The growing data traffic demands in short-to-medium-range optical interconnects and 200/400 Gb/s Ethernet (GbE) links require new transmitter components to meet rigorous requirements regarding cost, power efficiency, and device size^[1,2]. Wavelength division multiplexing (WDM) has emerged as a viable solution to increase transmission capacity by employing multiple wavelengths to send more signals in one single-mode fiber. To fulfill the requirements of the applications, much research has been done on the improvement of the semiconductor lasers, since they are the core components of the transmitters. Currently, low-dimension active region lasers, such as quantum dot and quantum dash lasers, have attracted a lot of attention as they are a promising technology that focuses on photonic transmitters in O-, C-, and L-bands, in comparison with the quantum well counterpart^[3-9]. Besides the improvement in the active region of the lasers, photonic integration is also important to future transmitting devices. The utilization of integrated high-speed directly modulated laser (DML) arrays presents a promising approach for WDM transmitters due to their cost-effectiveness, compact form factor, and reduced power consumption^[10-12]. In such integrated DML arrays, each channel should ideally support a data rate of no less than 25 Gb/s, operate in O-band, and maintain a channel spacing of 800 GHz, aligning with the local area network-WDM (LAN-WDM) grid^[11]. Besides, a mux component is also needed for the WDM transmitter using a multi-wavelength laser array (MLA) so that all output signals can be combined into one single-mode fiber. Such a combiner component is preferred to be an integrated planar light-wave circuit (PLC), such as the multi-mode interferometer (MMI) mux, arrayed waveguide grating (AWG) mux, or cascaded Y-branch waveguide mux, to keep the overall size small.

To achieve the integration of the MLA and the mux components, several methods have been studied, such as the butt-joint growth technique^[13] and the chip bonding method^[14]. However, due to the complexity of the fabricating procedure, such integration methods still need some improvements in cost and yield. Recently, a new technique named photonic wire bonding (PWB) is drawing attention in optical integration. By employing multi-photon polymerization and 3-D waveguide writing, the PWB method simplifies the intricate aligning process while ensuring good optical coupling^[15-17]. The fabrication of PWB is flexible, fast, and cost-effective, making it a promising technique in heterogeneous integration.

In this Letter, we propose and experimentally demonstrate an 8-channel WDM transmitter, which is composed of a DML array chip and a PLC mux. The laser array and the mux chip are hybrid-integrated by the PWB technique. The DML array serves as the core component of this transmitter, and in the array, the reconstruction equivalent chirp (REC) technique is employed to simplify the grating fabrication and achieve a multi-wavelength DML array with a channel spacing of 800 GHz^[18]. The experimental results show that precise wavelength control and good single-mode operation are achieved in this array. Clear eye diagrams of 25 Gb/s non-return-to-zero (NRZ) modulation are obtained for all channels; thus, in total, it supports a 200 Gb/s data rate. To the best of our knowledge, it is the first reported 200 Gb/s transmitter based on the REC-DML array coupled by PWB technology and is a promising transmitter for LAN-WDM systems.

To achieve a high modulation rate in our DML array, an improvement of the modulation bandwidth is needed. It is found that the electro-optic (EO) response bandwidth is mainly limited by the relaxation oscillation frequency (fr)^[19,20]. Thus, the fr should be high enough for the required modulation rate, and it can be expressed in the following relation^[21]: fr∝Γdg/dNLWNwLw(I−Ith),(1)where Γ is the optical confinement factor, dg/dN is the differential gain, Ith is the threshold current, L is the active region length, W is the active region width, Lw is the well thickness, and Nw is the number of wells. From Eq. (1), it can be found that reducing the active-region volume by shortening the cavity length is an effective method to improve the fr. However, in DFB lasers, a shorter laser cavity means weaker optical feedback of the grating, which results in a high threshold current and, therefore, deteriorates the direct modulation bandwidth. Thus, a high reflection and anti-reflection (HR/AR) coating scheme is used for the DFB lasers to compensate for the optical feedback and lower the threshold current. However, the random phase shift from the HR coating facet will deteriorate the single-mode properties and increase the uncertainty of the lasing wavelength. Therefore, an asymmetric phase shift (APS) structure is proposed to improve the single-mode yield of the laser array. In this APS design, the π phase shift introduced in the laser cavity is placed near the HR coating facet. To study its effect on single-longitudinal-mode (SLM) operation, the normalized threshold gain margin (NTGM) based on the transfer-matrix method is employed as a criterion. NTGM is characterized as the net normalized gain difference between the dominant mode and the most probable side mode. When the NTGM exceeds the threshold of 0.25, SLM operation is ensured^[22]. In the calculation, the uncertain phase from the HR-coated facet is evenly taking 100 values ranging from 0 to 2π, and for each value, the NTGM is calculated. Then the NTGM values exceeding 0.25 are counted and normalized to percentage, to find the optimized π phase shift position. The calculation result is shown in Fig. 1. From the figure, it is found that, when the π phase shift is located at 4/5 cavity length away from the AR-coating facet, the highest single-mode yield can be obtained for the proposed HR-AR coated DFB lasers. The theoretical single-mode yield of the HR-AR-coated DFB laser with the optimized APS structure is approximately 90%, considering that it is only affected by the uncertain phase from the HR-coated facet. Thus, the theoretical single-mode yield of the 8-DFB laser array is 43.1%.

