Packaging of micro-lens arrays to photonic integrated circuits using beam shape evaluation

We propose a method for aligning and attaching micro-lens arrays to photonic integrated circuits (PICs). Unlike the conventional approach of assessing power coupled to a fiber directly, our method utilizes a beam profiler. This profiler allows us to optimize the lens position by analyzing the transmitted beam shape from the PIC edge coupler through the lens. In conjunction, we employ grating couplers to introduce external light, acting as a ‘beacon’ for optimization. The use of grating couplers enables efficient coupling of external light into the PIC, providing a reference point for alignment. Importantly, our method accommodates both regular waveguide-side-up and upside-down (through-Silicon) orientations of the PIC. This versatility allows us to reproduce coupling results across a 6-channel array, demonstrating robust performance. This innovative approach not only ensures precise alignment and attachment but also opens up new possibilities for photonic packaging. The flexibility to work in different orientations is likely to lead to advancements in the design and assembly of photonic devices.


Introduction
Photonic packaging is a critical aspect of the photonic ecosystem following on from a rapid growth of Photonic Integrated Circuit (PIC) based technologies.There exist numerous highly innovative PIC-based solutions for a wide range of applications and markets such as sensors [1,2], telecoms [3,4], bio-medical [5][6][7] and LIDAR (Light Detection and Ranging) [8].However, as chips become more complicated, there are assembly and packaging bottlenecks that impede the commercialization of the full systems.Compared to micro-electronic devices, where assembly and packaging can take 10%-20% of the overall manufacturing cost, in the case of photonics, this typically amounts to more than 50% [9][10][11].This cost gap will become even more important as nascent artificial intelligence (AI) technology will require an order-of-magnitude increase in demand for high-throughput transceivers [12][13][14].
There are several reasons for the high cost of photonic packaging.One of them is the necessity of chip pigtailing, wherein the fibre array (FA) is permanently attached to the PIC using an adhesive.Despite being the current state-of-the-art, it has several drawbacks.FAs are bulky in comparison to the chip, which presents a logistic challenge for loading and unloading in the packaging machine and subsequent handling of the module as well as cable management.Additionally, the temperature sensitivity of optical adhesives makes it a challenge for any reflow procedure to be utilized for the module.Some of the handling issues can be mitigated using short stubs of fibre (several millimeters up to a few centimeters) terminated with pluggable fibre connectors.Such is the case in standard QSFP (Quad Small Form-factor Pluggable) transceiver packages which utilize LC (Little Connector) or MPO (Multi-fibre Push On) connectors [15] or TeraPHY TM I/O chiplets that use an MPO connector [16].
Another solution is to utilize micro-lenses (MLs) to expand and collimate the beam between the PIC and FA.This method leads to relaxed alignment tolerances which allow for 'truly pluggable' photonic packaging solutions and was previously demonstrated for a grating coupler (GC) [17] and edge coupler [18] paradigms.However, the methods of attachment of the MLs to the PIC are still time-consuming and expensive, therefore the benefit of PIC pluggability cannot be utilized on a commercial high-volume scale.There are methods of 3D-printing of MLs directly on the PIC facet [19], although they bear severe limitations.The main issue is that it is much more difficult to expand the beam to sufficiently large mode-field diameters (MFDs upwards of 50 µm-a minimal requirement for pluggability) because the lenses would have to be very long.For example, optical simulations for a 10 µm mode adapter at 1550 nm suggest that the printed lens would need to be at least ∼400 µm long, while for a 3 µm mode adapter this would be 115 µm, however at a cost of a small radius of curvature.This lowers the reliability.Additionally, not all polymers used for 3D-printing are reflow-compatible, limiting their use to lower-temperature applications.
The work reported below outlines a method of attaching a micro-lens to the edge couplers of the PIC which measures the parameters of the expanded beam to align the lens.Instead of coupling light from the front and using loopback power optimization [20,21], light is instead coupled through a grating coupler, which is connected to an edge coupler.The method has a potential for scalability, as well as allowing for alignment through the Silicon substrate for a simplified packaging process of increasingly complex package designs, e.g.highly-integrated co-packed photonic devices [22,23].The work is divided into several sections.First, we perform baseline measurements using a standard FA (see section 2 and figure 2).In the next step we assemble a reference package with micro-optics, which will be used to define the beam parameters necessary to perform packaging using the new approach (see sections 3.2, 4.2 and figure 3).Following that, we perform an assembly of a micro-lens to the PIC utilizing the grating coupler approach (see sections 4.3, 4.4 and figure 4).

