1. Introduction

Over the past decades, Internet-driven data center traffic has been rising exponentially in order to meet the ever-growing cloud computing and high-definition video distribution application requirements. Since the release of IEEE 802.3 40G/100G Ethernet standard in 2010, defining both short reach optical interconnect based on 850-nm vertical-cavity surface-emitting laser (VCSEL) and multimode fiber (MMF) technologies, and single mode fiber (SMF) based 1310-nm 10-km and 40-km transmission [1], the photonic industry has been gradually shifting from earlier generation 10Gb/s transceiver to Coarse Wavelength Division Multiplexing (CWDM) based technologies in order to scale up the transmission capacity of metro, large campus, and large data center optical networks [2,3]. The primary requirements of photonic device technology supporting future optical datacom applications are ever higher transmission data rate on each optical fiber, and compatibility with high volume low-cost manufacturing. Compared with existing optical datacom transceiver technologies which are mostly based on discrete semiconductor laser components, multi-wavelength optical chips with all the passive and active components integrated on a single wafer offer clear advantages in footprint, packaging cost, and stability [4]. 1310-nm distributed feedback (DFB) lasers based directly modulated lasers (DMLs) have been selected in the 100GBASE-LR4 standard for 10-km reach applications due to its relatively low cost [5]. Both InGaAsP- and AlGaInAs-based multiple quantum wells (MQWs) active region materials have been used as the active region material for 1310-nm DMLs. In general, AlGaInAs-based DFB lasers can reach higher modulation bandwidth and superior high-temperature performance [6,7]. However, the Al-containing active layers could be oxidized when exposed to the air, making AlGaInAs photonic integrated chip fabrication still challenging [8]. InGaAsP-based DFB lasers, on the other hand, are more insensitive to the oxidation, and thus well suited as active elements in monolithically integrated chips. The main tradeoff of the InGaAsP MQWs materials is relatively small conduction band offset resulting in high carrier leakage and typically smaller modulation bandwidth [8].

In order to improve the optical modulation bandwidth of DMLs based on InGaAsP MQWs materials, the traditional approach has been through increasing the carrier-photon resonance (CPR) frequency by optimizing the standard semiconductor laser parameters such as internal quantum efficiency, differential gain, RC time constant [9]. A more novel approach involves enhancing the modulation bandwidth by taking advantage of the photon-photon resonance (PPR) effect [10]. PPR effect has been shown to overcome the carrier-photon modulation bandwidth limitation of the DMLs and capable of enhancing laser modulation bandwidth beyond the CPR frequency. PPR effect induced bandwidth enhancement effect has been demonstrated through external optical injection with a master laser source operating at detuned wavelengths [11] or by optical feedback through an integrated passive optical waveguide [12,13]. State-of-the-art results reported optical injection-locked VCSEL with >80 GHz small signal modulation bandwidth demonstrating the great potential of the PPR effect for bandwidth enhancement, but external optical injection involves sophisticated test setups with high-power single-mode external-injection light source, and the test platform is very sensitive to mechanical vibrations and other environmental factors [11], making its real-world applications in optical datacom networks impractical. Furthermore, there are no reports of large signal time domain modulation results based on the optical injection approach up to now. In comparison, PPR bandwidth enhancement through monolithically integrated optical feedback section has the advantages of the simple device structure, small footprint, natural insensitivity to environmental factors, and compatibility with volume production based on standard semiconductor manufacturing technology. Passive feedback laser (PFL) with 40 Gb/s data rate has been demonstrated well exceeding the intrinsic semiconductor laser CPR modulation bandwidth for a single channel device [13].

