IEICE TRANS. ELECTRON., VOL.E100–C, NO.2 FEBRUARY 2017187

INVITED PAPER Special Section on Fabrication Technologies Supporting the Photonic/Nanostructure Devices

Assembly Technologies for Integrated Transmitter/Receiver OpticalSub-Assembly Modules

Keita MOCHIZUKI†a), Tadashi MURAO†, Mizuki SHIRAO†, Yoshiyuki KAMO†, Nobuyuki YASUI†,Takahiro YOSHIMOTO†, Daisuke ECHIZENYA†, Masaya SHIMONO†, Hidekazu KODERA†,

Masamichi NOGAMI†, and Hiroshi ARUGA†, Members

SUMMARY We have succeeded in developing three techniques, a pre-cise lens-alignment technique, low-loss built-in Spatial Multiplexing op-tics and a well-matched electrical connection for high-frequency signals,which are indispensable for realizing compact high-performance TOSAsand ROSAs employing hybrid integration technology. The lens positionwas controlled to within ±0.3 μm by high-power laser irradiation. All com-ponents comprising the multiplexing optics are bonded to a prism, enablingthe insertion loss to be held down to 0.8 dB due to the dimensional accu-racy of the prism. The addition of an FPC layer reduced the impedancemismatch at the junction between the FPC and PCB. We demonstrated acompact integrated four-lane 25 Gb/s TOSA (15.1 mm × 6.5 mm × 5.6mm) and ROSA (17.0 mm × 12.0 mm × 7.0 mm) using the built-in spa-tial Mux/Demux optics with good transmission performance for 100 Gb/sEthernet. These are respectively suitable for the QSFP28 and CFP2 formfactors.key words: hybrid integration, optical sub-assembly, 100 Gb/s Ethernet

1. Introduction

The rapid growth of data traffic due to expansion of theCloud and content streaming services via the Internet re-quires high-capacity optical transmission systems [1]. Tomeet this demand, Ethernet transmission systems used forcommunication within or between data centers are currentlymoving from 10 Gb/s to 100 Gb/s [2], with 400 Gb/s sys-tems on the horizon [3]. For long-reach applications over 10km, 100 Gb/s Ethernet (GbE) uses a four-lane 25 Gb/s opti-cal architecture employing a local area network wavelengthdivision multiplexing (LAN-WDM) technique to meet theneed for cost reduction and energy saving. In addition,the over-the-top operators (OTTs) which run their own datacenters have a strong need for low-cost Ethernet systems.Several multi-source agreements (MSAs), employing, e.g.,coarse WDM (CWDM) [4], parallel single-mode (PSM) [5]etc., are proposed to relax or avoid the difficulties presentedby high-cost LAN-WDM optics.

The downsizing of the optical transceivers is proceed-ing in parallel with the development of the transmission sys-tem, as shown in Fig. 1. The centum form-factor pluggable(CFP) standard [6] has been adopted for the first-generation100GbE transceivers, consisting of four individual 25 Gb/stransmitter optical sub-assemblies (TOSAs) and receiver op-tical sub-assemblies (ROSAs), and an optical multiplexer

Manuscript received June 16, 2016.Manuscript revised September 16, 2016.†The authors are with the Mitsubishi Electric Corporation,

Kamakura-shi, 247–8501 Japan.a) E-mail: Mochizuki.Keita@ab.MitsubishiElectric.co.jp

DOI: 10.1587/transele.E100.C.187

(Mux) and demultiplexer (Demux). The bit rate density (theratio of the bit rate to the footprint) of 0.87 Gb/s/cm2 forthe 100GbE CFP is rather smaller than the 1.31 Gb/s/cm2

of the small form factor pluggable plus (SFP+) [7] usedfor 10GbE. Downsized 100GbE transceivers, CFP2 (2.24Gb/s/cm2) [6], CFP4 (5.06 Gb/s/cm2) [6], QSFP28 (7.69Gb/s/cm2) [8] etc., are prerequisite for superseding thecurrent systems. Considering the manufacturing strate-gies for such small form-factors, only integrated opti-cal sub-assemblies (OSAs) can be used to construct suchtransceivers.

A number of research groups have proposed variousOSA structures with several integration technologies whichare basically categorized as monolithic and hybrid types.Mounting four laser diodes (LDs) and an optical Mux,such as a multi-mode interferometer (MMI), on the sameindium-phosphide (InP) chip is proposed for the formertype [9], [10]. While there is an advantage of simplified as-sembly, this structure involves the difficulty of chip fabrica-tion. For the latter, hybrid, type, an OSA constructed by in-tegrating a planar lightwave circuit (PLC) [11]–[14] or built-in spatial optics with four LDs has been developed [15]–[17]. For these structures, the optics tend to be complex,reducing the manufacturing yield due to unwanted misalign-ment of the assembly, so a precise alignment technique isessential.