Reference PIC and baseline measurements
The PIC employed in this work was fabricated at Ligentec on a SiN platform.The schematic is shown in figure 1(a).The PIC size is 5 × 2.4 mm 2 and it contains several loopback structures (labelled L1-L3) which optically short two adjacent edge couplers (ECs).On both sides of this array, there are edge couplers which are connected to grating coupler (GC) structures (channels 1 and 8) on the opposite side of the chip.Figure 1(b) shows a magnified image of the grating couplers and the waveguide taper, while figure 1(c) shows the edge couplers.The angle of incidence of the GCs is 10 • and the pitch of the waveguides is 250 µm.All experiments are performed at 1550 nm using a 1 mW Thorlabs© S5FC1550S-A2 Benchtop SLD Source.A standard FA is aligned to the PIC with Dymax© OP-4-20632 adhesive in the optical path to obtain baseline coupling efficiencies of the edge couplers and to measure the alignment tolerances.These results are shown in figure 1(d).The average per-coupler EC losses are 2.8 dB with a small ±0.1 dB variation between different loopbacks.The spectra are flat across a wide, 150 nm band.The coupling efficiency to the grating couplers is measured using an 8-channel FA with the facet polished at 40 • to satisfy the 10 • angle of incidence [20].The spectra are then normalized using the EC data.Grating couplers are significantly more wavelength-dependent with large power variations across the 150 nm bandwidth.At peak, the coupling efficiency is around 3.5 dB.
Typical tolerances of alignment of a single-mode fibre to the ECs are presented in figure 1(e).The curves in the x (along the edge) and y (surface-normal) directions are not identical, with the 1 dB tolerance in the x-direction being 2.5 µm and 2 µm in the y-direction.The mode-field diameter of the edge coupler spot-size converters is around 6-7 µm with a larger mode size in the horizontal direction, leading to the asymmetry of the tolerance curves.With the fibre MFD equal to 10.4 µm, the mismatch is the primary source of coupling losses.

Photonic packaging methods
The standard method of alignment of FAs to the PICs utilizes, in the simplest case, two sets of loopbacks on the opposite sides of the coupling array [20,21].In such an arrangement it is relatively easy to find 'first light' on one of those loopbacks during alignment and optimize the signal.A loopback on the other side of the array is subsequently aligned and the rotation of the array is corrected to match the PIC.The larger the distance between the alignment structures (i.e.loopbacks or GCs as used in this work), the more precisely the rotation can be corrected.This approach requires an overhead of additional 4 'passive' channels for optical alignment.This adds to the cost of the packaging and reduces the space available on the PIC for active channels.This can be mitigated by using active components on the PIC, either light sources or detectors, to sample light.This however requires electrical connectivity and sometimes thermal control during packaging.Additionally, not all platforms are capable of source integration, InP being the notable exception, however, this is currently changing with the rise of heterogeneous integration techniques, such as transfer printing [24][25][26].

Utilization of grating couplers in packaging
The new method, presented in this paper, is to utilize GCs routed to an edge-coupler to align and attach optical elements such as MLs and standard FAs to the PIC facet.In this method, the outer channels of the PIC coupler array connect the EC to a GC as shown in figure 1(a).When the light source is coupled to those GCs, the outer edge couplers will serve as a light source, or a 'beacon' for active alignment of the FA, replacing the need for an active on-PIC element (laser or detector) and reducing the complexity of fabrication.Therefore, in the outlined method there is no need for active electrical monitoring or powering of the PIC.
Figure 2(a) shows this process performed in laboratory conditions on a reference PIC using a Nanosystec© Nanoglue, equipped with dual 6-axis stages which allow for high precision of alignment (better than 0.1 µm translation and 0.005 • rotation accuracy).One stage of an alignment machine positions an 8-channel FA against the PIC facet and another stage positions an array for light injection into the outer channels.As the machine does not require much light to perform the active alignment of the FA, there is no need to employ a planar angled-facet FA (as is done for further measurements) and align it precisely to the GCs with single-micron tolerances.It is therefore replaced with an array which has a micro-lens array (MLA) attached to the front that collimates and expands the beam to ∼130 µm at 1550 nm.This reduces the time necessary to find first light, allowing the procedure to be executed much faster.The losses in this case are presented in figure 2(b).The data for the coupling efficiency through the loopbacks on the reference PIC is the same as presented in figure 1(d), showing the viability of the proposed method.We note that the transmission through the GCs is >20 dB lower than using an angled-facet FA (compare to figure 1(d)), however this is sufficient for alignment purposes.

