An intraboard chip-to-chip optical interconnection isone of the attractive candidates for solving the so-called pin-bottleneck problem for construction of anultrahigh-performance information-processing unit.1,2
A two-dimensional (2-D) massive parallel transmis-sion of optical signals will be used for greater thanterabits per second signal transmission since thetransmission bandwidth of a single channel is limitedby direct modulation of a laser diode to �10 Gbits�s.A topic is how to connect the optical signals from a2-D array of the vertical-cavity surface-emitting la-sers (VCSELs) connected to a transmitter chip to a2-D array of the photodiodes (PDs) connected to areceiver chip. Free-space interconnecting configura-tions with micro-optic components have been thosemostly demonstrated so far.3–9 As another approach,an integrated-optic configuration by use of a thin-filmwaveguide has been proposed and investigated.10–12
The integrated-optic configuration has the advan-tages of alignment stability, reduction of weight and
size, fabrication by planar processes, and compatibleconfiguration with a current circuit-printed board.
An issue is the input–output coupling of free-spacewaves, which are radiated from VCSELs or focused tobe detected by PDs, with the guided waves in a thin-film waveguide. Another issue is a 2-D optical trans-mission in the waveguide. To settle both issues, wehave investigated the utilization of a wavelength-division-multiplexing (WDM) technique with thefree-space-wave add–drop multiplexers. A conceptimage of the integrated-optic version is depicted inFig. 1. Each electronic chip with an optoelectronicinterposer integrating a 2-D array of VCSELs and a2-D array of PDs is surface mounted on a thin-filmoptical waveguide. VCSELs of different wavelengthsaround 850 nm are integrated in line along theguided-wave propagation direction to transmit WDMoptical signals. The multiplexer consists of a guided-mode-selective (GMS) focusing grating coupler (FGC)and a different guided-mode (DGM) coupling distrib-uted Bragg reflector (DBR). The device configurationand the operation behavior are described in Section 2.The device has been designed and fabricated and hasbeen demonstrated as a free-space wave drop demul-tiplexing function.12
One of the next key issues is the development of ahigh throughput fabrication process applicable forfuture mass production of the devices. We have fab-ricated both GMS FGCs and DGM DBRs by theelectron-beam (EB) direct writing technique. Thistechnique is not suitable for high throughput. How-ever, there is not a serious mass production problemwith respect to GMS FGCs, because they are fabri-
S. Ura ([email protected]), M. Hamada, J. Ohmori, and K. Nishioare with the Department of Electronics and Information Science,Kyoto Institute of Technology Matsugasaki, Sakyoku, Kyoto 606-8585, Japan; K. Kintaka is with the Photonics Research Institute,National Institute of Advanced Industrial Science and Technology,1-8-31 Midorigaoka, Ikeda 563-8677, Japan.
Received 1 February 2005; revised 26 April 2005; accepted 26April 2005.
cated by a photolithography technique with a lithog-raphy mask made by the EB direct writing technique.On the other hand, the photolithography technique isnot as effective for the fabrication of DGM DBRs,since the grating periods are precisely tuned to com-pensate for the fabrication errors in the wavelengthsand the guided-mode indices. To find a better solu-tion, an interference exposure system was devel-oped,13 and the applicability of a laser beaminterference exposure photolithography was tested.Our waveguide device has a high reflection substrate;such a high-reflective substrate is not used for high-resolution photolithography because of the patterndegradation that is due to reflection. We carefullyoptimized the exposure condition and could integrateDGM DBRs of an approximately 290 nm grating pe-riod with 0.03 nm period precision to form a two-channel drop demultiplexer.14 Here we describe theexposure system and report some preliminary exper-imental results of the drop demultiplexer.