We believe that, while monolithic integration is the ulti-mate solution for ultra-compact, low-cost high-performanceOSAs, the hybrid integration approach is the best current so-lution for the immediate productization of an OSA, becausedeveloping an assembly is rather easier than developing achip. In addition, the hybrid integration has the advantagesin improving output power of the TOSA and sensitivity ofthe ROSA compared with the monolithic integration due to

Fig. 1 Bit rate vs. footprint of optical transceivers for 10, 100 and400GbE.

Copyright c© 2017 The Institute of Electronics, Information and Communication Engineers

188IEICE TRANS. ELECTRON., VOL.E100–C, NO.2 FEBRUARY 2017

the low insertion losses of several optical components, e.g.Mux/Demux optics and an isolator.

In this paper, we report a compact integrated four-lane25 Gb/s TOSA (15.1 mm × 6.5 mm × 5.6 mm) and ROSA(17.0 mm × 12.0 mm × 7.0 mm) [17] with built-in SpatialMux/Demux optics, using three key techniques, a preciselens-alignment technique [14], [16], low-loss built-in Spa-tial Mux/Demux optics [17] and a matched electrical con-nection for the high-frequency signals [18]. We demon-strate electrical-optical performance which fully meets the100GbE specification.

2. Structures of TOSA and ROSA

Figure 2 outlines the structures of the four-lane integratedTOSA and ROSA [17]. For the TOSA, four 25 Gb/s electri-cal signals are input from a flexible printed circuit (FPC) andfour optical signals are generated at different wavelengths inthe 1.3 μm band by individual electroabsorption-modulatorlasers (EMLs), mounted on a thermos-electric cooler (TEC).The optical signals modulated at 25 Gb/s are combined in abuilt-in Mux and launched into an optical output interface.For the ROSA, the multiplexed 100 Gb/s optical input is di-vided into four independent 25 Gb/s signals in built-in De-mux optics, and the signals are received by a four-lane PDarray. The resulting four 25 Gb/s electrical signals are out-put through transimpedance amplifiers (TIAs) and an FPC.

Generally, in order to realize the various characteristics

Fig. 2 Structures in the integrated (a) transmitter and (b) receiver modulefor 100 Gb/s Ethernet.

required for a 100 Gb/s Ethernet system, namely an opticalmodulation amplitude (OMA) of at least −1.3 dBm for theTOSA [2], a receiver sensitivity of less than −8.6 dBm (ex-pressed as OMA) for the ROSA [2], and low-loss electricaltransmission of both, it is necessary to resolve three majorissues. First, a precise submicron lens-alignment techniqueis indispensable for realizing lens-coupled multi-lane opticalsystems. Second, the compact built-in WDM optics need tobe assembled in the integrated OSA with low losses. Third,an electrical transmission path with good performance forhigh speed signals is needed. The technologies used in thismodel to resolve these issues are described below [14], [16]–[18].

3. Precise Lens-Alignment Technique

For conventional single-lane optics, the optical alignmentis accomplished by positioning the optical output interface,e.g. a single mode fiber (SMF), receptacle etc., after theEML and lens have been mounted. However, this conven-tional approach cannot be applied to multi-lane optics, be-cause it is necessary to align the lenses after mounting theEMLs and the optical output interface. Figure 3 plots thecalculated coupling loss between the EML and the opticaloutput interface as a function of the position of the lens per-pendicular to the optical axis when the positions of the EMLand optical output interface are fixed. For comparison, wealso show the result for the conventional case as a dashedblue line in the same figure. We assume that the spot size ofthe EML is 0.83 μm and that of the optical output interface is3.3 μm, while the magnification of the lens is 4× to compen-sate for the difference in the spot sizes for each optic. Dueto the magnification of the lens, the allowable positional tol-erance of the lens assembly for the multi-lane optics reducesto ±0.3 μm from ±1.5 μm for the single-lane optics wherethe target coupling loss is assumed to be 1.3 dB (the solidline in Fig. 3). The variation of the lens position is usuallymore than a few microns when it is fixed by laser welding,gluing etc., so that excessive loss due to lens misalignmentis normally unavoidable with multi-lane optics.