In this paper, we report a 1.3-µm dual-wavelength monolithically integrated optical chip with modulation bandwidth enhancement based on the PPR effect for the first time. The chip is composed of dual-wavelength high-speed DFB lasers with integrated passive optical feedback sections, a 2 × 1 multimode interference (MMI) optical combiner, and a common optical output waveguide. The DFB laser wavelength definition was achieved through upper separate confinement heterostructure (SCH) selective area growth (SAG) approach [14], which represents a simple and cost-effective approach for fabricating multiwavelength DFB laser arrays compared to the traditional E-beam grating direct writing approach. We successfully demonstrated the PPR enhanced modulation bandwidth >17 GHz and 25 Gb/s Non-Return-Zero (NRZ) optical modulation in the dual-wavelength DFB optical chip, and an aggregate data transmission rate of 50 Gb/s from a single output waveguide. Our device design can be in principle easily scaled to four channel counts or higher and further bandwidth enhancement through device design optimization, thus providing an InP based monolithic photonic integration technology path for next generation optical datacom links beyond 100G.

2. Design and fabrication

The optical microscope photograph of the monolithically integrated device is shown in Fig. 1, with the total dimension of 3200 × 500 μm². The integrated chip consists of two sections: a Passive Feedback Laser (PFL) array section and a 2 × 1 MMI coupler section connecting to a single output waveguide, while the PFL is consisting of an active DFB laser and a passive integrated feedback (IFB) section, the latter with metal contact pads for current injection based phase tuning. The cavity lengths of the DFB and IFB section are 400 µm and 900 µm, respectively, and the spacing between the two channels is 250 µm. A passive 2 × 1 MMI coupler with dimensions of 12-µm in width and 190-µm in length serves as an optical combiner of the dual channel PFL array. Passive S bend waveguides connect the PFL array and the MMI coupler, measuring 1100 µm along the propagation direction.

 

Fig. 1 Optical microscope image of a fabricated dual DFB integrated optical chip.

Download Full Size | PDF

For multiple wavelength DFB laser array, several approaches have been used defining the DFB laser wavelengths, including electron beam lithography [15], sampled grating [16] and MOCVD Selective Area Growth (SAG) technique [17]. While the conventional electron beam lithography adjusts the individual channel DFB laser wavelength by varying the grating pitch physical dimension, in the SAG approach the DFB grating pitch remains unchanged, and the lasing wavelength is modified by changing the effective index of the optical mode through SAG enhanced optical confinement layer growth rate. The SAG process presents a low-cost approach for fabricating multiple wavelength laser arrays due to its simple fabrication process and demonstrated wavelength control precision, and SAG growth of the multi-quantum well (MQW) and the upper cladding layer has been demonstrated in fabricating multiple wavelength DFB laser arrays [17]. However, one challenge of this approach is that the MQW active region material quality is difficult to maintain in the SAG regrowth process. A modified process where the SAG regrowth involves only the upper-SCH layers resolved the MQW material quality issue and resulting in improved DFB array uniformities [14]. This is also the approach adopted in our monolithically integrated device fabrication.

The epitaxial structure of the monolithically integrated device was grown by metal organic chemical vapor deposition (MOCVD) in a three-step regrowth process. In the first MOCVD growth step, an n-type InP buffer, an InGaAsP lower SCH layer composed of 50-nm 1.05 Q (photoluminescence wavelength equals to 1.05µm) and 50-nm 1.1 Q layers, a 20-nm InP etch stop layer, and an InGaAsP MQWs layer were successively grown on n-type InP substrate. The MQWs layer consisted of eight compressively strained 7-nm wells and nine 10-nm barriers lattice matched to InP materials. Next, a Plasma Enhanced Chemical Vapor Deposition (PECVD) grown SiO₂ layer was deposited to form the SAG mask, which the gap between the mask pairs (W_g) is kept constant at 10 µm and the mask width (W_m) ranges from 14.4 to 43.2 µm in 28.8 µm increments. Then, the MQWs layer in the passive waveguides was selectively etched off by a wet chemical etching process. In the second MOCVD growth step, an InGaAsP upper SCH layer composed of a 50-nm 1.1-Q layer and an 80-nm 1.05-Q layer was deposited across both the active region in between the SAG oxide mask and the passive waveguides section. The resulting active/passive interface shown in Fig. 2(a) is very smooth with no visible sharp discontinuities and crystalline defects commonly observed in the direct butt joint approach, with a height difference equal to the thickness of the whole waveguide layers composed of MQWs and SCH layers [17]. This modified butt-joint technique is important for achieving high-quality active-passive interface and minimizing coupling loss and back reflections, which will directly affect the integrated photonic chip performance. After the second MOCVD growth step, the SiO₂ masks were removed by HF wet etching process. A grating with a uniform period was defined only in the DFB section of the integrated chip via a simple holographic lithography process and wet etched 80 nm depth into the upper SCH layer, the rest of the chip masked with photoresist. Figure 2(b) shows the fabricated DFB grating. Finally, in the last MOCVD regrowth step, a p-type InP cladding layer and an InGaAs contact layer were successively grown across the entire wafer completing the full epitaxial structure.

