1. Introduction

The continuous demand over the past several decades for increasingly higher performance computers has driven the need for high speed interconnects that are capable of operating at rates greater than 10 Gb/s. This demand for high data rates has pushed the traditional electronic interconnect links to their functional limits [1]. Metal wiring based interconnect technologies suffer from inherent disadvantages such as electromagnetic interference, power and heat dissipation issues at high operating frequencies [2]. Consequently, high speed and high bandwidth interconnects between racks, on backplanes, daughter cards and on the chip level are now needed. Optical interconnects provide a superior bandwidth × length performance in comparison with electronics [2]. Silicon photonics based interconnects are a solution for chip-level dimensions but are not realistic for board-level (1-20 cm) dimensions because of the manufacturing cost and mode size mismatch with optical fibers. An Optical Printed Circuit Board (OPCB) using single-mode optical polymer waveguides can be used to manipulate and transport the optical signal from the processor chips to single mode optical fibers attached at the board edge as explained in a recent paper from IBM [3]. There are a range of low loss optical polymers operating in the ‘telecom’ range of 1300-1600 nm such as Poly(methyl methacrylate) (PMMA) and organic-inorganic hybrid materials [4,5]. Particularly, Organically Modified Ceramic (ORMOCER) [6] is an excellent material for shaping optical structures via ultraviolet (UV) lithographic or stamping [7, 8] processes.

In this paper, we have demonstrated ORMOCER based single-mode waveguides and passive devices for both single and multi-level centimeter sized optical boards using direct UV Nano-Imprint Lithography (NIL) patterning. In-plane single mode waveguides, Directional Couplers, Multi-Mode Interference (MMI) and Y-splitters were designed, fabricated and optically characterized to show the workings of the process at a single optical level. In-plane and multi-level polymer waveguides at telecom wavelengths (1300-1600 nm) have already been reported in literature [9–16]. Previous reports of polymer based 1x2 Y-branch splitters by Oh et. al [17] show an insertion loss (IL) of 3.8 - 4.8 dB for 12 mm long splitter. Here 2 cm long splitters show an insertion loss of 0.9 ± 0.1 dB at 1550 nm. Polymer based directional couplers have been reported with an insertion loss of 0.7 dB at 1310 nm wavelengths by Yoshimura et al [18]. Here the directional couplers and MMI devices show less than 0.45 dB insertion losses at 1550 nm.

The fabrication tolerances associated with the stamping process were analyzed to prove that these fabrication tolerances do not affect the overall working of the devices making NIL suitable for optical interconnect applications. We have proposed a method to quantify a small variation in refractive index contrast between the core and cladding of the polymer waveguides using a tunable laser. Multilayer optical interconnects, being more challenging structures, were fabricated in the next step to show the workings of an optical via and a 1x4 optical 2-D port, Fig. 1. We have experimentally shown dual level waveguides with coupling of light from one level to another, similar to the working of an electrical via in copper based electrical interconnects. Coupling of optical signal from one signal plane to another has already been reported using curved surfaces [19] or partial mirrors using sloped reflectors [20] in polymer waveguides. Single mode optical vias have recently been demonstrated in Si₃N₄ waveguides [21, 22] but this is the first time that multi-level vertical coupling of telecom wavelengths by directional couplers in polymeric waveguides using NIL have been demonstrated to the best of our knowledge. The use of polymers and NIL makes the fabrication process suitable for board level optical interconnects being cost effective and suitable for mass production.

 

Fig. 1 Diagram showing a schematic combination of in-plane and vertical directional couplers.

Download Full Size | PDF

2. Fabrication

Single layer and multilayer single mode waveguide stacks have been fabricated by full-wafer direct UV NIL of polymer materials on 100 mm diameter silicon wafers. We have used commercially available ORMOCER materials: Ormocore with a measured refractive index, n, of 1.534 at 1550 nm and Ormoclad with a measured n = 1.519 at 1550 nm. Having less than 4 nm surface roughness [10] on NIL patterned waveguides and shrinkage of less than 5% on curing [6], ORMOCER is a very suitable polymer for waveguide applications. The imprints were made using an Eitre 6 imprinting tool from Obducat AB (Sweden).