Methods for packaging of micro-lens arrays
The extent of improvement of the proposed scheme to the automation of standard packaging (i.e.pigtailing) of PICs will be independently investigated in the future.The purpose of this paper is to provide a feasibility  study of utilizing the GCs for attachment of a MLA to the PIC by characterising the beam profile (BP).The proposed method offers potential advantages over the standard MLA attachment method cited in [18] in terms of speed and scalability through automation which will be investigated in future work.
Currently, the MLA-to-PIC packaging process is similar to the alignment of FA-to-PIC with loopbacks being ubiquitously used.The photograph of such an alignment procedure is shown in figure 3(a) and schematically in figure 3(b).This process is quite complex, as active simultaneous alignment of 3 elements is required: the PIC, the MLA and the pre-assembled FA (with attached MLA) used for probing of the loopbacks.The procedure involves angle correction of each component followed by many iterations of position optimisation of MLA and probing FA to obtain maximum coupling.This is a time-consuming process which is not conducive to scalable production.
The proposed method of using the grating couplers as a 'beacon' light source to illuminate the edge couplers along with a BP has the potential to reduce the time required to align and attach a micro-lens to the PIC.BPs have been used previously in photonic packaging for the alignment of micro-optics (typically ball lenses) to lasers [27,28] and for PIC-based free-space applications, e.g.bio-medical [29].To the best of the authors' knowledge, the method proposed in this paper was not used in the context of improvement of the packaging process of PICs.
Figure 3(c) shows a schematic of the proposed process.Light is injected from the outside source into the grating coupler, which illuminates the connected edge couplers.The edge couplers on the PIC facet are facing a BP, while a gripper manipulates the MLA in front of the PIC facet.Two parameters need to be monitored by the profiler: the size of the mode in 2 dimensions, as well as its shape (e.g.eccentricity).The combination of these parameters is an indicator of the relative position of the micro-lens with respect to the coupler: the further the lens is from its optimal position in the x-y directions, the more distorted from the optimal the resulting mode will be, while the z-direction will mostly determine the overall mode size.The packaging process begins with a preliminary visual angle correction of the lens with respect to the PIC, followed by coupling light into the GCs and manipulating the lens in cardinal directions to achieve the desired mode parameters.
In the proposed scheme, the position of only one element-the MLA-needs to be optimized using active alignment methods as the initial coupling to the GC on the surface of the PIC is much easier to perform and can be done passively e.g. using machine vision [30].If both, the standard and proposed, MLA attachment methods were fully optimised for automation, the full alignment and MLA attachment cycle of the proposed method would be faster, as it replaces active co-alignment of 2 elements with a combination of a passive and active alignment, making the method better-suited to large-scale manufacturing and commercialisation.
As flip-chip bonding is becoming more prevalent as an electronic and photonic packaging solution (an alternative to wire-bonding) [31,32], the waveguides and couplers are mostly hidden from view during the packaging process.Design and assembly of such a package thus brings more challenges as vision is necessary for alignment and required for automated machine vision systems to identify features on the PIC.The proposed method of using GCs can mitigate those issues by coupling into the waveguides through Silicon-either to attach an FA or a micro-lens.This process is shown schematically in figure 3(d).
In this paper, we demonstrate the laboratory-based implementation of the described GC process and compare the results to the standard process.Additionally, we demonstrate the implementation of the procedure on the PIC flipped upside-down, i.e. through Silicon (GCTSi).The critical parameters are the reproduction of the same expanded mode size of the PIC assembly and a comparison of coupling efficiencies (CEs) to a micro-lens equipped (probing) FA.