In order to overcome this problem, we propose a pre-cise lens-alignment technique to control the lens position af-ter the lens misalignment has occurred. To achieve the sub-micron control needed to compensate the misalignment, wecause plastic deformation of the stainless steel of the lens

Fig. 3 Coupling loss as a function of the relative position of the lens andoptical output interface, where 0 corresponds to the optimum position.

MOCHIZUKI et al.: ASSEMBLY TECHNOLOGIES FOR INTEGRATED TRANSMITTER/RECEIVER OPTICAL SUB-ASSEMBLY MODULES189

Fig. 4 (a) Construction of the proposed lens holder for controlling thelens position by high-power laser irradiation. (b), (c), (d) Mechanism forshifting the lens by state-variation of the top surface of the holder in (b) ±x,(c) −y, (d) +y directions calculated by 3D FEM (Quick Welder software).Blue corresponds to a larger shift in each direction.

Fig. 5 Experimental results for relative position of the lens in the y-axisas a function of the number of laser irradiations (9 samples). The grayregion indicates the target tolerance of ±0.3 μm.

holder by high-power laser irradiation. Figure 4 depicts theconcept and principles of the proposed technique. The lensis bonded to the lens holder at the top of the lens. Irradi-ating the top of the holder with a high-power laser, e.g. aYAG laser, causes a state change of the lens holder, induc-ing a compressive stress at the surface, with solidificationafter melting. The lens holder is thus distorted slightly. Wecalculate the shift in lens position by the 3-D finite elementmethod (FEM) as shown in Figs. 4 (b)–(d). When the top ofthe lens holder is irradiated, the bending of the holder shiftsthe lens in the ±x and −y directions as shown in Figs. 4 (b)and (c). For the +y direction, we deform the base underneaththe lens holder as shown in Fig. 4 (a). If the top of the baseis irradiated, the base adopts a downward convex shape. Asa result of this deformation of the base, the lens shifts in the+y direction as shown in Fig. 4 (d). In the proposed method,we can detect the direction and degree in which the lensesshould be moved by monitoring the beam angles and posi-tions (e.g., a CCD camera) at the end of the multiplexer.

Figure 5 plots the experimental results for relative lensposition in the y-axis as a function of the number of laserirradiations. The gray region indicates the target tolerance

Fig. 6 (a) Schematic structure of the narrow lens holder. (b), (c), (d)mechanism for shifting the lens by state-variation of the top surface of theholder or base in (b) ±x, (c) −y, (d) +y directions calculated by 3D FEM(ANSYS Software). Red corresponds to a larger shift in each direction.

of ±0.3 μm. We put nine lens samples in an initial position+1.7 μm away from the optimum position, the thickness ofthe lens holder being 0.4 mm, the width in the x-axis being2.8 mm, the depth in the z-axis being1.8 mm and the heightin the y-axis being 2.6 mm. The position of the lens can becontrolled to within the target range of ±0.3 μm from theoptimum position by less than three laser irradiations withenergies of 0.7 J and 0.5 J. The laser power is determined byconsidering both minimization of the number of irradiationsand the degree of shift. The irradiation cycle is not to belimited to three times. The irradiations are carried out untilthe lens is controlled to be within the target range. Theseresults indicate that the precise lens-alignment technique isfully established [14], [16].

However, it is difficult to apply the proposed techniqueto a small TOSA complying with a small transceiver form-factor such as CFP4 or QSFP28, because the lens holderslimit the ability to downsize the TOSA. To overcome thisproblem, we developed the new lens holder avoiding sidepillars shown in Fig. 6 (a). In order to control the lens posi-tion precisely, we use the same method as previously, irradi-ating the top of the lens holder or the base with a high-powerlaser, as shown in Figs. 6 (b)–(d). The new lens holder al-lows us to reduce the overall width of the lens-alignmentsystem.

4. Low-Loss Built-In Spatial Mux/Demux Optics

The low-loss built-in technique is applied to both OSAs.However, the Spatial Demux optics for the ROSA requiresmore precise alignment than the Spatial Mux optics for the

