Embedded optical interconnect on printed wiring board

Optical interconnect on printed wiring board Mikko Karppinen*a, Jukka-Tapani Mäkinena, Kari Katajaa, Antti Tanskanena, Teemu Alajokia, Pentti Kariojaa, Marika Immonenb, Jorma Kivilahtib a VTT Electronics, Technical Research Centre of Finland, P.O. Box 1100, FIN-90571 Oulu, Finland b Helsinki University of Technology, Electronics Production Technology, P.O. Box 3000, FIN-02015 HUT, Finland ABSTRACT Integration of high-speed parallel optical interconnects into printed wiring boards (PWB) is studied. The aim is a hybrid optical–electrical board including both electrical wiring and embedded polymer waveguides. Robust optical coupling between the waveguide and the emitter/detector should be achieved by the use of automated pick-and-place assembly. Different coupling schemes were analyzed by combining non-sequential ray tracing with Monte-Carlo tolerance simulation of misalignments. The simulations demonstrate that, with optimized optomechanical structures and with very low loss waveguides, it is possible to achieve acceptable total path loss and yield with the accuracy of automated assembly. A technical demonstrator was designed and realized to allow testing of embedded interconnects based on three different kind of optical coupling schemes: butt-coupling, and couplings based on micro-lens arrays and on microball lenses. They were implemented with PIN and flip-chip-VCSEL arrays as well as 10-Gb/s/channel electronics onto LTCC-based (low-temperature co-fired ceramic) transmitter and receiver modules, which were surface mounted on high-speed PWBs. The polymer waveguides were on separate FR-4 boards to allow testing and characterization of alignment tolerances with different waveguides. With micro-lens array transmitter, the measured tolerances (±10 µm) were dominated by the thickness of the waveguides. Keywords: optical interconnection, optoelectronics packaging, micro-optics, polymer waveguide, printed wiring board, alignment tolerance analysis 1. INTRODUCTION With ever-increasing processor clock frequencies and telecom data rates the electrical interconnects inside equipment is becoming a bottleneck. The performance of electrical interconnects has many physical limitations, for instance, due to dispersion, electromagnetic emission, and susceptibility against electromagnetic radiation. However, optical interconnects have very high bandwidth and, also, are insensitive to EMI/EMC problems. Fiber-based optical interconnects are already used in rack-to-rack and board-to-board applications, but they do not to meet the low-cost and high-integration requirements for board level and shorter distance applications. Free-space optical interconnects have been widely studied but it is difficult to achieve high performance at board or backplane level without the use of bulky optics.1 However, optical interconnects based on low-loss integrated waveguides are a promising solution to overcome the interconnection bottlenecks at board and module level.2,3,4 1.1. Optical-electrical wiring board Most likely, the novel technology with the high-bit-rate optical interconnects has to be developed based on existing PWB technology instead of developing a completely new one. The strip line-like optical waveguide approach fulfils compatibility requirements of PWB technology and enables design techniques analogous to the ones used for wiring electrical signals. Fig. 1 illustrates such a hybrid optical–electrical board (OE-PWB) concept with polymer optical waveguides embedded in/onto board as additional optical layers. 1.2. Optical coupling One of the key challenges of OE-PWB technology is to provide efficient and robust optical coupling between the waveguide and the laser diode/detector, i.e. the transmitter/receiver. Sufficient optical alignment should be achieved by Photonics Packaging and Integration IV, edited by Randy A. Heyler, Ray T. Chen, Proceedings of SPIE Vol. 5358 (SPIE, Bellingham, WA, 2004) · 0277-786X/04/$15 · doi: 10.1117/12.530350 * [email protected]; phone +358 8 551 2111; fax +358 8 551 2320; www.vtt.fi/ele 135 the use of pick-and-place assembly of the components in order to be compatible with automated electronics manufacturing. In addition, tilts of devices (i.e. angular misalignments) have to be rather small with optics – an aspect with not much importance in traditional electronics assembly. A simple approach is butt coupling, in which the optoelectronic devices are facing directly towards the waveguide ends. The optical alignment requirements may be relaxed and the coupling efficiency improved by the use of