Heterogeneously Integrated Membrane III-V Compound Semiconductor Devices With Silicon Photonics Platform

Silicon photonics is a key technology for constructing large-scale photonic integrated circuits (PICs) because it enables large-scale wafer processes with high uniformity and quality. To further improve device characteristics, heterogeneous integration of III-V compound semiconductors that provide optical gain, a high modulation efficiency, and optical non-linearity is desired. This paper describes the heterogeneous integration of membrane III-V compound semiconductor photonic devices that have a similar structure including thickness and refractive index. These devices provide efficient optical coupling with a Si waveguide using a simple taper waveguide structure. If the total thickness of the film structure is designed to be less than the critical thickness (calculated to be 430 nm for fabrication conditions such as bonding and growth temperatures), high-quality epitaxial layers can be grown on a thin InP layer directly bonded to the Si substrate. Therefore, regrowth techniques are employed on bonded InP layer on SiO2/Si substrate. We fabricate two kinds of laser-integrated Mach-Zehnder modulators using epitaxial regrowth on Si substrates. One uses Si phase modulators, and the other uses InP-based modulators. A micro-transfer-printing technology is also important when the number of III-V devices is relatively small. Furthermore, the micro-transfer-printing technology enables devices to be selected that meet the required characteristics before integration. For this purpose, we try to integrate a membrane laser on a Si substrate, in which the membrane laser is fabricated on InP substrate. The device shows a threshold current of 0.8 mA when the active region length is 140 μm. Finally, we briefly describe a transmission module, in which directly modulated membrane lasers and electronic drivers are integrated by flip-chip bonding through Au bumps. To reduce power consumption, it is important to design driver circuits that incorporate semiconductor lasers as electronic components. We demonstrate a 2-channel 53-Gbit/s 4-level pulse amplitude modulation (PAM4) transmitter front-end consisting of a 2-channel PAM4 shunt laser driver and 2-channel O-band directly modulated membrane lasers. The total power consumption is only 60.7 mW, resulting in 0.57 mW/Gbit/s.


I. INTRODUCTION
P HOTONIC integrated circuits (PICs) are becoming increasingly important because they represent a promising solution for providing high-throughput transmitters at a low cost [1], [2], [3], [4]. Silicon (Si) photonics technology is promising because it can be used with mature Si CMOS technology that provides nm-scale fabrication with high uniformity on a largescale wafer [6], [7], [8], [9], [10]. Therefore, Si photonics devices enable low-loss optical circuits including high-performance optical filters; however, fabricating the laser and high-efficiency modulators is difficult because Si is an indirect bandgap material. Thus, heterogeneous integration with InP-based compound semiconductor devices has attracted much attention.
Direct growth of InP-based materials is the most preferable solution [11], [12]; however, the quality of the epitaxial growth layer is poor due to the lattice constant mismatch and thermal expansion coefficient mismatch between Si and InP. Recently, quantum dot lasers overcome these problems and successfully demonstrated high-output power and high-temperature operations [13], [14]. However, the optical coupling between InP-based devices and Si waveguides is also an issue because hetero-epitaxial growth requires a thick buffer layer.
Wafer-scale fabrication using directly bonded InP dies or wafers is promising because it can use high-quality epitaxial layers grown on InP substrates [15], [16], [17], [18]. InP devices can be aligned with Si waveguides accurately by photolithography. Therefore, many research groups have demonstrated heterogeneous integration and heterogeneously integrated lasers that exhibit a high output power, which is suitable to use as a bias light for a Si-based modulator.
Chip-scale integration techniques, such as flip-chip bonding [19] and micro-transfer printing [20], [21], [22], are another important technology for integrating InP devices with Si photonics circuits. Chip-scale integration allows us to select devices after characterizing the device performance. However, there are challenges in fabricating PICs with low cost and high density because precise positioning that limits the fabrication throughput is required.
In this paper, we first describe the structure and features of the membrane III-V photonic devices. Next, a wafer-scale fabrication procedure is described that uses a combination of direct bonding and epitaxial growth for the InP-based layers. We demonstrate the heterogeneous integration of lasers with Si-based or InP-based modulators. For chip-scale integration, we use a micro-transfer printing technique to integrate a laser on a SiO 2 /Si substrate. We successfully transfer the laser on an InP substrate to a Si substrate without degrading the device performance. Integration with electrical integrated circuits (ICs) is also important to make compact transmitter modules with low power consumption [36], [37], [38]. We design an electrical IC to optimize the directly modulated membrane laser. A membrane laser array is integrated with an electrical IC by using the flipchip bonding technique, and it obtains 56-Gbit/s pulse amplitude modulation 4 (PAM4) operations with 0.57 mW/Gbit/s [39].