190IEICE TRANS. ELECTRON., VOL.E100–C, NO.2 FEBRUARY 2017

TOSA because the polarizations of incident light to the Spa-tial Mux and Demux optics are fixed and random. Hence,we explain the detail of this technique with focusing on theROSA. The built-in Spatial Demux optics require precisecontrol when assembling the optical components. Angularmisalignment of the band-pass filters will cause the angleand position of the reflected light to depart from the design,so that the aperture loss for light incident on the PDs in-creases. Since the insertion loss depends on the incident an-gle to the band-pass filters, the Spatial Demux optics needsome kind of adjustment structure. These extra structuresresult in a cost increase and larger footprint. In order toavoid the excess loss without adding extra structures, we in-tegrated all the optical components into one assembly blockas shown in Fig. 7. The Spatial Demux optics consist ofband-pass filters, a mirror, a triangular prism, and a lensarray attached together to a small prism 5 mm × 8 mm.The input light is demultiplexed by multiple reflections be-tween the band-pass filters and the mirror. Each light pass-ing through a band-pass filter is bent towards the PD arrayby the triangular prism. This array of optics allows us tocontrol the misalignment of the band-pass filters to within±0.05◦, because the band-pass filters and the mirror, whichrequire the most precise assembly of all the optical compo-nents, are bonded to the accurately shaped prism.

We investigated the relationship between excess inser-tion loss and band-pass filter misalignment in angle and po-sition. Figure 8 shows the calculated contour map of aper-

Fig. 7 Proposed built-in Spatial Demux optics with all optical compo-nents integrated as one block, with the band-pass-filters, mirrors, and lensarray attached together to a prism; (a) top view and (b) side view.

Fig. 8 Calculated excess aperture loss with angle and position of the inci-dent light to a band-pass filter. The region enclosed with a red line outlinesthe target aperture loss of less than 1.0 dB.

ture loss versus incident angle and position when the aper-ture of the PD array is 20 μm φ, the lens focal length is 1.0mm, and the incident beam diameter is 300 μm. The regionenclosed with a red line indicates the edge of the target aper-ture loss of less than 1.0 dB. For non-multi-reflection optics,the optical alignment is accomplished by accurately locatingthe band-pass filter for each lane independently as shown inFig. 8. On the other hand, for multi-reflection optics, thisconventional procedure is not suitable because the angularshift of the incident light is twice that of the band-pass filterwhich is reflecting the light. Moreover, these variations ofthe incident light are cumulative over the following band-pass filters. In particular, Lane 3 is the one most affected bymisalignment of the band-pass filters.

In order to estimate accurately the required precision ofband-pass filter alignment, we simulated the dependency ofthe yield ratio on the variation in band-pass filter positioningusing the Monte Carlo method (10,000 trials). Figure 9 il-lustrates the distribution of the incident angle and positionalerrors for the Lane 3 light, assuming that the tilts of the Lane0, 1 and 2 band-pass filters are random, following a normaldistribution with 2σ equal to (a) 0.05◦ and (b) 0.10◦ for eachtrial. Each point on the map indicates the results of a trialobtained by calculation based on ray tracing. In this simu-lation, the yield ratio of the Demux optics is expected to be99.93% for case (a) and 94.5% for case (b). We concludethat an assembly accuracy of ±0.05◦ for the band-pass fil-

Fig. 9 Distribution of the incident angle and positional variations on alens in Lane 3 vs. tilt angles of the band-pass filters using the Monte Carlomethod (10,000 trials) where the misalignment error of each band-passfilter is chosen randomly following a normal distribution of 2σ equaling(a) 0.05◦ and (b) 0.10◦.

MOCHIZUKI et al.: ASSEMBLY TECHNOLOGIES FOR INTEGRATED TRANSMITTER/RECEIVER OPTICAL SUB-ASSEMBLY MODULES191

Fig. 10 Histogram of angles of band-pass filters attached to a prism forthe fabricated built-in Demux optics.

ters is a good target, yielding a low excess loss of 1.0 dB.Figure 10 is the histogram of the absolute angle of the band-pass filters attached to a prism for the fabricated built-in De-mux optics. The angular precision fully meets the required±0.05◦ [17].

5. Electrical Connection for High-Frequency Signals

Coaxial cables and connectors have good high-frequencyperformance, so they are generally used as the electricalsignal interface for the high-frequency signals of an OSA.However, such electrical interfaces cannot to be used foran integrated OSA because of their large size. The use ofFPC instead of coaxial cables and connectors is indispens-able for a compact integrated OSA. The challenge is to min-imize any degradation of the optical waveform due to poorhigh-frequency characteristics at the FPC interface causedby reflection at its junction with the printed circuit board(PCB).