Preparation of the fibre array
To measure coupling efficiency into the PIC that will be equipped with a micro-lens, as well as to perform the standard method of attachment as described further in section 4.2, we first assemble the probing FA by attaching a MLA to the facet of the fibre.This is done using a retro-reflector method, where a mirror is positioned in front of the facet of the FA and the two surfaces are made parallel to each other.Next, a MLA is positioned in between and all the angles are aligned to the FA facet.Light from a source is passed through a circulator, collimated by the MLA, reflected off a mirror and coupled back into the same channel.The 3rd channel of the circulator is connected to a photodetector, which allows for active alignment and optimization of the MLA position.The coupling efficiency is optimized for a 6 mm air gap, which is the distance between the lens and the BP sensor.The lens is attached using a UV-curable Dymax© OP-4-20632 adhesive.UV curing is performed using Thorlabs© M365L2 LED source equipped with a collimating lens.
For all the assemblies used in this work, MLAs fabricated by Axetris© are used.These are 8-channel arrays made of fused silica, with 1 mm thickness, radius of curvature 315 µm and conic parameter k = −0.7.The lens pitch is matched to the PIC at 250 µm and the diameter of the lenses on the surface is 240 µm.These lenses, when placed on a single-mode FA (with an MFD of 10.4 µm at 1550 nm) produce a beam 135 µm-wide.These lenses were utilized previously for a pluggable package demonstrator [17].

Standard method of attachment of micro-lens array
After the probing FA has been assembled, we attach a MLA to the PIC using the method and shown schematically in figure 3(b) and on a photograph in figure 3(a).The PIC is placed on a stage and held by a vacuum.The FA is held by one pair of grippers in front of the PIC, while angled grippers are used to position the MLA.The coupling efficiency is optimized for a design air gap of 6 mm, the approximate distance between the PIC assembly and the BP sensor.The optimization of the lens position is performed for the loopbacks on the opposite sides of the array (i.e.L1 and L3) to align the entire array.The MLA is secured to the PIC by an optical adhesive.
Following this, the prepared PIC assembly is positioned facing the BP and light is coupled into the GCs while we observe the shape of the beam on the outer channels (i.e.GC#1 and GC#8).The Nanoglue is integrated with a Thorlabs BP209-IR Dual Slit Scanning BP, which allows for single-micron profile resolution.The input aperture is 9 mm, which means that in principle both outside channels can be observed at the same time, as they are separated by 1.75 mm.The measured spot size at a distance of ∼6 mm is 141 µm in the horizontal (x) direction and 187 µm in the vertical (y) direction.The asymmetry is due to the ellipticity of the mode, magnified and inverted by the lens.We note that while the standalone software allows full-frame mode analysis and is used to measure the reference beam width, the BP module integrated with the Nanoglue alignment software only returns the beam 1/e 2 width in x-and y-directions.This limits the information available during MLA alignment.