micro-optical elements, such as, lenses, prisms, mirrors and diffractive optical elements. They also allow the use of packages suitable for standard surface-mount methods. However, there are trade-offs with the optimization of the micro-optical system and the cross-sectional dimensions of the waveguide. For instance, to loosen the alignment tolerances there is a trade-off between coupling of a divergent laser beam to a waveguide and coupling of the waveguide output beam to the relatively small active area of a high-speed detector. The expanded beam method enables to loosen tolerance requirements, e.g. between the package and the board, but the use of this method is limited by the high-density requirement of parallel interconnects, typical channel pitch being 250 µm in VCSEL (vertical-cavity surface-emitting laser) arrays. In practice, reasonable waveguide cross-section is on the order of those of micro-strip lines, i.e., 50…100 µm, which also allows high interconnection density. Fig. 1. A vision of optical-electrical printed wiring board 1.3. Purpose of the work Our objective is to develop technologies to integrate high-speed (~10 Gb/s/channel) optical interconnects into conventional FR-4-based PWBs with dimensions up to 50 cm. The studies include design and modeling methods of high-speed optoelectronics, packaging and optical coupling techniques of the transmitters and receivers, as well as materials and processing of the optical waveguides5. In this paper, some potential solutions are demonstrated with simulation and test results. 2. DESIGN To study different concepts and technologies a test platform was designed. Three realizable optical coupling concepts were chosen for the demonstrator on the basis of optical path loss estimates obtained by ray-tracing simulations and alignment tolerance analyses. Also, the availability of commercial micro-optical components set tight boundaries for the demonstrated coupling methods. The demonstrator includes several parallel data transmission channels, comprising of transmitters, receivers and optical waveguide media. A module substrate utilizing LTCC technology was chosen for the optical front-ends, since it offers possibility to implement precision structures for optical alignment, and, thus, allows the integration of photonic devices and high-speed electrical circuits into a multilayer ceramic wiring board.6,7 2.1. Optical coupling schemes A prerequisite of the optomechanical design is that the feasible component alignment tolerances have to be achievable by standard electronics assembly methods. There are also a couple of basic principles in the design of optical coupling in an OE-PWB: 136 Proc. of SPIE Vol. 5358 • • In order to have optimized in-coupling efficiency the emitter output angular properties need to be matched to the numerical aperture (NA) of the waveguide. Thus, if the emitter beam divergence is larger than the waveguide NA, the source has to be imaged to the input of the waveguide with a magnification larger than unity. In order to have optimized out-coupling efficiency the optics has to match the waveguide output surface area to the detector area. Thus, if the waveguide output surface area is larger than detector area it has to be imaged to the detector with a magnification smaller than unity. Another requirement is that the beam of the surface emitting source has to be bend 90° to couple into the waveguide, and again the waveguide output has to be bend 90° to detector surface. Alternatively, butt-coupling with direct launch to a waveguide can be used, but the VCSEL and photodetector have to be tilted 90°, e.g. using an optical subassembly that is attached to the optically interconnected device (Fig. 2). A hole has to be drilled or etched to the PWB for the subassembly. Thus, a feasible implementation is not straightforward using surface-mount technology (SMT). In addition, good quality waveguide end facets have to be made into the hole to minimize coupling loss. Nevertheless, butt-coupling scheme was designed for the demonstrator for comparison purposes. In butt-coupling, the nominal coupling losses can be small only if the devices can be placed close to the waveguide ends, the VCSEL output divergence is small compared to the NA of the waveguide, and the detector area is not small compared to cross-section of the waveguide. Fig. 2. A butt-coupling scheme realized with an optical sub-assembly attached to the optically interconnected device. Second coupling concept that was designed uses two stacked micro-lens arrays. They form a simple collimatingfocusing-lens pair, suitable for both the in- and