II. STRUCTURE OF MEMBRANE PHOTONIC DEVICES
Cross-sectional views of membrane photonic devices without and with Si waveguides are shown in Fig. 1(a) and (b). The InP-based core region is buried in the InP layer, whose thickness is 250-350 nm. This buried heterostructure (BH) is important to confine the carriers and photons because InP has a larger bandgap and smaller refractive index than those of the core region. We use a lateral p-i-n structure because a vertical p-i-n structure makes it difficult to get this thickness range. Since the device is put on a SiO 2 layer, a large optical confinement factor is achieved in the case of no Si waveguide, as shown in Fig. 1(a), which results in a device with a high modulation efficiency. Optical confinement factor of III-V core and Si waveguide. For this calculation, thicknesses of InP and Si were assumed to be 250 and 220 nm, respectively, and SiO 2 thickness between InP and Si was 100 nm. III-V core was 0.6-µm wide and 150-nm thick. Fig. 2 shows the optical confinement factor of the III-V core and Si waveguide, respectively, when the Si waveguide width underneath the membrane device is changed. For this calculation, we assumed that the thicknesses of the InP and Si were 250 and 220 nm, respectively, and the SiO 2 thickness between the InP and Si was 100 nm. The III-V core was 0.6-µm wide and 150-nm thick. When the Si width was less than 0.2 µm, the optical mode was mainly confined in the III-V core region with an optical confinement factor of 0.61. Then, the optical confinement in the III-V core gradually decreased as the Si width increased. The optical confinement factor was 0.21 and 0.63 for the III-V core and Si, respectively, when the Si width was 1.0 µm. As is shown, the optical confinement factor is easy to control by changing the Si width because the effective refractive index of the membrane device is close to the Si waveguide. Fig. 3 shows calculated mode fields and a schematic diagram of the calculated structure. The Si width was set to 0.45 µm. In the laser section, the supermode was generated as shown in Fig. 3(a). For constructing the laser output waveguide, we used an InP taper waveguide, as shown in this figure. As shown in Fig. 3(b)-(d), the optical mode field was gradually transferred to the Si waveguide by decreasing the InP width. Fig. 4 shows the calculated coupling loss between the laser and Si waveguide as a function of InP taper length. The InP taper waveguide width was changed from 1.5 to 0.2 µm. As is shown, low loss coupling was obtained by using a simple taper waveguide structure with a short taper waveguide length. As shown in Figs. 3 and 4, we can easily transfer the optical mode field from the III-V core to Si with a simple taper waveguide.

III. WAFER-SCALE FABRICATION
In this section, we describe wafer-scale fabrication by using direct wafer bonding and epitaxial regrowth on a bonded thin  InP layer [23], [24]. The static and dynamic characteristics of the fabricated devices are also described.

A. Fabrication Procedure
The most important advantage of membrane devices in terms of the fabrication process is that we can grow InP-related materials on a directly bonded InP template on a silicon substrate. Thus, a regrowth process that is widely used for fabricating InP-based devices on InP substrate can be used. The fabrication procedure for BH using direct bonding and regrowth is shown in Fig. 5. First, a multiple-quantum-well (MQW) active layer and InGaAs sacrificial layer are grown on an InP substrate, which is directly bonded to a SiO 2 /Si substrate, as shown in Fig. 5(a). After removing the InP substrate and InGaAs sacrificial layer, an MQW layer sandwiched between InP layers is left on the SiO 2 /Si substrate ( Fig. 5(b)). After defining the core region by using a SiO 2 mask, the MQW layer is removed except for the core region. Finally, the core region is buried in the undoped InP layer, as shown in Fig. 5(c).
Thus, the proposed fabrication procedure does not require the epitaxial regrowth under lattice mismatch conditions. However, to obtain high-quality III-V layers, we have to overcome the thermal-induced stress that generates the difference in thermal expansion coefficients between Si and InP. For this purpose, the reduction in total III-V thickness is key. We calculated the critical thickness to be 430 nm when we assumed that the bonding and growth temperatures were 200°C and 610°C, respectively, and the thermal expansion coefficients of Si and InP were 2.62 ppm/K and 4.56 ppm/K, respectively [24]. Therefore, the membrane photonic device is suitable to fabricate on Si, but it is difficult to use the proposed fabrication procedure for a conventional III-V photonic device structure that uses a vertical p-i-n junction. This is because the vertical p-i-n junction is typically ∼3-µm thick.
After fabricating BH, selective doping is performed by using Si ion implantation for n-type doping and Zn thermal diffusion for p-type doping. When the laser is fabricated, a surface grating is formed. Then, we fabricate the electrodes. The details are given in 24.