2. Device Configuration
A schematic view of a part of the proposed integrated-optic version, namely, a single waveguide channel fortwo-wavelength add–drop multiplexing, is illus-trated in Fig. 2. A cross-sectional view of the wave-guide is depicted with a refractive-index profile inFig. 3. Two guided modes, namely, TE0 and TE1, wereutilized. An optical signal wave diverging from asingle-mode polarization-controlled VCSEL is cou-pled and collimated by a GMS FGC of a 100 �m �100 �m aperture into TE1 mode in the waveguide, iscontradirectionally coupled by a DGM DBR of a300 �m coupling length to the TE0 mode and propa-gates in the waveguide. The TE0 mode of only a spec-ified wavelength is coupled by a corresponding DGMDBR to the TE1 mode, and finally coupled out by a
GMS FGC to a free-space wave focused to the corre-sponding PD. An electric field of the TE0 mode issmall at the GMS FGC layer, indicating that the TE0mode passes through the GMS FGCs without seriousdiffraction. As is well known, the DBR shows wave-length dispersion. As a result, WDM signals sup-ported by the TE0 mode pass through all the GMSFGCs and DGM DBRs except for the correspondingDGM DBR. Thus the TE0 mode serves as a signaltransmission mode. On the other hand, an electricfield of the TE1 mode is sufficiently large at the GMSFGC layer to be coupled by the diffraction of GMSFGC with the free-space wave. Thus the TE1 modeserves as an interface mode by bridging the gap be-tween the waveguide and the free space. In this way,an optical signal of wavelength �1 is transmitted fromthe first VCSEL to the first PD whereas another sig-nal of wavelength �2 is transmitted from the secondVCSEL to the second PD. The number of WDM chan-nels can be increased by the addition of VCSELs ofdifferent wavelengths and integration of GMS FGCsand DBRs for the corresponding wavelengths. Thewidth of a waveguide channel is determined by thewidth of a GMS FGC to be 100 �m. A divergenceangle along the propagation that is due to diffractionin the waveguide is 5 � 10�3 rad. The propagationloss is not a big issue since the propagation length isseveral millimeters. In other words, the GMS FGCwidth should be designed from the interconnectionlength.
Use of VCSELs of approximately 850 nm wave-length is considered. The waveguide consists of amain-guiding core layer of Ge-doped SiO2 (refractiveindex of n � 1.53) and a subguiding core layer of SiO2(n � 1.46) on a glass substrate coated with a Auhigh-reflection layer (n � 0.2 � j 5.5). GMS FGC isformed by the corrugation of a Si–N layer (n � 2.01)inserted into the middle of the subguiding core layer.A DGM DBR is formed by the corrugation of an UVphotoresist (n � 1.56) coated on top of the main-guiding core layer. The calculated mode profiles forboth modes in the GMS FGC area are also depicted inFig. 3. The TE0 mode is formed by a total internalreflection at the boundary between the main-guidingand the subguiding cores and is confined mainly tothe main-guiding core layer. On the other hand, theTE1 mode is supported by a high reflection by the Au
Fig. 1. Concept image of intraboard chip-to-chip optical intercon-nection with a thin-film waveguide for 2-D parallel transmissionfrom a VCSEL array to a PD array.
Fig. 2. Schematic view of the proposed free-space-wave add–dropmultiplexing configuration with a thin-film waveguide.
Fig. 3. Cross-sectional view of a waveguide with a refractive-index profile, and the calculated electric-field profiles of the TE0
reflection layer and is spread over both the main-guiding core and the subguiding core layers. The cou-pling coefficient of the GMS FGC is proportional tothe electric field strength of the coupled mode. There-fore, the structure was designed so that the electricfield of the TE0 mode is small but that of the TE1mode is large at the GMS FGC layer. The Au reflec-tion layer plays an important role to yield such acondition. Another role of the Au reflection layer isenhancement of the coupling efficiency of the GMSFGC. In an output-coupling scheme of a grating cou-pler without a special design, a guided wave is cou-pled and radiated to the substrate as well as to theair. In our design, on the other hand, the substrateradiation wave is reflected by the Au reflection layerand added in phase to the air-radiation wave. A pre-dicted power distribution ratio to air radiation versusthe total diffraction is almost 100%. There is somepropagation loss for the TE1 mode that is due to ab-sorption by the Au layer, whereas propagation loss ofthe TE0 mode can be neglected. However, the TE1mode appears only twice for one signal channel asmentioned above, and its propagation length is asshort as the submillimeter range then the resultantpropagation loss is small.
3. Development of an Exposure System
A grating period of a DGM DBR is determined fromthe TE0 and TE1 guided-mode indices and a couplingwavelength. The resultant grating period is approx-imately 290 nm for a glass waveguide, and the perioddifference is 1 nm for an example spacing of 3 nm inthe coupling wavelengths. In addition, DGM DBRsare integrated in the final process to compensate forthe errors in the wavelengths and the guided-modeindices by precise tuning of the grating period. ALloyd mirror configuration is well known for use inthe construction of a compact interference exposuresystem15 and is good at precise control of the inter-ference fringe period.