out-coupling. The optical design of a transmitter is shown in Fig. 3. The VCSEL beam is collimated using the first lens and the second lens focuses the beam to the waveguide. In addition, a 45°-mirror surface is needed to deflect light into the waveguide. Correspondingly, at the receiver end of the system, the waveguide output beam is first collimated and then focused towards the detector. Basically, aforementioned design principles will cause the optimal distance between the lenses to be similar at the both ends. The double microlens system can be realized by processing microlens arrays on both sides of a substrate or by joining the bottom surfaces of two microlens substrates (as shown in Fig. 3). The lens parameters used in the design were the following: radius of curvature 266 µm, lens array pitch 250 µm, lens aperture diameter 240 µm, surface profile spherical. The lens arrays were made of acrylate photopolymer resin on 0.3mm-thick glass substrates. In the demonstrator, two such array substrates were stacked together with index-matched adhesive; thus the distance between the stacked lens arrays became 0.6 mm. The third designed coupling scheme (illustrated in Fig. 4) is based on two micro-sized ball lenses for each optical channel, which are glued to a mirror surface. Ball lenses are used to image the source to the input of a waveguide, when micro-mirror tilts the beam by 90°, and vice versa at the receiver end. The resulting component is placed under the VCSEL (or PIN) array chip and the whole transmitter (or receiver) module is positioned to the waveguide array. This allows to use separate (good quality) mirror surface. Just like in the butt-coupling scheme, the SMT assembly may become difficult, since the component have to be fit into a hole made onto the board and good waveguide end facets are needed. Also, due to several small components this scheme does not seem attractive for mass production. Although it is probably possible to design a kind of an alignment tool, for instance a frame, for reproducible gluing of the ball lenses. In the designed system, the lenses have diameter of 250 µm and are made of BK7 glass. Proc. of SPIE Vol. 5358 137 LTCC substrate VCSEL chip Cavity formed on LTCC substrate microlens substrates microlenses cladding core reflecting surface cladding Fig. 3. VCSEL-to-waveguide coupling using stacked microlens arrays (and LTCC module substrate) LTCC substrate VCSEL chip Ball lenses glued to micromirror cladding core cladding Fig. 4. VCSEL-to-waveguide coupling scheme based on ball lenses (and LTCC module substrate) 2.2. Optoelectronics For high-density interconnects flip-chip VCSEL and photodetector arrays are favored. A large detector area is preferred to loosen the alignment tolerances in optical coupling, but in practice the area is limited by the trade-off in the receiver bandwidth and sensitivity due to the increased capacitance. Bare die VCSEL arrays were selected for the demonstrator with laser driver chips for data rates up to 10 Gb/s. The VCSELs provide 8 GHz modulation bandwidth at nominal 3 mW output power at 850-nm wavelength. Each array included 12 VCSELs but only four of them were connected to drivers due to layout restrictions on LTCC substrate. For the receiver, InGaAs-based PIN diode arrays were chosen. They include 4 PINs each having active area diameter of 65 µm, 9 GHz bandwidth, 260 fF capacitance, and 0.5 A/W responsitivity at 850 nm. The chips were wedge bonded. Each receiver channel was implemented with a transimpedance amplifier (TIA) and a limiting amplifier. With the used detector they provide around 9-GHz bandwidth and the calculated receiver noise floor is around -27 dBm. These devices allow high loss margin, around 18 dB with VCSEL nominal output and bit-error-ratio of 10-12, but as single channel devices the ICs require rather large board area even for a 4-channel link. 3. SIMULATIONS OF PATH LOSS AND TOLERANCES The feasibility of the chosen optical coupling schemes was studied with ray-trace simulations using 3-D system models. The optical systems were optimized in terms of highest nominal coupling efficiency. After the optimization procedure, the alignment accuracy requirements were analyzed for each system. Both sensitivity and Monte-Carlo tolerance analyses were performed. 