Grating coupler method (GC)
Figure 4(a) depicts a photograph of the alignment of an MLA to the PIC using the BP.PIC is placed on a stage and held by a vacuum.Angled grippers are used to handle the MLA.The assembly is facing the aperture of the BP at a distance of ∼6 mm.Due to space constraints, we cannot in this case use a collimated beam to couple light into the GCs (compare to figure 2(a)), opting instead for a standard 40 • angle-polished FA. Figure 4(b) shows a side-view of the experimental setup.
The MLA is first aligned to the PIC visually, including correcting all the angles before alignment using the BP is considered.Figure 5 depicts the width of the beam as measured by the BP in the (a) x-and (b) y-direction respectively as a function of the displacement of the micro-lens with respect to the edge coupler.We use a logarithmic scale to showcase small differences in beam y-width that occur close to the lens centre.The lens outline can be clearly discerned and is shown as a dashed black line.Small defect on the lens (dust particle) can be also observed.Figures 5(c) and (d) show the corresponding beam width slices taken through the centre of each lens in linear scale.Dashed lines are the beam width target values taken from the reference measurement, i.e. 141 µm in x-and 187 µm in y-directions respectively.The beam width in x (figure 5(c)), exhibits a global minimum at the lens centre when scanned in the x-direction, while the y-direction scan shows quite a flat dependence within the central area of the lens with a small, 4% difference at ±50 µm.For the beam width in the other direction (figure 5(d)), the beam is not perfectly collimated therefore the local maximum at the centre of the lens is pronounced (lighter blue area in figure 5(b)) with 11% difference at ±75 µm.
The starker features occur at the edges of the lens, where the beam switches from passing through one lens to the adjacent.Moving away from the geometrical centre, close to the edges, the beam width in both directions increases sharply: 4-fold in the x-direction and up to 7.5-fold in the y-direction, depending on scan direction.Additionally, at the exact boundary there is a sharp maximum, most significantly in the y-direction.The distance between the extrema is exactly the pitch of the lens of 250 µm.From these graphs, we can find the centre of the lens in x-direction by scanning the beam width in y and, correspondingly, the centre of the lens in the y-direction can be found by scanning the beam width in x.The entire position correction procedure needs to be performed on both alignment channels, i.e. #1 and #8 to correct the rotation of the lens and align it to the coupler array, however-as mentioned previously-this could be performed simultaneously since both spots land within the BP aperture.
Finally, the optimal distance (i.e.z-direction) between the lens and the PIC can be obtained by directly measuring the beam width on the BP and changing the lens distance until the desired size is achieved.If the differences in these values are significant (more than a few microns), the entire procedure, including x-y optimization, might require several iterations.After the optimal position is found, the MLA is attached to the PIC using the same adhesive as in the standard process.

Grating coupler method-through silicon (GCTSi)
When the PIC is assembled face-up in the package, it allows simple machine vision to be employed to roughly align the lens (or FA) to the PIC using features such as couplers, alignment markers or lens facets.In this case, though, the electrical packaging is limited to wire bonding to the pads at the edge of the PIC.The increasing number and speed of electronic connections necessitates however a switch to solder ball bonding, utilizing a ball grid array in the same manner as in microelectronics.This means that, going forward, an increasing number of assemblies will be performed face-down, making the optical packaging much more challenging due to limited visibility.It requires a special package design, with a window at the bottom of the package through which the visual alignment can be performed with the package being additionally placed upside-down in the packaging machine.This is a significant complication to the design of the package and the packaging process.
Using grating couplers to through-Silicon coupling of light (GCTSi) has been previously employed [33].In our work, we are employing this concept to align and attach the MLA to the PIC.In this method, the reference PIC is placed upside-down on the alignment stage and held using vacuum (in the future it will be bonded to e.g. the electrical interposer).Similarly to before, the edge couplers are facing the BP and the grippers are handling the lens array.The light is coupled using the same facet angle-polished FA as for the GC method.Inset in figure 4(b) shows a top-view of the alignment, which shows that the bottom face of the PIC is not mirror-polished, nor is it very rough.
In these circumstances, finding the first light is difficult, as there are no alignment markers or any other visual cues.Figure 6 depicts the optical power incident on the BP as a function of displacement of the fibre in the edge-parallel (x) and optical axis (z) directions.The wide band is light that is coupled from the fibre through the back facet of the PIC (shown in the inset in figure 6) and is guided by reflections at the Si/Air interface.There are also some artefacts present in the scan window, which are not sensitive to polarization rotation and therefore disregarded.There is only one feature that shows polarization dependence, which is recognized as light being coupled through the GC.For future applications, one can use PIC geometry, e.g.distances from PIC corners to the GCs, or process the bottom side of the PIC to include alignment markers for fast machine vision alignment.
Additionally, in the flip-chip regime, the coupling losses between the fibre and the coupler are much greater, which is due to several factors.One of these is reflections at the Silicon/Air interface, amounting to around −1.61 dB (30%) losses and then at the Si/SiO2 for further −0.81dB losses.Another source is scattering at the rough Si/Air surface.The final source is the coupling at the GC, which occurs at a different AOI due to refraction through the Si substrate.Due to this, the power of the SLD is increased from 0dBm to +4 dBm to achieve similar power impinging on the BP as in the previous method.
Following the discovery of first light, the rest of the alignment procedure follows the same steps outlined in section 4.3.