To overcome this, we propose the new approach to animpedance-matched PCB-FPC connection shown in Fig. 11.The proposed FPC has three metallic layers, a padding layer,a signal-line layer and a ground layer, whereas a conven-tional FPC has only two metallic layers, a signal-line layerand a ground layer. A 50 Ω microstrip transmission line(MSL) is formed on the FPC by the 50 μm thick polyimidebase material. The PCB has just two layers, a signal layerand a ground layer. An MSL is formed on the PCB usingthe 250 μm thick core material. The FPC is mounted on thePCB such that the FPC’s padding layer faces the PCB’s sig-nal layer. We design the FPC pad to have the same structureas in a conventional 10 Gb/s XMD-MSA module, whichhas a pad spacing D of 0.79 mm, and a pad width WP of0.45 mm. In the junction region, we create a coplanar wave-guide (CPW) with a wide FPC signal track in the signal-linelayer and a notched PCB ground pattern as shown in Fig. 12.With the proposed structure, we can control the impedanceby the width of the wide signal track alone. We do not needto change the form of the padding layer. This means thatthis type of connection enables us to reduce the impedance-mismatch between the PCB and FPC signal lines withoutcompromising the mountability.

We simulated the high-frequency characteristics by theFEM. Figure 13 shows the reflection parameter S 11 where

Fig. 11 Structures of (a) PCB, (b) FPC and (c) PCB to FPC connection.The pink tracks indicate transmission line patterns and the brown tracks in-dicate ground plane patterns. An impedance matched coplanar waveguideis formed at the PCB to FPC connection by introducing a notch in the PCBground plane and a wide signal track on the FPC.

Fig. 12 Cross section of PCB to FPC connection. The impedance can becontrolled by the width of wide signal track WS.

Fig. 13 Calculated reflection parameter S 11 of the PCB to FPC connec-tion as a function of frequency for a conventional two-layer FPC and theproposed three-layer FPC.

WS is 0.8 mm, yielding a 50 Ω CPW. For comparison, wealso show the calculated results for a conventional two-layerFPC. The input and output ports are assumed to be 2.5 mmlong 50 ΩMSLs formed at the PCB and FPC edges. For theconventional two-layer structure, the maximum S 11 is worsethan 10 dB in the frequency range from 0 to 30 GHz. On theother hand, the proposed three-layer structure improves S 11

to better than 22 dB. These results indicate that the proposed

192IEICE TRANS. ELECTRON., VOL.E100–C, NO.2 FEBRUARY 2017

three-layer FPC structure has good enough high-frequencycharacteristics for use in the integrated OSA [18].

6. Characteristics of Prototype TOSA and ROSA

Figure 14 shows photographs of the fabricated 100GbE inte-grated TOSA and ROSA with the built-in Spatial Mux andDemux optics. The narrow lens holder structure shown inFig. 6 is applied to the TOSA. The TOSA and ROSA pack-ages excluding the LC receptacle, fiber and FPCs are 15.1mm L × 6.5 mm W × 5.6 mm H, and 17.0 mm L × 12.0mm W × 7.0 mm H [17], respectively, which comply withthe QSFP28 and CFP2 form factors. For both of the OSAs,two FPCs are attached directly to the ceramic feedthroughsformed at the rear of the packages without any lead-pins soas to reduce the package width. One of the FPCs is for thehigh frequency signals and the other is for the DC bias of theEMLs, TEC, PDs and TIAs. The width of the ROSA is dueto the four discrete TIAs which were mounted on the base.

6.1 TOSA

Figure 15 shows the measured relative output power spec-trum of the fabricated 100 Gb/s integrated TOSA. The fourEMLs are driven simultaneously with an operating cur-rent of 60 mA. The wavelengths of the lanes were 1295.5,1300.2, 1304.9 and 1309.3 nm, respectively, satisfying thespecification of 100 Gb/s Ethernet. The DC output pow-ers of the lanes were +4.40, +4.22, +4.37 and +4.4 dBm,respectively, which enables the average output power to bemore than +1.0 dBm with 28 Gb/s non-return to zero (NRZ)modulation of all lanes. The coupling loss for each lanewas less than −2.6 dB which included the insertion losses

Fig. 14 Photographs of developed (a) 100 Gb/s integrated TOSA and(b) 100 Gb/s integrated ROSA.

of isolators and the built-in spatial Mux optics. In addi-tion, the variations in the output powers for all lanes wereless than ±0.5 dB over a case temperature range from −5to +80◦C. The precise lens-alignment technique allows usto obtain high output power. Figure 16 shows fiber outputpowers depending on the operating currents with the EMLand case temperatures of 55 and 30 ◦C. There was no kinkwhen the operating current is less than 75 mA for each lane.

We also evaluated the four 28 Gb/s waveforms in aback-to-back configuration as shown in Fig. 17. All the datawere measured with all the lanes modulated at the same time

Fig. 15 The relative output power spectrum of the fabricated 100 Gb/sintegrated TOSA. The four EMLs are driven simultaneously with an op-erating current of 60 mA. The colored regions indicate the bandwidths ofeach lane per IEEE802.3ba.