138 Proc. of SPIE Vol. 5358 3.1. Methodology of ray tracing and tolerance analysis The simulations of the optical path were made using ASAP, a non-sequential Monte-Carlo ray-tracing software package by Breault Research Organization. Although ray-tracing approach is not as accurate as the wave propagation approach, it is the only one that can be used to analyze complicated optical systems with such physical properties as multimode sources, scattering, and aperture clipping. Ray tracing can also be used to analyze the cross-talk introduced to the parallel systems. Furthermore, it is possible to make propagation simulations through a multimode waveguide and to analyze the system impulse response from the calculated optical path lengths of the rays propagated through the whole system, thus allowing to estimate, for instance, the maximum data transfer rate of the interconnect.8 The root-sumsquare and worst-case analysis that can be made from the sensitivity analyses of individual misalignment effects alone is not a sufficient method for setting the tolerances of the components. This is because sensitivity analysis does not take into account the simultaneous interaction of all variables. This is problematic, for instance, in our case of in-coupling optics (transmitter end), where a single parameter does not necessarily affect the system very much or at all, since the in-coupling performance (loss) is unaffected until the focused beam drops out of the waveguide end and the loss suddenly increases. Therefore, two types of analyses are needed. In sensitivity analysis each tolerance variable (xyzdisplacements and tilts) is simulated separately and most critical tolerances are found. In Monte Carlo analysis the effect of all variables is simultaneously by creating and analyzing a large number of randomly chosen perturbed systems, thus giving statistical information about the system performance. In addition, the correlation between the loss and the tolerance parameters gives information, which parameters are affecting the most to the loss of the whole system. In the Monte Carlo tolerance analysis, the random misalignments were generated using normally distributed random numbers. However, the tails of the normal distribution having probability less than e-1 were cut off in order to avoid very high (probably unrealistic) alignment errors. In the simulations, all surfaces of the optical components were defined without anti-reflection coatings (except the detector). However, in a real product it is preferred to use AR coatings, where possible, to reduce the coupling losses. In all simulations the wavelength was 850 nm. 3.2. Simulation results with Gaussian beam The three designed optical systems were studied by Monte Carlo tolerance analysis. Some results are illustrated in Fig. 5 and the results are summarized in Table 1, which also gives the tolerance values (maximum displacements) used in the modeling as well as the nominal loss, i.e. the loss without any misalignments. A thousand randomly generated tolerance sets (component misalignments) were generated to calculate the loss histograms for the both the transmitter and the receiver ends. Further, the results were used to calculate by convolutions the loss histograms for the system with both the transmitter and receiver ends. However, no waveguide losses were included into the total losses. The simulations were performed using a Gaussian source, which is only an approximation because VCSEL usually has multimode output. VCSEL emitting area diameter was defined 10 µm and the beam divergence (full width at 1/e1) was 14°. The active area diameter of the detector was 65 µm. The waveguide had NA of 0.18 and cross-section of 100 µm x 100 µm. In the butt-coupling system (Fig. 2), we chose for the separation between the waveguide ends and the VCSEL/detector. 200-µm separation was also used in the micro-ball system between the lenses and the VCSEL/detector/waveguide ends. In the micro-lens array system, the separations were 200 µm between the lens systems and the waveguide ends, and 250 µm between the lens system and the VCSEL/detector. The tolerance parameters were chosen based on the planned demonstrator system. The results (Fig. 5 and Table 1) show that the losses are considerable higher at the receiver than at the transmitter in all coupling schemes. This is mainly because of the relatively large waveguide cross-section (100 µm x 100 µm) compared to the 65-µm diameter of the circular active area of the detector. Also, the low losses at the transmitter reflect the fact that VCSELs can be well imaged to the waveguide facets. Consequently, the losses at the receiver can be reduced by diminishing the waveguide cross-section area, up to a certain point without affecting the nominal losses at the transmitter. However, the sensitivity to the misalignments is increased at the transmitter. The highest nominal coupling efficiency is obtained with micro lens