Comparison of the methods
Figure 7 shows per-coupler coupling efficiency measured for each L1-L3 loopback, for each of the described methods of attachment of the MLA to the PIC edge: standard process utilizing probing FA (figure 7(a)), grating coupling process (figure 7(b)) and coupling process through-Silicon (figure 7(c)).All the measurements are performed after the packaging process is completed and utilizing the probing FA (FA + MLA).Typically the losses across a broadband spectrum are lower than 3 dB, with few outliers (L1 in standard process).
Figure 7(d) compares the L2 coupling spectra for all the processes, also including the baseline measurement using a pigtailed FA.We observe that between all the methods the coupling efficiency per-coupler is within 0.4 dB range across a 200 nm bandwidth, showing excellent performance-limited only by the design and fabrication limitations of the edge coupler mode converter.Previously, it was shown, that the losses can be sub-1 dB with an optimized optical design [18].
To further demonstrate the consistency and reliability, in figure 7(e) we plot the coupling efficiency at 1550 nm for all presented methods and loopbacks.It shows a very good performance, with only one coupler crossing the −3 dB threshold.The presented results demonstrate that the new methods of packaging using GCs outlined in this paper reproduce the standard method very well and all are comparable in performance to chip pigtailing using a FA.We can successfully align the lens to within 1 µm of the optimal position, as well as correct the rotation, which enables consistent coupling across the entire array.
Table 1 compares the measured beam widths at GC#1 channel.We show that the GCTSi process reproduces the reference value to within 1 µm.The GC process shows a <10 µm deviation in the width of the beam in the y-direction.This can be either due to coupler/MLA fabrication tolerances or imperfect alignment of the MLA to the PIC.This deviation however does not lead to any significant degradation of coupling performance, as demonstrated in figure 7, showing that the method can be quite robust.

Conclusions and outlook for process optimization
We have demonstrated for the first time a method of attachment of an MLA to the PIC using GCs as a 'beacon' to highlight the edge coupler locations.This approach offers multiple advantages over conventional means and enables high-speed alignment without the need of inline co-alignment of several optical elements (FA + MLA, MLA and PIC).We are able to reproduce coupling efficiency results across the entire 6-channel array.
The new method simplifies the process by removing the probing fibre and substituting it with a BP, replacing power optimization with beam shape optimization.This way, the process can be made simpler and faster, as only the position of the MLA is optimized.The light source still needs to be aligned to the GCs, but this can be simplified using several methods.Surface features are easier to automatically align to using machine vision.While we use a standard array for this purpose due to the limitations of the packaging machine, it can be re-designed e.g. to take advantage of contactless optical coupling using an expanded beam connector (as in figure 2(a)).Alternatively, the FAs can be equipped with 3D-printed light-turning lenses [34] for an even more simplified geometry.
For the case where the PIC is flip-chip-bonded to the substrate, finding the first light is a major hurdle.It can be solved in automation by using geometrical referencing (e.g.corners), or by processing the back of the PIC to include alignment markers.To improve coupling efficiency, the coupling array can be, again, equipped with an MLA to expand the beam and increase the alignment tolerances, or the lenses can be fabricated on the back of the PIC [33]-which will allow efficient coupling using standard fibres.These post-processing methods however are quite expensive, therefore simple referencing, coupled with beam expansion, would be the most cost-effective way going forward.
The process of aligning the MLA using the GCs as explained in this paper is currently slow and complex in its initial iteration.At the moment we require at least a 2-axis linear scan over a wide range (>250 µm) to find a geometric center of the lens by detecting its edges.This is followed by an evaluation of eccentricity to optimize the distance.Currently, the alignment and attachment of a micro-lens to the PIC may take up to two hours, aligning closely with the duration required by traditional methods.However, the innovative approach presented in this work lays the groundwork for substantial improvements.Future enhancements have the potential to transform the process by incorporating optimization algorithms, such as hill-climbing.As such, we anticipate automation will significantly accelerate the alignment, potentially reducing the time required to just about a minute.Another major improvement stems from the fact that both alignment channels can be potentially aligned at the same time due to the size of the BP sensor.
The automation of the outlined process will be a subject of further study, centered on the development of algorithms aided by statistical analysis.All these efforts will lead to faster packaging time and lower costs of MLA integration into the packages, enabling the pluggable photonic solutions to the edge couplers to be widely adopted.For different PIC coupling schemes, i.e. grating couplers, a very fast method of transfer printing can be used for packaging instead [35], however, this is not applicable to edge couplers.
Finally, while we note that the focus of the presented methods is on attaching an MLA to the PIC, we also demonstrate pigtailing of the PIC using the upright GC method (figure 2).It is therefore entirely possible to perform attachment of the FA to the PIC through Silicon.This opens the door for new approaches to package design and more streamlined packaging processes where the electrical connectivity requires solder-ball flip-chip bonding.