The nano-imprint process broadly consists of two steps; the formation of the stamp and the nano-imprint lithography. The applied fabrication processes were described earlier [9, 15, 16] thus, in the following, only the main process steps are overviewed and the specific parameters for the presented waveguide samples are mentioned. An example stamp is shown in Fig. 3(c). It was formed by lithographic patterning of positive photoresist. The inverted rib process outlined in Fig. 2(a) begins with the application of a 30 µm thick lower index cladding material, Ormoclad, by spin coating to ensure optical isolation between the waveguide mode and the underlying substrate, Fig. 2(a1). An imprint stamp was brought in contact with the lower cladding to form trench waveguides upon UV curing, Fig. 2(a3). An example of a trench is shown in Fig. 2(b) (top-right). These trenches were filled by a material of a higher refractive index, namely a mixture of Ormocore and Ormoclad, which was applied via spin-coating, resulting in the formation of a slab layer of higher index material above the trench, Fig. 2(a4). It is important that the layer is flat, filling the trench completely. The core was cured. Spin coating and curing of a lower index material, Ormoclad, on top of this slab completed the fabrication of a single level optical waveguide.

 

Fig. 2 (a) Step by step fabrication process for multilayer inverted rib waveguides using UV nano-imprinting. (b) Diagrams and Scanning Electron Microscope (SEM) images of inverted rib and rib waveguide processes are shown in the top and bottom part of the image respectively.

Download Full Size | PDF

Multilayer optical interconnects were fabricated by repeating the same imprinting process for the addition of each subsequent layer. Both the lateral alignment and the vertical separation between signal planes are very important factors in the fabrication of multilayer optical interconnects. The upper layer can be laterally aligned with respect to the lower layer(s) using an optical alignment method during the imprinting with alignment patterns made on the stamp. We have achieved lateral alignment tolerances between 1 – 2 μm across a 100 mm diameter wafer which is sufficient for many applications. The low refractive index cladding layer between optical planes should be flat and uniform as it defines the vertical layer-to-layer separation. The thickness variations were found to be less than ± 0.15 µm across the wafer [9]. A 30 µm thick cladding was used as the top-most cladding layer on the multilayer stack after fabricating the desired number of signal planes. In the rib waveguide process, the lower cladding was first spin coated and then the core layer was spin-coated and imprint patterned.

3. In-plane passive optical devices

We have designed the single mode optical waveguides to consist of a 5 µm square trench of higher index material clad by two layers of lower index material with an index contrast of 0.6% between the core and the cladding. The small refractive index contrast results in low optical confinement with the mode extending into the cladding material making sidewall roughness potentially an important factor in the waveguide loss. This is not a factor here as the roughness is less than 4 nm [10] with NIL patterned ORMOCER waveguides. The majority of the waveguide loss is associated with material loss. The loss of the waveguide cladding material Ormoclad is slightly smaller (<0.1 dB/cm difference) than the loss of core material, i.e. the mixture of the Ormocore and Ormoclad. The loss of similar waveguides was measured from the attenuation of the optical signal in a 27 cm long spiral. The overall loss was measured as 0.84-0.89 dB/cm at 1530 nm and 0.75-0.80 dB/cm at 1550 nm, composed of material loss of 0.7 dB/cm at 1530 nm and 0.6 dB/cm at 1550 nm (from material datasheet). This results in an excess waveguide propagation loss of < 0.2 dB/cm [9, 12]. The low optical confinement allows high efficiency coupling to single mode fibres while placing a lower limit on the bend radius to prevent light escaping [12].

