens designs investigated, this one provided the best compromise between optical performance and mechanical design. The calculated lateral misalignment tolerances for the displacement of the PIC lens and the connector are 3.5 μm and 23 μm respectively. The simulation was performed at an air gap of 2.6mm, where the minimum loss is expected to be 1.02dB. The performance can be marginally improved at smaller gap values (to 0.84dB insertion loss at 1.4mm distance), but smaller gaps were unachievable due to mechanical constraints, as discussed in the following section. Due to the use of an off-the-shelf lens not specifically designed to the application requirements, it is expected that the optical link will have somewhat degraded performance.

IV. MECHANICAL DESIGN
The 3D CAD model of the prototype pluggable package for the reference PIC is shown in Fig. 4. The mechanical package consists of a main aluminum housing, a commercial MTP Array Adapter and a separate housing lid (not shown) which is added in the final package assembly step ( Fig. 4(a)). The PIC is bonded with adhesive to a ceramic sub-mount, yielding along with the micro-lens attached to the PIC a "PIC sub-assembly". This sub-assembly is held by the assembly system ∼100 μm above the bonding surface of the aluminum housing, in rough alignment with the alignment pin holes shown in Fig. 4(b). The space between the subassembly bottom and the bonding surface in the housing is provided to allow clearance for motion during fine alignment of the sub-assembly in a subsequent process step, as discussed below. The MTP adapter is attached to the housing with screws, and requires no fine alignment. An MTP Male Array Connector is plugged into the MTP adapter, which provides coarse alignment between the connector and the subassembly. It also provides a compliant pluggable force to the MTP connector in the semi-constrained axial direction as well as strain relief for the connector.
The aluminum housing is designed with a cut-out through which optical signals can travel into and out of the package. This slot is referenced to precision alignment holes on either side, which mate with the alignment pins on the MTP connector ( Fig. 4(c)), thereby providing repeatable high-precision registration between the connector and the aluminum housing. These precision alignment holes were measured to be machined to ±22μm center-to-center spacing tolerance as shown in Fig. 4(d). The holes provide an interference fit with the MTP connector constraining 5 degrees of freedom to 0.022mm parallelism of the pluggable MTP to the aluminum housing base.
The vertical locations of these alignment holes above the aluminum housing base were designed to provide enough clearance for active alignment of the PIC sub-assembly to the axis of the MTP connector micro-lenses. To avoid needing excessive adhesive to affix the PIC subassembly to the aluminum housing, a 1-D tolerance analysis was carried out to minimize the height of the alignment holes above the aluminum base. The error stack-up consisted of the machining precision, ceramic sub-mount thickness variance, PIC attachment to the sub-mount variance, variance of the PIC waveguide height and lateral optical beam shift, and the variance of the adhesive thickness. The tolerance analysis of these variances gave the required 3σ clearance needed during the active alignment between the PIC sub-assembly and the expanded beam connector alignment holes in the aluminum substrate. This tolerance allows enough room for the active optical alignment assembly to compensate for the errors in the mechanical components. In this design, the analysis yielded a nominal gap between the PIC subassembly and the housing of 100 μm.
To avoid collisions between the pins and the PIC edge facet, the length of these alignment pins sets a minimum optical working distance of 2.6mm, which is the reason for using non-optimal lens-to-coupler distance. If not for this requirement, this working distance could be reduced to ∼1.4mm, which falls within the target working distance of the optical design. The required clearance between the PIC and the alignment pins limited the dimension that could be used for the mechanical design of the alignment pin holes. The combination of the MTP adapter and the mechanical housing alignment holes, despite their limited size, provided a robust assembly capable of being successfully mated with the MTP connector alignment pins repeatedly over multiple connect/disconnect cycles.