Figure 1 .
Figure 1.(a) Floor plan of the reference PIC with grating-and edge-coupler labels.Defined axes are used throughout this work.(b, c) Microphotographs of the PIC surface depicting the area of the grating coupler and edge coupler respectively.Distance between adjacent couplers is 250 µm.(d) Baseline per-coupler coupling efficiency spectra collected using a single-mode fibre array at the edge and grating couplers.Data for the GC couplers was renormalized using the EC spectra.(e) Alignment tolerance of a fibre to the EC in x-and y-directions.The dashed line shows 1 dB loss cut-off.Curves are asymmetric due to the asymmetry of the mode at the coupler.

Figure 2 .
Figure 2. (a) Photograph of alignment of FA to the PIC using a grating coupler as a 'beacon' .PIC is placed on a stage and held with vacuum.FA is held by mechanical grippers.Other set of grippers holds a fibre array equipped with micro-lenses that collimate the beam that is then injected into the GCs.(b) Per-coupler coupling efficiency for the edge coupler loopbacks L1-L3.Transmission through GCs is much lower due to the use of a collimated beam to inject light.

Figure 3 .
Figure 3. (a) Photograph of the standard method of alignment and attachment of an MLA to the PIC using loopbacks and an MLA-equipped fibre array.(b) Schematic of the standard method of MLA attachment.(c) and (d) Schematic of the proposed method of alignment of the MLA to the PIC using the injection of light to the GCs to highlight the ECs and a beam profiler to measure the beam characteristics.

Figure 4 .
Figure 4. Experimental implementation of attachment of a micro-lens array to the PIC using a beam profiler (compare with figures 3(c) and (d)) (a) Photograph of the alignment of the MLA to the PIC using a beam profiler.(b) Side-view photograph of the alignment of the MLA to the PIC.Arrow shows the path of light, while the inset is a top-view microphotograph (in the GCTSi regime as described in section 4.4) depicting unpolished bottom surface of the PIC.Bright spot is red light passed through the fibre for indication purposes.

Figure 5 .
Figure 5. (a) and (b) Measurement of the width of the beam passing from one of the 'beacon' edge couplers through a micro-lens in (a) x-and (b) y-directions as a function of displacement of the MLA in the x-and y.Colour scale is logarithmic to highlight small differences of width in the lens aperture.Lens outline can be easily discerned and is highlighted by a dashed contour.A defect on lens (possibly dust particle) can likewise be recognized.(c)and (d) Slices of beam width in (c) x-and (d) y-directions through the lens centre.The width rises sharply as the light passes through the edge between adjacent lenses, creating a 'singularity' .The spacing between the peaks is exactly the pitch of the micro-lenses in the array (250 µm)-presented by arrows.The dashed line demarcates the target width as taken from the reference measurement.

Figure 6 .
Figure 6.Optical power measured by the BP as a function of displacement of the FA.Signal from the GC is highlighted.Inset illustrates light coupled through the back facet of the PIC due to proximity of the GC to the PIC edge.

Figure 7 .
Figure 7. (a)-(c) Spectral coupling efficiency per-coupler measured after the packaging of the MLA using the probing FA equipped with a matching lens array for all loopbacks.(d) Comparison of coupling efficiency for all processes for the central (L2) loopback.(e) Comparison of the coupling efficiency at 1550 nm for all loopbacks and methods.

Table 1 .
Comparison of measured beam widths for all MLA attachment processes.