 

Fig. 2 (a) The SEM cross-sectional image of the active/passive interface after the second MOCVD regrowth step. (b) The SEM of the grating with a uniform period only on the DFB section.

Download Full Size | PDF

Afterward, inductively coupled plasma (ICP) etching process was used to define both the shallow ridge structure in the PFL array section and deep ridge structure in the passive waveguide section in two steps. First, a shallow ridge was defined with SiO₂ masks, and a single step ICP dry shallow etching process was used to etch both the active and passive waveguide sections above the InGaAsP MQW active region. Next, a SiO₂ mask was selectively deposited on the active waveguide sections. A second ICP etching process was subsequently applied to deep etch the passive waveguide section. Electrical isolation between the DFB lasers and the passive IFB sections was then finished by etching off InGaAs contact layer and subsequent He ion implantation. Ti/Au p-metal layers were sputtered and patterned, and then after backside lapping and polishing, Au/Ge/Ni metallization layers were evaporated on the backside for n-metalization. After metal contact annealing at 425 °C for 35 seconds, the monolithically integrated device was cleaved into chips, completing the process flow.

3. Device characterization

The fabricated laser chips were mounted onto AlN heatsinks and temperature stabilized at 25 °C for detailed device characterization. The light output versus injection current (L-I) characteristics was measured by using an integrating sphere, while the optical output power from the PFL side output facets and the common MMI output port are shown in Fig. 3(a) and Fig. 3(b), respectively. The threshold currents of the two channels are 72 mA and 36 mA, respectively. Mode hoping and thermal rollover can be observed in the L-I curve. Mode hoping happens due to the influence of the IFB cavity mode. Route from single mode, period one oscillation, period two oscillation into chaos can be observed with proper phase current. For the bandwidth enhancement, it is desirable to make the laser work at the onset of appearance of the second mode, which corresponds to a set of combination of phase current and laser bias current. For the laser reported in the paper, we found that an optimum bandwidth enhancement performance could be obtained when the DFB section is biased at 280 mA despite of the thermal rollover. The output powers of the two DFB channels measured from the PFL facet ends are 0.91 mW and 3.44 mW at a bias current of 280 mA, while the corresponding power measuring from the common MMI output port are 0.33 mW and 0.67 mW, respectively. The power conversion efficiency of both lasers is relatively low because the relatively high internal losses of the structures, mainly affected by the uneven epitaxial material quality and the material defect. There is also some disparity between the L-I properties of the two DFB lasers. The difference in the performance of the two DFB channels could be attributed to different loss levels between the two channels, which was induced by manufacturing errors. The insertion losses of the two channels are estimated to be 4.4 dB and 7.1 dB respectively. While the 2 × 1 MMI combiner accounts for 3 dB of theoretical insertion loss, the rest are excessive loss relating to the fabrication error of the passive waveguides, which can be improved in the future process runs by optimizing the waveguide etching process. Rest of the device characterization was based on measurement of the common waveguide optical output after the MMI combiner.

 

Fig. 3 L-I properties of the fabricated device: (a) from the PFL facets; (b) from the MMI output port.

Download Full Size | PDF

Fig. 4 shows the optical spectrum of the monolithically integrated device when the injection currents of the DFB lasers were set at 280 mA at 20 °C. The lasing wavelengths of the two DFB lasers were 1329.99 nm and 1332.06 nm, respectively. Both DFB channels maintained stable single mode operation with side mode suppression ratio (SMSR) greater than 34 dB. From the zoomed in view of channel 2 optical spectrum, as shown in the Fig. 4 (inset), a weak side mode alongside the main lasing wavelength can be observed, the wavelength difference between these two modes corresponds to the PPR RF resonance frequency, which serves to extend the DFB modulation bandwidth under appropriate optical feedback conditions [18].