The REC technique is used to simplify the grating fabrication and obtain a multi-wavelength array with precise wavelength spacing. This technique is described in our previous work^[23,24]. It utilizes the predesigned sampling grating to introduce the needed equivalent phase shifts. Instead of expensive and time-consuming electron beam lithography (EBL), the REC technique only requires a common holographic exposure and micrometer-level photolithography in grating fabrication to achieve certain resonant responses. Usually, the +1st-order subgrating is used to select the lasing mode. The Bragg wavelength of the +1st-order subgrating can be expressed as follows: 1λ+1=12n(1Λ0+1P),(2)where λ+1 is the Bragg wavelength of the +1st order subgrating, n is the effective index, Λ0 is the period of the uniform grating, and P is the period of the sampling grating.

The sampling grating with uniform grating is expressed as Δn(z)=12s(z)Δn0exp(j2πΛ0)+c.c.(3)

When π phase shift is located at z0 in a sampling grating, the expression of the sampling grating-based REC is Δns(z)={12Δn∑mFmexp(j2mπzP+j2πzΛ0)+c.c.z≤z012Δn∑mFmexp(−j2mπΔPP+j2πzΛ0+2mπzP)+c.c.z>z0,(4)where m denotes the mth-order Fourier series and Fm is the Fourier coefficient.

The π phase shift is introduced in the sampling structure with 2mπΔP/P according to Eq. (4). When the period of a uniform grating is determined, the lasing wavelength is solely dependent on the sampling period; thus, the grating fabrication tolerance is relaxed by the factor (P/Λ0+1)2, with the value of several hundred. The schematic of the DFB laser array and grating is shown in Fig. 2.

The lasers in the array are common ridged waveguide structures. The active layer utilizes a compressively strained InAlGaAs MQW structure due to its good high-temperature performance. The grating layer is p-InGaAsP with a thickness of 60 nm and a photoluminescence (PL) wavelength of 1.36 µm. The microscope photo of the laser array chip is shown in Fig. 3(a). The size of one laser unit is 250μm×250μm. The cross-section scanning electron microscope (SEM) image of the laser is shown in Fig. 3(b). The SEM image of the sampling grating is shown in Fig. 3(c). The front facet is AR-coated with a reflectivity below 1%, and the rear facet is HR-coated with a reflectivity above 97%. The coupling coefficient of the uniform grating is approximately 93cm−1 and 30cm−1 after sampling^[24]. Considering the cavity length of the laser is 250 µm, the κL of the laser is 0.75. The period of the uniform grating is 215.2 nm. The sampling grating periods of the eight lasers are designed as 3.54, 3.82, 4.11, 4.48, 4.93, 5.40, 6.06, and 6.87 µm. As a result, the lasing wavelengths of the four lasers are 1309.1, 1313.9, 1318.2, 1322.9, 1327.6, 1331.8, 1336.5, and 1341.2 nm.

The DML array chip is bonded on an aluminum nitride (AlN) base with the ground coplanar waveguide (GCPW) leads, which is shown in Fig. 4. The GCPW electrode serves both to transmit the RF signal efficiently and mitigate the crosstalk between channels. To ensure impedance matching, 40 Ω tantalum nitride (TaN) resistors were fabricated near the lasers. The impedances of the transmission line are usually designed at 50 Ω, and the resistance of the DFB laser is approximately 10 Ω. Thus a 40 Ω resistance is designed in series with the DFB laser to achieve a total resistance of 50 Ω at the load end, ensuring impedance matching with the 50 Ω transmission lines.

Figure 5 shows the schematic and a top view photo of the PWB integrated sample. In PWB integration, an important consideration is the alignment of the waveguides of different chips. So a specially designed WuCu submount is used to compensate for the height differences between the laser array chip and the PLC chip.

To optimize the loss of PWB, the mode field diameter (MFD) of the DML is simulated by the finite difference time domain method, as shown in Fig. 6. Overlap factor η is used to characterize the overlap between the two field profiles (modes). In general, an ideal interconnection between different photonic components is given a maximized overlap. The material for the PWB waveguide is ultraviolet (UV) photoresist, with a refractive index of approximately 1.5 at 1310 nm. The overlap η is typically represented as η=Re[(∫E1×H2*·ds)(∫E2×H1*·ds)∫E1×H1*·ds]/Re(∫E2×H2*·ds).(5)

According to Eq. (5), the overlap η between the MFDs of PWB waveguides and lasers with varying widths and heights is calculated. As shown in Fig. 7, good mode matching can be achieved with a PWB cross-section size of 4.5μm×1.5μm at the laser end. At the PLC end, the cross-section size of the PWB is 12μm×10μm to match that of the PLC waveguide, whose MFD is very close to a single-mode fiber.