In-plane passive optical devices namely directional couplers, MMI and Y-branch power splitters were designed. The devices were simulated using a commercially available mode solver, FIMMWAVE [23]. Directional couplers operate on the principle of resonant power transfer between two adjacent waveguides whose modes overlap. The amount of light split between waveguides can be controlled by varying the length and spacing of the waveguides in the interacting region. The gap between waveguides has been chosen to be 5 µm as gaps less than 2 µm are difficult to fabricate given the waveguide feature size of 5 µm. Directional couplers were fabricated with interaction lengths varying from 500 µm to 5000 µm in 500 µm steps. Simulations predict that the coupling length, where all the light input from one waveguide is coupled to the adjacent waveguide, is 2000 µm. The devices were measured on a waveguide measurement system using a fiber-coupled tuneable laser source set at 1550 nm. Single mode lensed fibers were coupled to the input and output waveguides of the directional couplers and the power maximized to measure the ‘through’ and ‘coupled’ output powers, Fig. 3(c). The total output was normalized to the input power for directional couplers of different lengths. The fabricated directional couplers display a coupling length of slightly less than 2000 µm with less than 0.45 dB insertion loss and 0.02 ± 0.01 dB power imbalance between outputs.

 

Fig. 3 (a) Comparison of the splitting of light in the ‘coupled’ and ‘through’ ports of polymeric directional couplers with different coupling lengths at 1550 nm. (b) Comparison of the simulated and measured efficiencies of the MMIs of different lengths measured at 1550 nm. (c) Diagrams showing the definition of through and coupled ports and a comparison between the nominal (ideal) and actual fabricated waveguides. The slope of the sidewalls and a residual layer of core material can be seen in the microscope image. A SEM image of a comparable imprinting stamp is shown at the bottom. (d) The measured and simulated modes for MMIs diced at different lengths of the multimode regions.

Download Full Size | PDF

Multi-mode interference devices employ self-imaging in a multi-mode section of the waveguide. The input field is reproduced in single or multiple images at intervals along the direction of propagation of the waveguide in the multimode region of the device. We have chosen an MMI design employing single mode input and output waveguides and a multi-mode region of 40 µm width. This width keeps the length of a 1 x 2 splitter to less than 1 mm with the simulations predicting a peak efficiency of approximately 90% (IL = 0.45 dB) at a length of 970 µm. The efficiency is calculated as the sum of both outputs divided by the total output power through an adjacent straight waveguide which is used as a control. MMIs were fabricated with lengths varied from 880 µm to 980 µm in 20 µm steps. Some additional MMIs of lengths 1090 µm, 1100 µm and 1120 µm were also added to the chips to measure the variation in coupling efficiency. The measured efficiencies for MMIs of different lengths are shown in Fig. 3(b) and agree well with the simulations, with the efficiency increasing with the length of the multimode region to a maximum at a length of approximately 970 µm. Multimode regions of the fabricated MMI devices were diced at different lengths along the direction of the propagation to observe the self-imaging of the input mode. Devices were diced at the lengths of 450 µm, 650 µm and 950 µm marked with red dotted lines in Fig. 3(d). An infra-red camera was used to image the output intensity at these lengths. Good agreement between the simulated and measured intensity profiles at the different lengths of the MMI region is shown in Fig. 3(d).

Y-branch splitters were designed and fabricated to divide the input power into half. Cascaded splitters were also designed to get different fractions of the input power shown in Fig. 4(a). Correct splitting ratios were achieved for different devices with 0.9 ± 0.1 dB insertion loss and 0.02 ± 0.01 dB power imbalance between the outputs. Insertion loss can be improved by further optimizing the design and fabrication process.

 

Fig. 4 (a) Sketches of Y-splitters designed for 50-50% and 50%, 25%, 12.5%, 6.25% and 6.25% splitting. Microscope images of (b) the fabricated 50-50% Y-splitter, (c) the start of the cascaded Y-splitter and (d) the continuation of the cascaded Y-splitter.