Fig. 16 The fiber output powers depending on the operating currentswith the EML temperature of 55 ◦C and the case temperature of 30 ◦C.

Fig. 17 Four 28 Gb/s waveforms of the fabricated integrated TOSA ina back-to-back configuration for (a) Lane 0, (b) Lane 1, (c) Lane 2 and(d) Lane 3.

MOCHIZUKI et al.: ASSEMBLY TECHNOLOGIES FOR INTEGRATED TRANSMITTER/RECEIVER OPTICAL SUB-ASSEMBLY MODULES193

using 28 Gb/s PRBS31 (231 − 1 pseudo-random binary se-quence) NRZ electrical signals with a low voltage of 1.75Vpeak-to-peak. Clear eye openings were successfully obtainedwith large mask margins of 15, 12, 11 and 17% for eachlane. The extinction ratios were 9.4, 9.1, 9.2 and 9.5 dB,respectively.

6.2 ROSA

The measured transmittances of the four lanes are plotted inFig. 18; the lower plot covers 0 to −25 dB, the upper part isa magnified plot covering 0 to −2 dB. These transmittancesare evaluated from the received currents of the four-lane PDarray normalized by the response of each PD. The adjacent-lane isolation for all lanes was more than 20.0 dB, and theoptical losses of Lanes 0, 1, 2 and 3 were 0.21, 0.10, 0.19and 0.8 dB respectively. The loss of Lane 3 was larger thanthose of the others. This is due to the aperture loss of thecollimated light path to Lane 3, which is the longest.

Figure 19 is the measured bit error ratio (BER). Theinput optical signal was NRZ modulated with PRBS31 at25.78125 Gb/s, the extinction ratio being 10.5 dB. The sen-sitivity criterion of IEEE 802.3ba [2] is also marked at the−10.8 dBm average power level, which is converted from−8.6 dBm OMA. The average power sensitivities of Lanes

Fig. 18 Measured transmittances of the four lanes evaluated from thereceived currents of the four-lane PD array normalized by the response ofthe four-lane PD array.

Fig. 19 BER results for the four lanes with and without adjacent lanesignals.

0, 1, 2 and 3 are −13.1, −12.7, −12.1 and −12.7 dBm, re-spectively. Figure 18 also shows each lane’s BER resultsin the presence of an adjacent lane. The loss of sensitivitydue to the crosstalk is less than 0.1 dB. These results indi-cate that the electrical and optical isolation is high enoughto suppress the crosstalk.

7. Conclusions

We have developed three assembly technologies for the in-tegrated TOSA and ROSA, a precise lens-alignment tech-nique, low-loss built-in Spatial Multiplexing optics and awell matched electrical connection for the high-frequencysignals. This combination of three techniques allows us torealize a compact TOSA 15.1 mm L × 6.5 mm W × 5.6 mmH, and a compact ROSA 17.0 mm L × 12.0 mm W × 7.0mm H, with good 100 Gb/s Ethernet transmission perfor-mance. They are respectively suitable for the QSFP28 andCFP2 form factors. We expect these assembly technologieswill contribute to the realization of the next generation of in-tegrated optical devices for transmission systems operatingat 100 Gb/s and over.

References

[1] CISCO, “The zettabyte era –Trends analysis,” White paper, May2015.

[2] IEEE P802.3ba 40 Gb/s and 100 Gb/s ethernet.Available: http://www.ieee802.org/3/ba

[3] IEEE P802.3bs 400 Gb/s ethernet.Available: http://www802.org/3/bs/

[4] 100G CWDM4 MSA Technical Specification.Available: http://www.cwdm4-msa.org/

[5] 100G PSM4 Specification. Available: http://www.psm4.org/[6] CFP Multi-source Agreement. Available: http://www.cfp-msa.org/[7] SFF-8431 Specifications for Enhanced Small Form Factor Pluggable

Module SFP+. Available: ftp://ftp.seagate.com/sff/SFF-8431.PDF[8] SFF-8665 Specifications for QSFP+ 28 Gb/s 4X Pluggable

Transceiver Solution (QSFP28).Available: ftp://ftp.seagate.com/sff/SFF-8665.PDF

[9] T. Fujisawa, S. Kanazawa, Y. Ueda, W. Kobayashi, K. Takahata,A. Ohki, T. Ito, M. Kohtoku, and H. Ishii, “Low-loss cascadedMach-Zehnder multiplexer integrated 25-Gbit/s × 4-lane EADFBlaser array for future CFP4 100 GbE transmitter,” IEEE J. QuantumElectron., vol.49, no.12, pp.1001–1007, Dec. 2013.