 

Fig. 4 The optical spectrum of the dual-wavelength DFB integrated optical chip measured from the output waveguide, inset showing side mode with wavelength separation from the main DFB mode matching the PPR effect resonance peak.

Download Full Size | PDF

The S₂₁ small-signal modulation responses of the dual channel DFB photonic integrated chip were measured from the common output waveguide with a vector network analyzer (VNA, HP 8510C), and are shown in Fig. 5. For the S₂₁ response of each channel both the CPR relaxation oscillation peak and the PPR resonance peak can be clearly observed. In this case, the f_3dB modulation bandwidths of both channels are extended to 17 and 18.3 GHz respectively, under optimal biasing conditions with the DFB laser biased at 280 mA. By comparison, identical discrete DFB test lasers have an f_3dB bandwidth of only ~10 GHz under the same biasing condition. The PPR effect is quite sensitive to the phase of the optical feedback from the end facet of the IFB section, which in our design can be adjusted by varying the injection current to the passive 900-µm IFB sections. The small-signal modulation response of channel 2 under different IFB currents is shown in Fig. 6, with the DFB laser section, biased at 280 mA. For IFB section phase tuning current of 13 mA, a relatively flat modulation response and maximum f_3dB modulation bandwidth of 18.3 GHz is achieved.

 

Fig. 5 Small-signal modulation responses of the dual-wavelength monolithically integrated device with each DFB section biased at 280mA.

Download Full Size | PDF

 

Fig. 6 Small-signal modulated the response of channel 2 of the chip with varying IFB current levels modifying the phase of integrated optical feedback, DFB laser section biasing held constant at 280 mA.

Download Full Size | PDF

Optical eye diagrams of the device are shown in Fig. 7. The measurements were carried out using a ground-signal-ground (GSG) RF probe providing a Non-Return-to-Zero (NRZ) 25 Gbit/s signals with 2⁷-1 pseudo-random bit sequence (PRBS) data streams and DC injection currents to the DFB laser section, combined externally with a broadband bias Tee. 25 Gb/s eyes opening can be observed. However, the eye quality does suffer from poor signal to noise ratio issues, due to the impedance mismatching of the DFB laser that created RF back reflections and artificial ringing in the optical eye, which can be improved by adding a small RF serial resistance to the DFB laser. This can be implemented using a 45-Ohm series matched RF probe [19], which will be carried out in subsequent experiments currently underway.

 

Fig. 7 25Gb/s optical eye diagram of the dual channel integrated photonic chip.

Download Full Size | PDF

4. Summary

A 1.3-µm dual-wavelength photonic integrated chip fabricated in the InGaAsP material system with modulation bandwidth enhancement based on the PPR effect was demonstrated for the first time. The CWDM photonic chip consisted of a pair of DFB laser sections with optical feedbacks provided by integrated passive waveguides and a 2 × 1 MMI coupler, where the phase of the optical feedback can be controlled by current tuning. Dual operating wavelengths with separation of 2 nm were realized by modifying the upper DFB cladding thickness using the SAG-Upper SCH technique, which provided a low-cost alternative for fabricating WDM DFB laser arrays to the E-beam approach. A modified butt-joint technique was also adopted producing a smooth active/passive interface, and minimizing unintentional intra-cavity optical feedbacks. The fabricated photonic chip operated at stable single mode for both channels with SMSR >35 dB. Clear PPR effects were observed, and the 3-dB modulation bandwidths of both channels were extended to greater than 17 GHz. 25 Gbit/s eye diagrams were demonstrated. Our design principle can be easily scaled up to higher channel counts and higher data rate with further design optimizations. Overall we believe our design represents a promising technology path for future cost-effective, ultra-compact and high data rate optical transmitters operating at 100-Gb/s and beyond.