B. Heterogeneous Integration With Si-Based Mach-Zehnder Modulator
We demonstrate the heterogeneous integration of a distributed feedback (DFB) laser and Si-based Mach-Zehnder modulator (MZM) [29]. Fig. 6 shows schematics of the top and crosssectional views of a device consisting of a membrane distributed feedback (DFB) laser and a Si MZM. As shown in the crosssection of the DFB laser, a Si waveguide was underneath the BH region, which creates an optical supermode. This supermode reduces the optical overlap with the MQW region, which suppress the spatial hole-burning effect, and the p-InP layer that has a large optical loss, so a single mode lasing with a high output power can be obtained [28]. The light emitted from the DFB laser was coupled to the Si waveguide through a 50-µm-long InP inverse-taper waveguide that was long enough to achieve good optical coupling with the Si waveguide as shown in Fig. 4. The Si waveguide was connected to a 4-mm-long Si-phase modulator, consisting of multimode interferometers (MMIs) and carrier depletion-type Si phase shifters.
The Si-MZM was fabricated in a Si photonics foundry using an 8-inch Si-on-insulator (SOI) wafer. The thickness of the Si layer was 220 nm. We used carrier-depletion-type Si phase modulators. After performing p-and n-type doping and etching, a SiO 2 film was deposited on a Si photonics wafer and planarized by chemical mechanical polishing (CMP) for direct bonding of the InP wafer. Then, the wafer was cut into 2-inch wafers for our in-house process, and we fabricated DFB lasers by using our proposed fabrication procedure including buried regrowth on the Si substrate. The active region length was set to 500 µm, and a SiN grating with a λ/4 phase shift was formed on the top surface of the InP layer. Finally, we deposited the electrodes for the Si and InP devices.
Microscope images of the fabricated device including a closeup of the laser and InP taper waveguide are shown in Fig. 7, where the Si Phase modulators are 4-mm long, the DFB lasers are 500-µm long and the InP taper waveguide is 50-µm long. As shown in these images, we successfully demonstrated heterogeneous integration with precise positioning since we used markers on the Si substrate to fabricate the laser section.
We measured the characteristics of the integrated laser through the monitor port. Fig. 8(a) shows a measured optical spectrum with a bias current of 49 mA at a stage temperature of 25°C. The optical spectrum was measured through a lensed optical fiber. Single mode lasing at a wavelength of 1.55 µm with a side-mode suppression ratio (SMSR) of 42 dB was obtained. Fig. 8(b) shows the maximum output power and threshold current as a function of the stage temperature from 25 to 80°C. In this experiment, the optical power was measured with a large-area photodiode placed in front of the monitor port. The threshold current was 7 mA, and the maximum output power was 4.3 mW at room temperature. We achieved lasing at 80°C, and a maximum output power of 1.5 mW was obtained. Since a membrane laser integrated on SiO 2 has a poor thermal conductivity and the electric resistance per unit length is larger than that of the conventional laser structure, it is not suitable for use at large current densities that generate large amounts of heat. However, thanks to a large optical confinement factor, efficient laser oscillation is obtained in the region where the injection current density is small. Thus, by increasing the active region length, a higher output power is expected [28].
Next, we measured the output power dependence of the reverse bias voltage applied to the Si phase shifter, and the modulation efficiency was estimated to be a V π L of 2.4 Vcm, which is almost the same as the normal Si MZM performance. Then, we measured the dynamic characteristics of the integrated device. In this experiment, the laser was biased around 45 mA. Electrical signals were input to a one-side phase shifter from a pulse pattern generator (PPG) through a linear amplifier, in which pre-emphasis was applied to the pulse patterns. Fig. 9 shows eye diagrams with 32-Gbit/s non-return-to-zero (NRZ) signals with a pseudo-random binary sequence (PRBS) 2 31 -1 under temperatures of 25, 50, and 80°C. Clear eye-openings with extinction ratios over 5 dB were observed. These results were obtained thanks to the small temperature dependence of the Si MZM. Therefore, we have successfully an integrated membrane DFB laser with a Si-based MZM, and there seemed to be no degradation in the Si layer due to epitaxial regrowth process.