arrays. Nevertheless, the shape of the histogram reveals there are several cases with much lower coupling efficiency both in in-coupling and out-coupling, i.e. the variation of total loss is high with these tolerance parameters. The ball lens system is less sensitive to alignment tolerances than the micro-lens system but the nominal and mean losses are higher. Also with the ball-lens system, there are some systems in which the transmitter side has very low coupling efficiency (not seeable in Fig. 5). The butt- Proc. of SPIE Vol. 5358 139 400 Source end Detector end Total 300 200 100 0 -10 Number of Monte Carlo systems Number of Monte Carlo systems Number of Monte Carlo systems coupling system is rather insensitive to misalignments with these parameters. This is mainly thanks to the large waveguide cross-section. 600 500 -8 -6 -4 Loss [dB] -2 0 -2 0 300 Source end Detector end Total 250 200 150 100 50 0 -10 -8 -6 -4 Loss [dB] -2 0 Source end Detector end Total 400 300 200 100 0 -10 -8 -6 -4 Loss [dB] Fig. 5. Coupling loss histograms of the transmitter and receiver ends (separately simulated 1000 cases at both ends) and of the combined system (calculated by convolution of 1 million systems and scaled to 1 thousand): up-left) butt-coupling; up-right) coupling with double-sided micro-lens array and mirror; down-left) coupling with double micro-ball lenses and mirror. Table 1. The statistical data from the Monte-Carlo tolerance simulations (Fig. 5) of the three coupling schemes, for both the transmitter and the receiver ends as well as the combined (convoluted) systems. Also, the maximum misalignments used in the simulation are listed. XYZ-displacements XYZ-tilts XYZ-tilts of mirrors Coupling Nominal loss Max loss Mean loss Min loss Mode loss (peak loc.) 3.3. Butt-coupling ± 20 µm ± 4° In Out Both 4.5 2.98 7.68 10.66 0.39 4.89 5.27 0.06 4.44 4.49 0.09 4.69 4.95 Micro lens array ± 20 µm ± 2° ± 1° In Out Both 1.6 ∞ 21.32 ∞ 1.55 3.25 4.81 0.34 1.23 1.58 0.36 1.67 4.20 Micro ball lenses ± 20 µm ± 2° ± 1° In Out Both 5.0 ∞ 15.11 ∞ 0.74 5.60 6.34 0.40 4.39 4.79 0.46 4.69 5.23 Simulation of ball-lens system with waveguide and VCSEL-like beam The micro-ball-lens-based optical channel was also analyzed using more realistic model of the demonstrator (Fig. 6). This meant a realistic multimode VCSEL beam, and the propagation through the waveguide was included in order to 140 Proc. of SPIE Vol. 5358 obtain more realistic characteristics of the waveguide output beam. The multimode VCSEL source was modeled into ASAP with a non-Gaussian beam having angular distribution characteristics of the VCSEL. The cross-section of the waveguide had two sharp (45°) and two rounded edges. The input and output facets of the waveguide were defined with light scattering properties (Gaussian distribution with 5° scatter angle). The waveguide material was defined to cause 0.05 dB/cm attenuation. Fig. 6. Simulation model of the ball-lens-based optical interconnect demonstrator The simulated attenuations through the nominal optical channel are shown in Fig. 7 with two different sizes of squared profile waveguides. The coupling loss is practically the same at transmitter end, but at the receiver the loss is much less with the waveguide having 50 µm sides than with the one having 100 µm sides. This is because the optical cannot image the whole output beam of the thicker waveguide towards the detector. However, what is not seen is that the system with thicker waveguide is less sensitive to misalignments. 5.1 5.0 Waveguide 100 um Loss [dB] 4.0 Waveguide 50 um After waveguide 3.0 before waveguide 2.0 waveguide (loss 0.05 dB/cm) 1.5 1.0 0.0 0 10 20 30 40 50 60 70 Distance from VCSEL [mm] Fig. 7. Simulated attenuation through nominal optical channel of the micro-ball-lens-based demonstrator The tolerance analysis of the full optical channel was made with the 50-µm waveguide. The modeling parameters are listed in Table 2. The simulation was performed with two different sets of tolerance values (i.e. misalignment distributions). The set number 1 modeled the 'active alignment' case that is realizable in the laboratory where the demonstrator can be assembled using translation stages and optical power measurements. The set number 2 modeled the 'passive alignment' case using proposed electronics assembly methods. Results of the tolerance simulation are presented in Fig. 8. It shows that with the tighter tolerance set most systems have total transmission loss between 5.0 and 5.5 dB, i.e. the assembly tolerances have only a minor effect on the optical power budget. With the looser set most systems (55 %) have transmission