[10] S. Kanazawa, W. Kobayashi, Y. Ueda, T. Fujisawa, K. Takahata, T.Ohno, T. Yoshimatsu, H. Ishii, and H. Sanjoh, “30-km error-freetransmission of directly modulated DFB laser array transmitter op-tical sub-assembly for 100-Gb application,” J. Lightwave Technol.,vol.34, no.15, pp.3646–3652, Aug. 2016.

[11] Y. Beak, Y.T. Han, C.W. Lee, D.H. Lee, O.K. Kwon, J.W. Shin,S.H. Park, and Y.A. Leem, “Optical components for 100G ethernettransceivers,” The 17th Opto-Electronics and Communication Con-ference, 4D1-2, Busan, Korea, July 2012.

[12] T. Ohyama, Y. Doi, W. Kobayashi, S. Kanazawa, K. Takahata,A. Kanda, T. Kurosaki, T. Tanaka, T. Ohno, H. Sanjoh, and T.Hashimoto, “Compact hybrid integrated 100-Gb/s transmitter op-tical sub-assembly using optical Butt-coupling between EADFBlasers and silica-based AWG multiplexer,” J. Lightwave Technol.,vol.34, no.3, pp.1038–1046, 2016.

[13] Y. Doi, M. Oguma, T. Yoshimatsu, T. Ohno, I. Ogawa, E. Yoshida, T.Hashimoto, and H. Sanjo, “Compact high-responsivity receiver op-tical subassembly with a multimode-output-arrayed waveguide grat-

194IEICE TRANS. ELECTRON., VOL.E100–C, NO.2 FEBRUARY 2017

ing for 100-Gb/s ethernet,” J. Lightwave Technol., vol.33, no.15,pp.3286–3292, 2015.

[14] T. Murao, N. Yasui, K. Mochizuki, M. Shimono, H. Kodera, D.Morita, T. Yamatoya, and H. Aruga, “Lens alignment technique us-ing high-power laser for hybrid integrated multi-channel transmitteroptical sub-assemblies,” IEEE Photon. Technol. Lett., vol.25, no.20,pp.1958–1960, 2013.

[15] F. Nakajima, M. Kawamura, K. Oki, Y. Koyama, A. Yasaki,Y. Fujimura, and H. Hara, “100Gbit/s compact receiver mod-ule with the built-in optical de-multiplexer,” IEEE Photon. Conf.,pp.305–306, Sept. 2013.

[16] T. Murao, N. Yasui, T. Shinada, Y. Imai, K. Nakamura, M. Shimono,H. Kodera, Y. Morita, A. Uchiyama, H. Koyanagi, and H. Aruga,“Integrated spatial optical system for compact 28-Gb/s × 4-lanetransmitter optical subassemblies,” IEEE Photon. Techbol. Lett.,vol.26, no.22, pp.2275–2278, 2014.

[17] K. Mochizuki, H. Aruga, H. Itamoto, K. Yamagishi, Y. Horiguchi,S. Nishikawa, R. Takemura, M. Nakaji, and A. Sugitatsu, “Four-channel integrated receiver with a built-in spatial demultiplexer op-tics for 100 Gb/s ethernet,” IEICE Trans. Electron., vol.E96-C, no.7,pp.981–988, July 2013.

[18] M. Shirao, N. Ohata, N. Yasui, K. Uto, T. Fukao, T. Hatta, H.Aruga, and T. Mizuochi, “A 1.55 μm 40 Gbps EML TOSA Employ-ing a novel FPC connection,” J. Lightwave Technol., vol.32, no.19,pp.3344–3350, 2014.

Keita Mochizuki received the B.E. de-gree in Electric and Electronic Engineering fromKyoto University in 2006 and M.E. degree inElectronic Science and Engineering from KyotoUniversity in 2008. In the same year, he joinedthe Mitsubishi Electric Corp., Japan, where he iscurrently engaging in the study of optoelectron-ics devices for optical communication. He is amember of the Institute of Electronics, Informa-tion and Communication Engineer (IEICE).