Besides the simulation on mode overlapping, an additional experimental optimization is performed to get a low-loss PWB link. Due to the difficulty of directly characterizing the depth in the Z-direction (along the ridge waveguide to the quantum well) in practical fabrication, the optimum Z-coordinate value is obtained experimentally. This experimental optimization, which is called Z-scan, uses a test sample made of two components. One is a laser array or PLC chip, and the other is a fiber array. In such test samples, different Z values are used for the fabrication of each PWB in the array, and the loss measurement is performed to determine the Z value with the lowest loss. The Z-scan results are shown in Fig. 8. From these tests, it is found that the optimized Z-offset values are −0.4μm and −0.5μm for the laser waveguide and the PLC waveguide, respectively.

A 3-D plot of the PWB wire is shown in Fig. 9. In this plot, Sections 1 and 3 are two taper structures, which are working as spot converters to ensure mode matching for both laser and PLC waveguides. Section 2 is a linking section to guide the light from one taper to another. At the linking section, the designed cross-section size is 1.2μm×0.8μm. Figure 10(a) shows the microscope photos of the PWB polymer waveguide. The SEM images of the PWB waveguides and single PWB waveguide are shown in Figs. 10 (b) and 10(c), respectively. The wires are considered to have a good coincidence with the 3-D plot, and no obvious flaw is seen.

The spectral diagrams of all channels are measured. In the measurement, the sample is placed on a temperature-controlled stage whose temperature is kept at 25°C. The lasers are driven by a DC power source (Rigol DP832, with a high-resolution option), and an optical spectrum analyzer (Yokogawa, AQ6370) is used to measure the lasing spectra. The optical spectra of the DML array are measured at 100 mA, as shown in Fig. 11(a). Good single-mode operations are obtained, with all SMSRs above 50 dB, and the channel spacing is very uniform. Figure 11(b) shows the power-voltage-current characteristics of the DML array before integration. The thresholds of all lasers are around 18 mA, and when the injected current reaches 120 mA, the power outputs of all channels exceed 25 mW.

Figure 12 shows the measured power-current curves through the single-mode fiber after PWB integration. The output powers are above 1 mW for all channels at 100 mA. The total loss is about 12–13 dB obtained from Figs. 11 and 12. The intrinsic loss of the pigtailed PLC chip (fabricated in Henan Shijia Photonics Technology Co., Ltd.) is about 10 dB, which is provided by the producer; thus, the insertion loss introduced by the PWB is about 2–3 dB.

The EO response of the hybrid-integrated transmitter is measured by using a 70 GHz vector network analyzer (Anritsu MS4647A). The typical EO response of a unit laser is shown in Fig. 13. The range of injected current variations from 40 to 100 mA results in an expansion of the 3 dB EO response bandwidth from 15.8 to 21.9 GHz. The enhancement is attributed to the high relaxation oscillation frequency at higher injected currents. The achieved 3 dB bandwidth exceeding 20 GHz is deemed sufficient for supporting 25 Gb/s NRZ large-signal modulation.

The 25 Gb/s NRZ large-signal modulation experiment using a pseudo-random bit sequence (PRBS) of 231−1 is performed. The PRBS signal, generated by a 100G bidirectional encoder representation from transformers (BERT, Golight), is amplified by a 38 GHz electrical amplifier (SHF L810A). The peak-to-peak current supplied to the DFB laser is about 60 mA. Optical eye diagrams are measured and analyzed by an optical sampling oscilloscope (Keysight N1092A). In Fig. 14, the back-to-back optical eye diagrams are present for all 8 channels. The bias currents injected into the DML array are 60 mA. The dynamic extinction ratios for all channels are above 5 dB. The average output power ranges from 0.7 and 1.1 mW at the bias current of 60 mA for all 8 channels. To assess the quality of these eye diagrams, the 25GBASE-LR_ER_TX mask is embedded. Notably, all eye diagrams exhibit mask margins exceeding 20%, which implies a good transmission performance.

In this Letter, a 200 Gb/s hybrid-integrated WDM transmitter is reported. In the DML array design, the laser cavity is shortened to improve the 3 dB EO response. An HR/AR coating scheme is utilized to provide enough feedback for the lasers with short cavities. An APS structure is utilized to compensate for the random phase shift from the HR coating facet and keeps every channel of the array single-mode working. The REC technique is employed to simplify the grating fabrication and achieve a multi-wavelength DML array with a channel spacing of 800 GHz. The DML array is hybrid-integrated with a PLC combiner by PWB, whose insertion loss is measured to be about 2–3 dB for all channels. The output power of each channel of the integrated transmitter is above 1 mW at 100 mA, and the measured SMSRs are above 50 dB. The 3 dB bandwidth is beyond 20 GHz at the bias current of 60 mA, and clear eye diagrams of 25 Gb/s non-return-to-zero (NRZ) modulation are obtained for all 8 channels.