A compact interference exposure system with aLloyd mirror configuration was combined with acontact-type mask aligner. A schematic diagram ofthe developed system is illustrated in Fig. 4. A244 nm wavelength laser beam from an Ar-ion laserfollowed by a second-harmonic generation unit is spa-tially filtered by a 5 �m diameter pinhole, expandedin 1 m propagation, and collimated by a lens to be an
almost uniform plane wave of 50 mm diameter. Thewave is reflected downward by mirror M1 mountedon a rotation stage on which angle � can be controlledto a precision of 1.25 � 10�5 rad/div. Half of the re-flected wave is directly illuminated with an incidenceangle � to a metal mask fixed beneath a mask holder,and another half is reflected by mirror M2 and isilluminated with an incidence angle �� to the metalmask. M2 is fixed to the mask holder to be perpen-dicular to the metal mask. The optical interferencefringes are generated by the two waves, and period �is given by
� ��
2 sin ��
�
2 cos 2�, (1)
where � is the exposure wavelength. The dependenceof � on � is almost proportional to �, indicating thata shorter � is better for precise control of � by �. Wechose a photoresist developed for the KrF laser expo-sure. A highly coherent beam is necessary for wide-area interference exposure, which is why we used the244 nm laser system. The variation of � around0.29 �m is 0.015 nm�div against a scale of �. Themetal mask has two slit windows for the interferenceexposure of a pair of DBRs of the same grating period.The distance between the slits is 20 mm, as deter-mined from an optical interconnection length. Thediameter of the incident beam should be larger thandouble the distance for the folded-beam interference,which is why the laser beam was expanded to 50 mm.The mask is 50 �m thick.The sidewalls of the slitwindows are tapered to reduce the shadow of thewalls on the waveguide since the exposure beamshave tilted angles. The waveguide coated with a pho-toresist is set on an x–y–z–� stage, aligned and incontact with the mask. DBRs of the different gratingperiods are integrated by displacing the waveguideand adjusting � one after another.
The system performance was checked first. A0.14 �m thick photoresist (KRFM151Y provided byJSR Corporation) was coated on a glass waveguide ona glass substrate. Incident angle � was 0.44 rad togenerate � � 0.29 �m. The exposure dose at the max-imum point of the interference fringes was0.26 J�cm2. After development, the obtained gratingswere observed by a scanning electron microscope; theobtained photographs are shown in Fig. 5. The cross-sectional structure was sinusoidal and the groovedepth was 0.10 �m, which is deep enough to fabri-cate DGM DBRs. The different period gratings wereintegrated and their periods were measured. Two ex-amples of the period differences are summarized inTables 1 and 2. Five gratings, #1, #2, #3, #4, and #5listed in Table 1, were integrated with the variationof � by 2�, �, 0, ��, and �2�, respectively,with � � 1.5 � 10�4 rad; the five gratings shown inTable 2 were integrated with � � 2.5 � 10�4 rad. Agrating period for grating #3 in Table 1 was measuredto be 290.3 nm, whereas that for Table 2 was285.7 nm. A predicted period difference for � in Ta-
Fig. 4. Compact interference exposure system developed for in-tegrating the DGM DBRs. The Lloyd mirror optics and contact-type mask aligner are combined.
ble 1 is 0.09 nm and for � in Table 2 it is 0.15 nm.The deviations of the measured values from the pre-dicted values were within 0.03 nm for both cases.This means that the developed interference exposuresystem provides sufficient precision for the integra-tion of DBRs with a 1 nm grating-period difference inthe construction of the proposed add–drop multi-plexer.
4. Fabrication and Characterization ofDrop Demultiplexer
Two pairs of GMS FGC and DGM DBR were inte-grated to form a two-channel free-space-wave dropdemutliplexer. After thermal evaporation of a Au filmon a glass substrate, a 0.6 �m thick SiO2 first sub-guiding core layer and a 40 nm thick Si-N layer weredeposited by plasma-enhanced chemical vapor depo-sition (PECVD) and reactive sputtering, respectively.An EB resist was coated, and the GMS FGC patternwas exposed by EB direct writing and developed. Thepattern was transferred by reactive ion etching ontothe Si–N layer to be grating grooves. A SiO2 secondsubguiding 0.66 �m thick core layer and a Ge-dopedSiO2 main-guiding 0.7 �m core layer were sequen-tially deposited by PECVD. The 0.14 �m thickKRFM151Y photoresist was coated and DGM DBRs
were exposed by the developed interference exposuresystem.