Tadashi Murao was born in Ibaraki, Japan,on December 16, 1982. He received the B.S.,M.S., and Ph.D. degrees in media and networktechnologies from Hokkaido University, Sap-poro, Japan, in 2006, 2008, and 2010, respec-tively. From 2008 to 2011, he was a ResearchFellow of the Japan Society for the Promotionof Science (JSPS). He is currently with the In-formation Technology R&D Center, MitsubishiElectric Corporation, Kanagawa, Japan. Hiscurrent research interests include the optimiza-

tion of optelectronic-device design and theoretical investigation of opti-cal phenomena in band-gap devices, and so on. Dr. Murao is a memberof the Institute of Electronics, Information and Communication Engineers(IEICE), and the Institute of Electrical and Electronic Engineers (IEEE).

Mizuki Shirao (S’08–M’11) was born inKanagawa Prefecture, Japan, in 1984. He re-ceived the B.E., M.E., and Ph.D. degrees in elec-trical and electronic engineering from the To-kyo Institute of Technology, Tokyo, Japan, in2007, 2009, and 2011, respectively. He wasa Research Fellow with the Japan Society forthe Promotion of Science, Tokyo, from 2010 to2011. During his Ph.D. work, he demonstrateda 1.3 μm AlGaInAs/InP quantum well transis-tor laser. He joined Mitsubishi Electric Inc. at

Kanagawa, Japan, in 2011, and works with the Optoelectronics Group. Hiscurrent research interests include electro-absorption modulator lasers, highspeed optical transmitters and receivers. Dr. Shirao is a member of theIEEE Photonics Society and the Institute of Electrical, Information, andCommunication Engineers (IEICE) of Japan.

Yoshiyuki Kamo received the M.E. degreein Material Chemistry and Engineering fromOsaka University in 2009. In the same year,he joined the Mitsubishi Electric Corp., Japan,where he is currently engaging in the Develop-ment of Assembly Technology for Communica-tion Devices. He is a member of the Electro-chemical Society of JAPAN.

Nobuyuki Yasui received the B.E. and M.E.degrees in electronic engineering and systemdesign engineering from Kanazawa Institute ofTechnology in 1998 and 2000, respectively. In2000, he joined Mitsubishi Electric Corporation,Japan, where he is currently engaging in the ap-plication of optoelectronics devices for opticalcommunication.

Takahiro Yoshimoto received the B.E. de-gree in Mechanical Engineering from KyushuInstitute of Technology in 2005 and M.E. degreein Mechanical Engineering from Kyushu Insti-tute of Technology in 2007. In the same year,he joined the Mitsubishi Electric Corp., Japan,where he is currently engaging in the study ofstructural reliability and fracture mechanics forelectronic products.

Daisuke Echizenya received the B.E. de-gree in Mechanical Engineering from HokkaidoUniversity in 2000 and M.E. degree in Mechan-ical Engineering from Hokkaido University in2003. In the same year, he joined the MitsubishiElectric Corp., Japan, where he is currently en-gaging in the study of design optimization forstructure of electronic products.

MOCHIZUKI et al.: ASSEMBLY TECHNOLOGIES FOR INTEGRATED TRANSMITTER/RECEIVER OPTICAL SUB-ASSEMBLY MODULES195

Masaya Shimono received the B.E. de-gree in Mechanic and Electric Engineering fromTohoku University in 2004 and M.E. degree inAerospace Engineering from Tohoku Universityin 2006. In the same year, he joined the Mitsubi-shi Electric Corp., Japan, where he is currentlyengaging in the study of production engineering.

Hidekazu Kodera received the B.E. degreein Mechanical Engineering from Osaka Univer-sity in 1997 and M.E. degree in Mechanical En-gineering from Osaka University in 1999. Inthe same year, he joined the Mitsubishi ElectricCorp., Japan, where he is currently engaging inthe study of production engineering for opticaldevice etc.

Masamichi Nogami graduated from theToin Technical College, Kanagawa, Japan, in1986. In the same year he joined MitsubishiElectric Corporation, Kanagawa, Japan, wherehe has been engaged in research and develop-ment on fully integrated optical transceivers. Heis currently engaging in the study of optoelec-tronics devices for optical communication. He isa member of the Institute of Electronics, Infor-mation, and Communication Engineers (IEICE)of Japan.

Hiroshi Aruga received the B.E. degreein materials science form Tohoku University,Miyagi, Japan, in 1992, and the M.E. degrees inelectrical engineering from Sophia University,Tokyo, Japan in 1994. He is with MitsubishiElectric Corporation, Kanawaga, Japan, work-ing in the field of design and development of mi-crowave components for space satellite and op-tical components for telecommunication systemfrom 1994. He is a member of IEEE PhotonicsSociety and the Institute of Electronics, Infor-

mation and Communication Engineers (IEICE).