Download Full Size | PDF

3.1 Fabrication tolerances

The use of NIL results in two parasitic effects: the presence of a residual layer (or slab) of core material and sloped side walls of the waveguides, Fig. 3(c). The residual layer is inherent in the NIL process, whereas the sloped waveguide side walls originate from the stamp fabrication process. At least a small tilt of the side walls is also preferred in the imprinting, because it eases the stamp release (compared to vertical walls), thus improving the fabrication yield. Processing of polymers may also result in a variation in the actual refractive index contrast between the core and cladding materials if the mixing ratio of the two polymers used in “tuning” the core index is not controlled precisely. Simulations to investigate the effect of these variations show that directional couplers are much more sensitive as compared to MMI devices. The sloped sidewalls reduce the effective gap between the waveguides of the directional coupler in the coupling region, thus reducing the coupling length, Fig. 5(d). The side wall angles were consistently measured to be less than 20° in the fabricated samples. A thinner residual layer increases the coupling length of the directional coupler by increasing the effective index contrast between adjacent waveguides and thereby increasing the modal confinement and reducing the evanescent field overlapping the adjacent waveguide in the coupling region, Fig. 5(f). Note that the residual layer thickness in the fabricated structures was consistently measured to be less than 0.5 µm. Geometry-related variations in device performance are well controlled. Tight control of the polymer mixing ratio is essential for control of the index contrast. A larger index contrast increases the coupling length due to increased modal confinement inside the core, Fig. 5(e). Good control of polymer mixing ratio is evident as coupling lengths of the fabricated directional couplers were consistently measured to be 1900 ± 100 µm in line with the simulations. MMI devices, whose modal expansion occurs in the core material, do not depend on the overlap of the evanescent tail of the mode into an adjacent waveguide and therefore are less affected by the changes in refractive index contrast compared to directional couplers, Fig. 5(b). The increase in effective width due to sloped sidewalls is small and leads to a small increase in efficiency, Fig. 5(a). The MMIs are therefore more tolerant to all these process related effects.

 

Fig. 5 Simulations of the maximum efficiency as a function of MMI length for: (a) an increase in the side wall angle, (b) a change in refractive index contrast, (c) a change in residual layer thickness. Simulations of the maximum efficiency as a function of directional coupler length for: (d) wall angle showing that the coupling length increases from 1100 µm to 1800 µm compared to 100 µm change in the optimum for the MMI, (e) a change in refractive index contrast. (f) a change in residual layer thickness.

Download Full Size | PDF

Simulations show that a small change in refractive index contrast, Δn, between the core and cladding results in a significant change in coupling length of the directional coupler. Thus a method to experimentally quantify a small variation in Δn of a fabricated device is investigated. This method can be used for small (5 µm) waveguides in comparison with, for example, ellipsometry which has difficulty in distinguishing small index changes in multilayer structures with small feature sizes. Waveguides of fixed dimension but with different Δn were simulated to calculate the output mode for both center and off-center excitations of the waveguides. Simulations show that the waveguides with Δn = 0.6% will remain single mode irrespective of the position of input excitation for all wavelengths across our measurement spectral range as represented by three wavelengths i.e. 1440 nm, 1550 nm and 1640 nm. The waveguides will become multimode for the shorter wavelengths i.e. 1440 nm and 1550 nm for off-center excitation if the index contrast is increased to Δn = 0.7%. A further increase in the index contrast will excite multiple modes in the waveguides for all three wavelengths with off-center excitation. The mixing ratio of the polymers was changed to get an index contrast of 0.8%, which is higher than the single mode value of 0.6%. Devices with the different index contrasts were measured on a waveguide measurement system using a fiber-coupled tunable laser source for all three wavelengths. The output waveguide was scanned in horizontal and vertical directions to measure the output mode shape. It was experimentally observed that a waveguide having a higher index contrast was multimode for all three wavelengths and we measured a coupling length of 4000 µm for directional couplers on that chip which is consistent with the 0.8% index contrast, Fig. 6. Note that variations in sidewall angle and residual layer thickness were measured with a microscope and would not account for the observed difference. Matching these results with simulation results confirmed that the refractive index contrast of the tested sample was 0.8% as calculated form the applied 80% - 20% mixing ratio of Ormocore and Ormoclad for the sample.