loss below 6 dB, but there are also several systems with much higher losses. In other words, the optical channel performs well, but the yield is probably not adequate. Proc. of SPIE Vol. 5358 141 Table 2. Parameters used in the tolerance analysis of the optical channel based on micro-ball lenses Tolerance value set #1 active alignment #2 passive alignment Tolerance parameter Shift x Shift y Shift z Tilt x Tilt y Tilt z Shift x Shift y Shift z Tilt x Tilt y Tilt z Ball lenses & mirror (transmitter) ± 0 µm ± 0 µm ± 0 µm ± 0° ± 0° ± 0° ± 10 µm ± 10 µm ± 10 µm ± 0.5° ± 0.5° ± 1° Tolerance value (maximum misalignment) VCSEL Transmitter Ball lenses & module mirror (receiver) ± 2 µm ± 5 µm ± 0 µm ± 2 µm ± 5 µm ± 0 µm ± 25 µm ± 10 µm ± 0 µm ± 0.5° ± 1° ± 0° ± 1° ± 1° ± 0° ± 0° ± 1° ± 0° ± 5 µm ± 15 µm ± 10 µm ± 5 µm ± 15 µm ± 10 µm ± 5 µm ± 15 µm ± 10 µm ± 1° ± 1° ± 0.5° ± 1° ± 1° ± 0.5° ± 0° ± 0.5° ± 1° Detector ± 1 µm ± 1 µm ± 25 µm ± 0.5° ± 1° ± 0° ± 5 µm ± 5 µm ± 5 µm ± 1° ± 1° ± 0° Receiver module ± 2 µm ± 2 µm ± 5 µm ± 1° ± 0.5° ± 1° ± 15 µm ± 15 µm ± 15 µm ± 1° ± 1° ± 0.5° Number of Monte Carlo systems 300 Tolerance set 1 (active alignment) 250 Tolerance set 2 (passive alignment) 200 150 100 50 0 4 6 8 10 12 14 Coupling loss [dB] Fig. 8. Optical channel loss histograms of the tolerance simulation of the demonstrator based on micro-ball lenses 4. EXPERIMENTAL The designed 10 Gb/s/channel interconnection demonstrator was implemented. It included the three different optical coupling schemes: butt-coupling, micro-lens arrays, and micro-ball lenses. The performance of the optical data channels was characterized. 4.1. Implementation of the demonstrator A modular realization (Fig. 9) was adopted for the demonstration platform to obtain flexibility for characterization. Physically, the system consists of three separate parts: transmitter, receiver, and optical waveguide board. The optical front-ends of the transmitters and the receivers are based on LTCC modules, which include most of the electronics as well as all optoelectronic devices and optical components. LTCC modules are assembled using BGAs on transmitter and receiver PWBs, which include all connectors as well as some electronics, such as limiting amplifiers and control circuits for VCSEL bias and modulation currents. The optical outputs and inputs of LTCC modules are located on areas that extend a bit outside the PWB area, thus allow placing the separate board (e.g. FR-4) with waveguides to serve as an 142 Proc. of SPIE Vol. 5358 optical channel. This kind of mechanical concept enables to measure the alignment tolerances with precise positioning using translation stages. optical transmitters waveguides optical receivers Fig. 9. Schematic illustration of the modular demonstration platform. 4.2. Results of the coupling measurements with micro-lens arrays The coupling optics based on double-sided micro-lens array (as presented in Fig. 3) was realized by gluing a stacked micro-lens array pair to the LTCC after the electronic devices had been assembled. The micro-lens array stack was aligned to the through holes on the LTCC (Fig. 10). The coupling efficiencies from the VCSELs to the waveguides (through micro-lenses) were characterized as a function of misalignment between the transmitter and the waveguide. This was carried out moving the waveguides with a translation stage and measuring the coupled optical power. The 40µm thick polymer waveguides were made on FR-4. Four different core widths between 50 µm and 75 µm were used in the experiment. The waveguide NA was 0.25 and attenuation less than 0.1-dB/cm. The output powers from 10.5-cm waveguides were collected with a large-core fiber. The measured coupling efficiency sensitivities to the alignment of the waveguide perpendicular to the optical path are presented in Fig. 11. The coupling efficiencies are lower and the misalignment sensitivities higher than in the simulations mainly for two reasons: the VCSEL has a multi-modal and bit wider beam in the experiment, and the separation between the VCSELs and the microlenses is longer (550 µm) in the experiment. The alignment tolerances are well explained by the cross-sectional shape and dimensions of the waveguide. Fig. 10. Double-sided microlens array on top of the LTCC substrate that has VCSEL array flip-chipped on the other side. Active areas of the VCSELs are seen through the lenses and LTCC via holes. Proc. of SPIE Vol. 5358 143 0,12 58.3 um waveguide 66.7 um waveguide 50 um waveguide 0,1 Coupling efficiency 75 um waveguide 0,08 0,06 0,04 0,02 0 -50 -30 -10 10 30 Misalignment along x-direction [um] 50 -30 -10 10 30 50 Misalignment along y-direction [um] Fig. 11. Coupling efficiency of microlens transmitter to waveguides as a function of misalignment horizontal (left) and vertical (right) to the waveguide substrate. 