The indices of the TE0 and TE1 modes in the DGMDBR area were theoretically predicted to be N0� 1.494 and N1 � 1.469, respectively. The propaga-tion loss for the TE0 mode that is due to the radiationand the absorption by Au was theoretically predictedto be 1.7 dB�cm in a guiding area without a GMSFGC or a DGM DBR but was experimentally esti-mated to be approximately 5 dB�cm. We believe thatthe main cause is the in-plane scattering that wouldbe suppressed by improvement of the process condi-tions. The propagation losses of the TE1 mode werecalculated to be 9 dB�cm in the DGM DBR area and15 dB�cm in the GMS FGC area. In other words, theloss is estimated to be less than 0.4 dB in a single pairof GMS FGC and DGM DBR since GMS FGC andDGM DBR are integrated side by side and the prop-agation length of the TE1 mode is 0.4 mm at its long-est point. The radiation decay factor of the GMS FGCwas theoretically calculated to be 17 mm�1, and thecoupling efficiency was predicted to be 97% for a0.1 mm coupling length. However, the decay factorwas experimentally measured to be approximately3 mm�1. The cause of the difference between theoret-ical and experimental values is now under investiga-tion. A coupling coefficient of the DGM DBR and theresultant coupling efficiency for the 0.3 mm couplinglength were estimated to be 9.8 mm�1 and higherthan 99%, respectively. The grating periods of theDGM DBRs were chosen to be �1 � 284.6 nm and�2 � 286.1 nm. The drop wavelengths �1 and �2 arecalculated from the relation �i � �N0 N1��i, wherei � 1 or 2. They were estimated to be �1 �843.3 nm and �2 � 847.7 nm for the theoretical N0 N1, whereas they were predicted to be �1 �844.5 nm and �2 � 848.9 nm from the measured N0 N1.
The drop demultiplexing function of the fabricateddevice was tested. A wave from a wavelength tunablelaser diode was coupled by a coupler prism to be theTE0 mode in the waveguide as illustrated in Fig. 6.The output powers of the focusing waves from GMSFGCs were measured when the wavelength was var-ied. The measured dependence of the output efficien-cies of the waves radiated from the first pair of GMSFGC and DGM DBR and the second pair are repre-sented by open and closed circles, respectively, in Fig.7. The efficiency shows its peak at around 844.5 nm
Fig. 5. Scanning electron microscope photographs of the fabri-cated gratings: (a) top view and (b) cross-sectional view.
Fig. 6. Schematic of the experimental setup. A wavelength tun-able laser diode was used instead of VCSELs to characterize wave-length drop demultiplexing function.
Table 1. Measured Period Differences �� for Gratings Integrated byVarying � with Pitch �� � 1.5 � 10�4 Rad
for the first grating pair, and that for the second pairhas its peak around 849 nm. They were exactly thesame as the predicted values within the measure-ment accuracy. The obtained full width at half-maximum for both peaks was 2 nm, which coincideswith the predicted value of 2 nm. However, the effi-ciency of the output beam from the second pair isconsiderably lower than that from the first pair. Webelieve that this difference is enhanced by the prop-agation loss of the TE0 mode including a diffractionloss by the GMS FGC and a mode conversion loss atthe boundaries between the areas with and withoutgratings. Research is under way to determine how tosuppress propagation loss.
5. Conclusions
An interference exposure system combined with amask aligner was newly developed for integratingDGM DBRs of 0.29 �m grating periods withnanometer-order period differences. It was demon-strated that the gratings can be integrated on awaveguide substrate with period errors of less than0.03 nm. DGM DBRs of 300 �m coupling length wereintegrated with GMS FGCs on a thin-film waveguideto form a free-space-wave drop demultiplexer of twochannels. The demultiplexing principle was con-firmed with the result that the measured drop wave-lengths were exactly the same as the predicted ones.The reduction of the propagation loss is required andnow under study to construct a WDM chip-to-chipoptical interconnection board.
This research was undertaken as part of the Coop-erative Studies with Researchers of Industries “De-velopment of Innovative Designing�ManufacturingProcess and Creation of New Opto-Electronic Systemwith Ultra High Performance and Integration” in Ja-pan. A part of this study was financially supported bythe International Communications Foundation in Ja-
pan and another part was funded by Grant-in-Aid forScientific Research (A) 15206008 of the Japan Societyfor the Promotion of Science.