 

Fig. 6 Comparison of the measured modes in a directional coupler at different wavelengths with the simulations confirms that refractive index contrast of the sample is 0.8%. The measured coupling lengths are also in agreement with the simulations.

Download Full Size | PDF

4. Multi-level optical interconnects

Multilevel optical structures were designed to achieve coupling from one signal plane to another analogous to electrical interconnections. The vertical coupler works on the same principle as the in-plane coupler but light resonantly transfers between the waveguides closely stacked in the vertical direction. The vertical coupler thus can act as an ‘optical via’ to transfer light from one signal plane to another if designed to fully (100%) transfer the power. The spacing between the waveguides (top of lower waveguide to bottom of upper waveguide) in the vertical coupling region is 5 µm as for the in-plane couplers so as to obtain similar coupling lengths. Vertical directional couplers were fabricated on 100 mm diameter silicon wafers using multiple NIL stages as indicated in Fig. 2(a). Vertical couplers with coupling lengths varying from 500 µm to 1250 µm in 250 µm steps were fabricated. The insertion loss of the fabricated devices was measured to be less than 0.45 dB. The normalized powers from the ‘through’ and ‘coupled’ outputs of the vertical directional coupler are plotted in Fig. 7(a). It can be seen that almost 50% of the input power was up-coupled for a coupling length of 900 µm. It can be inferred from the measured results that complete transfer of power will take place at a coupling length of 1800 µm which is consistent with the coupling length of in-plane couplers. So, a vertical directional coupler having a coupling length of 1800 µm can act as an ‘optical via’ to transfer power from one signal plane to another. Note that a lateral misalignment between upper and lower signal plane(s) will increase the separation between the waveguides resulting in larger coupling lengths in vertical directional couplers.

 

Fig. 7 (a) Measured normalized powers from ‘through’ and ‘coupled’ ports of the vertical directional coupler. A crossing at 900 µm was also measured for in-plane directional couplers. (b) Schematic of a vertical coupler with the measured image of the light split laterally and vertically. (c) Microscope images of a multilevel device to show the alignment of the vertically connected waveguides. The top image was taken after imprinting the 2nd level waveguides and before spinning the core material. (d) Schematic of 1x4 2-D port device with camera image of light split into the 4 ports. The interference fringes were measured by defocusing the camera.

Download Full Size | PDF

A combination of in-plane and vertical couplers as shown in Fig. 7(d) was designed and fabricated to realize a two dimensional 1x4 port. This 2-D port was designed to equally split the input power into four waveguides spatially placed on two vertically stacked optical planes. An in-plane Y-branch coupler was used to equally split the input light into two waveguides. Two identical vertical directional couplers of coupling length 900 µm were placed on the output waveguides of the in-plane coupler to up-couple half of the available power. The fabricated devices were characterized using the setup explained earlier to measure one output at a time. Measurement results show that incident power can be split into 4 spatially placed waveguides with an insertion loss of 1.2 dB. The imbalance between the combined power in the two ports on the top layer and the corresponding power in the bottom layer was 0.2 ± 0.02 dB. The output ports of the fabricated devices were imaged using a camera by exciting an input port. The image showing the splitting of light into the 4 ports is shown in Fig. 7(d). As the light is from a single laser source and split into four ports so the output signals should be coherent and show interference. The resulting interference fringes are shown in Fig. 7(d) which were obtained by defocussing the camera to allow the overlap of the light from the different ports confirming the coherence of the output light.

5. Summary and conclusions

The use of nano-imprint lithography for board-level optical interconnects has been demonstrated using single mode polymeric waveguides. The characteristics of Y-branch splitters, directional couplers and MMIs were presented. An experimental method to identify variations in the index contrast of the waveguides is shown. Functioning optical vias and a 1x4 two dimensional optical port have been demonstrated which will pave the way for scalable, polymer based, single mode multi-level optical interconnects. The efficiency of these multi-level devices can be increased by further optimizing the designs and the fabrication processes. These 2D devices can also be used as excitation and receiving channels in spatial division multiplexing applications.