V. PACKAGING PROCEDURE AND RESULTS
Two package prototypes were fabricated using the above design and process approaches (P1 and P2 respectively). P1 was used for initial process design and validation, therefore not all optical channels were evaluated on P1. P2 utilized an improved assembly and attachment procedure that produced better stability and allowed testing of more optical channels to demonstrate feasibility of multi-channel operation.
The PICs were first cleaned using acetone and isopropanol to remove residue from the dicing procedure, and then bonded with adhesive to a larger, 6x6mm 2 ceramic substrate.
The optical alignment procedure was performed on a nanosystec NanoGlue system, equipped with two 6-axis stages which allow for high-precision alignment (better than 0.1 μm translation and 0.005°rotation). For the prototype development, the process was not automated. There are two phases in the optical alignment procedure. In the first phase, the micro-lens array is attached to the PIC, forming a lensed PIC assembly. For this, 1550nm SLD light is circulated from the connector (Molex VersaBeam) through the PIC lens array and the loopback waveguide on the PIC, and back out into another channel of the coupler. The signal is maximized by adjusting the positions and rotations of both the PIC lens and the connector which are held by separate high-precision 6-axis stages, while the PIC lens is fixed in place. The distance between the lenses during this procedure is fixed at 2.6mm. A fiber-based paddle polarization controller is used at the input to optimize transmission. The lens is fixed to the PIC using DELO DualBond OB6268 UV-curable epoxy.
In the second phase of the optical alignment procedure, the MTP connector is plugged into the MTP adapter in the mechanical housing and the position of the PIC assembly is optimized in the same manner as in the previous phase and then attached to the aluminum housing using the same DELO OB6268 epoxy. Fig. 5 shows experimental alignment tolerances of various elements of the optical link, measured in the first phase of the packaging procedure. While the alignment tolerance of the micro-lens to the PIC (solid lines in Fig. 5(a)) is same as for alignment of a fiber array to the PIC, 2.35 μm, the connector allows for the tolerance to be expanded to 14-15 μm, as seen by the dashed lines in Fig. 5(a). As noted earlier, by expanding the beam the angular alignment tolerances between the beams are tightened, however these are typically easier to control than lateral alignment tolerances since alignment machines are capable of delivering angular alignment accuracies an order of magnitude better than 0.17°, which is the worst-case scenario in Fig. 3(b).
The connector distance (Fig. 5(b)) tolerance is relaxed as well, to 1.7mm from the original position. Since bringing the connector closer to the PIC would risk contact between the alignment pins and the PIC lens gripper during the alignment procedure, coupling at closer distances was not measured. The distance between the PIC and its lens (Fig. 5(c)) can be comfortably varied within typical capabilities of a packaging machine without losing significant coupling. Shrinkage of the epoxy (0.7% vol. [17]) should not lead to significant change of the gap during curing; in any case, the parameters requiring precise control during packaging are the transverse position (X-Y plane) and the angles.
Measurements of coupling efficiency take into account 0.3dB loss due to an FC/APC connector in the sample arm, as well as two MTP connections, each contributing 0.16dB insertion loss on average as measured by the part manufacturer. The optical fiber connection configuration is shown in Fig. 6(a). Fig. 6(b) shows the evolution of coupling efficiency at different stages of micro-lens attachment. In air, at 1550nm, the coupling loss is around 1.9dB per coupler. With epoxy between the PIC and the lens, the losses drop to 0.7dB/coupler (Arrow 1 in Fig. 6(b)).
Comparing these values to ones achieved using standard fiber ( Fig. 2(a)) we observe improvements in coupling: ∼1dB in Air and 0.5dB with epoxy. This indicates that the mode of the PIC is not matched to the fiber and with a lens better matching can be achieved by changing the distance between the lens and PIC. After curing the epoxy (Arrow 2 in Fig. 6(b)), a small decrease in coupling is observed, on the order of 0.2dB, that is attributed to shift and rotation of the lens due to shrinkage of the adhesive. During Phase 2 (Arrow 3 in Fig. 6(b)), where the PIC assembly is being placed and cured in place in the package, further loss of signal is observed, to 2.3dB. This is again attributed to shrinkage of the epoxy, however with process development (e.g., by incorporating "windage" into the alignment process to  compensate movement due to shrinkage or active re-alignment during the curing process) these losses can be minimized and the package can potentially exhibit sub-1dB losses.
Both prototypes exhibited similar behavior during packaging, however with P1 the losses were mostly due to epoxy laterally displacing the PIC assembly out of the optical axis, while in the case of P2, the assembly was rotated. These differences stem from the differences in primary attachment points between the assemblies. For P1, the PIC assembly was attached to the bottom of the aluminum housing, while in P2 the edge of the ceramic submount was attached to the vertical face of the housing, close to the alignment holes.
As mentioned earlier, for the first prototype (P1) only one channel (S1) was tested, yielding a loss of 2.3dB/coupler. In the other prototype (P2), all channels were tested and uniformity was assessed. Table I shows coupling loss measured at 1550nm for all the loopbacks in P2. Variation is around 0.86dB, and could be a result of variation between the couplers on the PIC, or possibly rotation of the micro-lens array during curing. Note that in the final assemblies, of the five loopbacks tested, two satisfied the requirement of sub-2dB loss per coupler. Fig. 6(b) however strongly suggests that with improved design, better process control and automation, the losses per-coupler can achieve much better performance, even sub-1dB.  The final assembled package is shown in Fig. 7 without-and with a cover.

VI. PLUGGABLE CONNECTOR PERFORMANCE
Both prototypes were tested for loss reproducibility in repeated mating of the pluggable interface immediately after the packaging procedure was completed. For P1 the number of mate/de-mate cycles was 30 and for P2 -50 cycles. Additionally, P2 pluggability was tested the following day with 30 additional cycles. For P2, only 1 of the channels was tested (S3). The resulting loss histograms are shown in Fig. 8.
In the case of P1 (Fig. 8(a)), the majority of matings (2/3) yield losses in the 2.4dB range with a high-loss tail extending to 3dB. The high-loss tail may be due to mechanical wear of the alignment slots in the aluminum housing.
In the case of P2 (Fig. 8(b)), there is a wider spread of loss values, however the values are still centered around 2.3-2.4dB. There is also negligible difference between measurements taken on different days.