References1. N. Savage, “Linking with light,” IEEE Spectrum 39(8), 32–36
(2002).2. D. A. B. Miller, “Rationale and challenges for optical intercon-
nects to electronic chiops,” Proc. IEEE 88, 728–749 (2000).3. A. G. Kirk, D. V. Plant, T. H. Szymanski, Z. G. Vranesic,
F. A. P. Tooley, D. R. Rolston, M. H. Ayliffe, F. K. Lacroix, B.Robertson, E. Bernier, and D. F.-Brosseau, “Design and imple-mentation of a modulator-based free-space optical backplanefor multiprocessor applications,” Appl. Opt. 42, 2465–2481(2003).
4. C. Debaes, M. Vervaeke, V. Baukens, H. Ottevaere, P. Vynck,P. Tuteleers, B. Volckaerts, W. Meeus, M. Brunfaut, J. VanCampenhout, A. Hermanne, and H. Thienpont, “Low-cost mi-crooptical modules for MCM level optical interconnections,”IEEE J. Sel. Top. Quantum Electron. 9, 518–530 (2003).
5. S. Voigt, S. Kufner, A. Kufner, and I. Frese, “A refractivefree-space microoptical 4 x 4 interconnect on chip level withoptical fan-out fabricated by the LIGA technique,” IEEE Pho-ton. Technol. Lett. 14, 1484–1486 (2002).
6. M. Chateauneuf, A. G. Kirk, D. V. Plant, T. Yamamoto, andJ. D. Ahearn, “512-channel vertical-cavity surface-emitting la-ser based free-space optical link,” Appl. Opt. 41, 5552–5561(2002).
7. H. Sasaki, K. Kotani, H. Wada, T. Takamori, and T. Ushikubo,“Scalability analysis of diffractive optical element-based free-space photonic circuits for interoptoelectronic chip intercon-nections,” Appl. Opt. 40, 1843–1855 (2001).
8. M. Gruber, “Multichip module with planar-integrated free-space optical vector-matrix-type interconnects,” Appl. Opt. 43,463–470 (2004).
9. G. Li, D. Huang, E. Yuceturk, P. J. Marchand, S. C. Esener,V. H. Ozguz, and Y. Liu, “Three-dimensional optoelectronicstacked processor by use of free-space optical interconnectionand three-dimensional VLSI chip stacks,” Appl. Opt. 41, 348–360 (2002).
10. S. Ura, “Selective guided mode coupling via bridging mode byintegrated gratings for intraboard optical interconnects,” inOptoelectronic Interconnects, Integrated Circuits, and Packag-ing, L. A. Eldada, R. A. Heyler and J. R. Rowlette, Sr., eds.,Proc. SPIE 4562, 86–96(2002).
11. K. Kintaka, J. Nishii, Y. Imaoka, J. Ohmori, S. Ura, R. Satoh,and H. Nishihara, “A Guided-Mode-Selective Focusing Grat-ing Coupler,” IEEE Photon. Technol. Lett. 16, 512–514 (2004).
12. K. Kintaka, J. Nishii, J. Ohmori, Y. Imaoka, M. Nishihara, S.Ura, R. Satoh, and H. Nishihara, “Integrated waveguide grat-ings for wavelength-demultiplexing of free space waves fromguided waves,” Opt. Express. 12, 3072–3078 (2004).
13. M. Hamada, T. Noritomo, K. Nishio, and S. Ura, “Integrationof fine-tuned 0.28 �m period gratings for integrated-optic add�drop multiplexing,” in Technical Digest of International Sym-posium on Contemporary Photonics Technology (NationalInstitute of Information and Communications Technology, To-kyo, Japan, 2004), pp. 97–98.
14. S. Ura, M. Hamada, J. Ohmori, K. Nishio, and K. Kintaka,“Integrated-optic free-space-wave drop demultiplexer fabri-cated by using interference exposure method,” in DiffractiveOptics and Micro-Optics, Technical Digest of OSA TopicalMeeting (Optical Society of America, 2004), paper DWA3.
15. X. Mai, R. Moshrefzadeh, U. J. Gibson, G. I. Stegeman, andC. T. Seaton, “Simple versatile method for fabricating guided-wave gratings,” Appl. Opt. 24, 3155–3161 (1985).
Fig. 7. Wavelength dependences of output efficiencies from thefirst GMS FGC and DGM-DBR (open circles) and the second GMSFGC and DGM DBR (closed circles).