5. DISCUSSION The tolerance insensitivity of the designs could be improved by optimizing the parameters based on the loosest tolerances instead of the highest nominal efficiency. This means that the nominal value is most likely reduced form the current value. One issue not taken into account here, is the potential underfill for the emitter and detector chips. Underfills are probably needed for reliability reasons, especially, to avoid dust or moisture from deteriorating the optical path. The underfill may cause problems or as limitations to the optics design if the refractive index is close to the one of the lens surfaces. However, the butt-coupling design is less affected by the use of underfill. 6. CONCLUSIONS Design and ray-trace simulation of the coupling optics for optical-electrical PWBs was carried out. By Monte-Carlo tolerance simulations it was shown that micro-optical elements allow enhancing coupling efficiency and loosening alignment accuracy requirements. The possibility to simulate tolerances of the whole interconnect system was demonstrated by combining the 3D models of the in- and out-coupling optics and the waveguide. The performance of the micro-ball-lens system was feasible, but the tolerance analysis showed that with passive alignment the yield is not necessarily adequate. A demonstration platform with parallel data links based on three different coupling schemes was implemented. The electronics and optics of the transmitters and receivers were integrated on LTCC substrates. The presented coupling efficiency measurements with micro-lens arrays to the waveguides revealed around ±10 µm alignment tolerances, which are challenging for today's high-end assembly techniques. ACKNOWLEDGEMENTS The authors wish to thank Jussi Hiltunen, Kari Kautio, Kimmo Keränen, Jyrki Ollila, and Jarkko Tuominen for their assist with design, implementation and characterizations. The industrial partners of the OHIDA (Optics on future printed circuit board in high-speed data transmission applications) project, namely Aplac Solutions, Asperation, Aspocomp, Elcoteq Network, and Perlos, are acknowledged for their support. This research is financially supported by Tekes, the National Technology Agency of Finland, as a part of its ELMO Technology Program. 144 Proc. of SPIE Vol. 5358 REFERENCES 1. 2. 3. 4. 5. 6. 7. 8. D. Plant, and A. Kirk, "Optical Interconnects at the Chip and Board Level: Challenges and Solutions", Proceedings of IEEE, Vol. 88, No. 6, pp. 806–818 (June 2000). E. Griese, "A High-Performance Hybrid Electrical-Optical Interconnection Technology for High-Speed Electronic Systems", IEEE Transactions on Advanced Packaging, Vol. 24, Number 3, pp. 375–383 (2001). Y. Ishii, S. Koike, Y. Arai, and Y. Ando, "SMT-Compatible Large-Tolerance “OptoBump” Interface for Interchip Optical Interconnections", IEEE Transactions on Advanced Packaging, Vol. 26, No. 2, May 2003. R. T. Chen, L. Lin, C. Choi, Y. J. Liu, B. Bihari, L. Wu, S. Tang, R. Wickman, B. Picor, M. K. Hibbs-Brenner, J. Bristow, and Y. S. Liu, "Fully Embedded Board-Level Guided-Wave Optoelectronic Interconnects", Proceedings of the IEEE, Vol. 88, No. 6, June 2000. M. Immonen, M. Karppinen, J. Kivilahti, " Fabrication and characterization of optical polymer waveguides embedded on printed wiring boards", Proceedings of Polytronic 2003, 3rd International IEEE Conference on Polymers and Adhesives in Microelectronics and Photonics, Montreux, Switzerland, 21–23 Oct. 2003, pp. 91–95 (2003). M. Karppinen, K. Kautio, M. Heikkinen, J. Häkkilä, P. Karioja, T. Jouhti, A. Tervonen, and M. Oksanen, "Passively aligned fiber-optic transmitter integrated into LTCC module", Proceedings of 51th Electronic Components and Technology Conference (ECTC), Florida, USA, May 29–June 1, 2001, pp. 20–25. J.A. Hiltunen, K. Kautio, J.-T. Mäkinen, P. Karioja, and S. Kauppinen, "Passive Multimode Fiber-to-Edge-Emitting Laser Alignment Based on a Multilayer LTCC Substrate", Proceedings of 52th Electronic Components and Technology Conference (ECTC), San Diego, USA, May 28–31, 2002, pp. 815–820. M. Karppinen, K. J. Kataja, J.-T. Mäkinen, S. Juuso, H. Rajaniemi, P. Pääkkönen, J. Turunen, J. Rantala, and P. Karioja, "Wireless infrared data links: Ray-trace simulations of diffuse channels and demonstration of diffractive element for multi-beam transmitters", Optical Engineering, vol. 41(4), pp. 899-910 (2002). Proc. of SPIE Vol. 5358 145