VII. DISCUSSION
While the performance of the pluggable connector was satisfactory in terms of coupling efficiency and reproducibility, there was initially a discrepancy between the calculated and measured average tolerances of lateral displacement of the MTP connector: 23 μm and 14.5 μm respectively. Additionally, we observed a small difference in the shape of the coupling efficiency as a function of displacement in the x-and y-directions (see Fig. 5(a)), which suggests that the mode at the edge coupler is not circular.
In order to investigate these discrepancies, we have experimentally measured the size of the PIC mode. The experimental set-up is presented in Fig. 9. We have used an Allied Vision Goldeye G130 TEC1 InGaAs camera with 1280x1024 pixel array and a 5 μm pixel size. The imaging objective was a Mitutoya 20x with NA of 0.4. To access the waveguides used in the prototypes, given the space restrictions, we have selected the long loopback L1 and have aligned a single-mode fiber to the waveguide input facet; at the output we have placed a beam-splitter cube. The PIC and the cube sat on a 6-axis hexapod platform to align the PIC with the fiber.
In order to calibrate the camera, we have substituted a singlemode fiber for the PIC waveguide in the same configuration. Knowing the mode size of the SMF at 1550nm (10.4 μm) we have calibrated the image magnification to be M = 0.398 μm/pixel. Applying this calibration to the reference PIC, we arrive at MFD values of 6.8 μm x 6.2 μm, where the wider dimension is along the chip edge, i.e., X. Since these measured values are substantially smaller than the ones assumed during the simulations performed to select the lenses used, it is possible that the disagreements between simulated and measured optical performance and tolerances are due to inaccurate input parameters to the simulations.
We have also measured the MFD of the beam leaving the MTP connector using a beam profiler. The value came out significantly larger (>20%) than predicted in our initial model 1 . 1 Optical design parameters of the VersaBeam© connector are a trade secret of Molex©, as such we cannot publish the theoretical and experimental values. Using new experimentally-measured values of beam sizes, we re-visited the ZEMAX model. Fig. 10 shows a comparison between the revised model and the experiment. Due to the fact that the optical system is not optimized and both sides do not couple to the same expanded beam, the properties of the link will depend on the direction the light takes -either into or out of the PIC. In the PIC-to-Connector direction the expanded beam is wider and therefore has a more relaxed alignment tolerance of around 19.45 μm. In the opposite direction, a narrower beam results in tighter theoretical alignment tolerance of 13.7 μm.
In the experiment the light travels through the loopback, therefore the measured tolerance is a product of contributions from both directions of propagation, which undergo identical displacement. Experimental data in Fig. 5(a) was averaged over both directions by taking a square root of the raw translation curve. It represents a single-coupler tolerance in a case where the optical link is symmetrical, i.e., when both sides couple to the same expanded beam. The follow-up investigation revealed high discrepancy in the assumed and measured optical parameters of the PIC and the MTP connector lens. The product of the PIC-to-Connector and Connector-to-PIC contributions is shown as dashed line in Fig. 10. The resulting 1dB tolerance is 11.4 μm. This is comparable with the experimental data, shown in Fig. 10 as a blue line, where the tolerance is 9.1 μm. In the original theoretical model, the product tolerance was 20 μm, more than 2x the experimental result.
This discussion reveals possible difficulties in system design. In the re-visited model the discrepancy is smaller, however the simulation still does not fully reproduce the experimental curve. It has to be noted however that while the new model relies on experimentally-reverse-engineered parameters of the PIC and MTP connector, there are still a few imprecisely known parameters, such as thickness of the epoxy in the gap between the PIC and the lens. This ambiguity stems from the fact that during the active alignment the epoxy thickness is not pre-determined, but rather is used to compensate for less-than-ideal MFD mismatch between the components and optimize the coupling efficiency. This makes the theoretical comparison challenging. The issues created by a propagation direction dependent link can be mitigated by redesigning one, or both sides of the connection to produce the same beam parameters, regardless of direction.

VIII. CONCLUSION
We have demonstrated a pluggable single-mode fiber-to-PIC connection. By using micro-lenses to relax the alignment tolerances by a factor of 6 in comparison to typical butt-coupled connections (14 μm using micro-lenses versus 2.35 μm using standard fibre), we were able to fabricate a housing that allowed for efficient (∼2dB) multichannel connections, with a good coupling reproducibility over up to 80 mate/de-mate cycles. We have also highlighted specific design issues that have to be taken into consideration when developing a contactless, plugabble optical link based on micro-lenses.
While active alignment was used for assembly of these first two prototypes, future work will concentrate on designing scalable approaches which will allow for passive-alignment-only assembly to the required tolerances, e.g., machine vision, fiducials and mechanical stops for automated assembly. Reduction of misalignment resulting from the shrinkage of adhesive during curing can be addressed by using reduced-shrinkage adhesives (e.g., with nanoparticle loading), and/or by controlling the geometry of the adhesive body to minimize the net shrinkage-related misaligning force working on the optical interface. Additionally, the redesign of optical elements can lead to reproducible pluggable <1dB loss per coupler. These methods, when incorporated into Design Rules and Assembly Design Kits [13], [18] and standardized across the industry, provide a path to significantly reducing the costs of photonic packaging.