C. DFB Laser With InP-Based Mach-Zehnder Modulator
A MZM using InP-based membrane phase modulators is expected to improve the modulation efficiency. However, it requires additional III-V bandgap material in addition to the laser active region. Thus, we used epitaxial regrowth on an InP template on a Si substrate to fabricate different bandgap core regions. Schematic diagrams of a device are shown in Fig. 10,    where DFB lasers and InP-based phase modulators were connected by Si waveguides including a Si MMI [30], [31]. As shown in the cross-sections, the phase modulator had no Si waveguide to increase the optical confinement factor. The DFB laser had a Si waveguide underneath the BH region as described before. In this demonstration, a 600-nm-wide MQW core was optically coupled to an 840-nm-wide and 220-nm-thick Si waveguide core. To obtain single-mode lasing, we used a combination of a SiN-based uniform surface grating and Si waveguide width modulation, in which the Si waveguide width was increased to 920 nm at the center of the DFB laser [32]. Fig. 11 shows the procedure of epitaxial regrowth. (a) To use direct bonding, the fabricated Si waveguides are covered with a SiO 2 layer, and the top surface of the SiO 2 is made flat by using chemical mechanical polishing (CMP). (b) An InP substrate with a MQW layer for the laser active region and a sacrificial layer are bonded to the Si substrate, and then the InP substrate and  sacrificial layer are removed. (c) The MQW layer is removed except for the area including the laser active region. (d) The core layer for the phase shifter grows on the InP template. In this experiment, we used an n-type doped InGaAsP bulk layer. (e) The MQW and bulk layers are removed except for their core regions. The core regions for the laser and phase shifter are defined by using markers on the Si substrate. As a result, the position of the Si waveguides and III-V core regions are precisely aligned within the accuracy of the stepper. (f) Finally, the undoped InP layer is regrown to bury the core regions.
A microscope image and cross-sectional SEM images are shown in Fig. 12. As shown in the cross-sectional SEM image of the laser section, the BH and Si waveguide were precisely aligned. The phase shifter used a bulk n-doped InGaAsP layer, whose PL wavelength was 1.3 µm. The large carrier plasma effect, band filling effect, and Franz-Keldysh effect contributed to improving the modulation efficiency compared with the Si phase shifter. In addition, without coupling to the Si core, a high optical confinement factor was obtained. We designed the lengths of the laser and phase shifter to be 500 µm. Fig. 13(a) shows a lasing spectrum at an LD current of 50 mA. We obtained single-mode lasing at a wavelength of around 1550 nm. The side-mode suppression ratio was 55 dB. To check the modulation efficiency of the MZM, we measured the halfwave voltage of a stand-alone MZM with various input wavelengths. This stand-alone device was fabricated on the same chip. The V π L was around 0.4 Vcm, as shown in Fig. 13(b). A high-efficiency MZM was successfully fabricated by using the on-silicon regrown high-quality bulk InGaAsP layer. An eye diagram with a 50-Gbit/s NRZ signal is shown in Fig. 13(c). We used a lumped electrode with 50-Ω termination. As shown, we have successfully demonstrated NRZ signal modulation.

IV. CHIP-SCALE FABRICATION
Chip-scale integration is another important technology for heterogeneous integration because devices can be selected before integration, and a variety of different functional devices and materials can be used. However, it requires precise alignment with the Si waveguide, which causes the problem of throughput and cost. As described before in reference to Figs. 3 and 4, since the membrane III-V device is easy to connect with a Si waveguide with a simple structure, it is suitable for use in chip-scale fabrication. In this section, we describe a micro-transfer-printing method for integrating membrane lasers on Si substrates.

B. Transfer Printed Membrane Lasers
Here, we show our first demonstration of the transfer printing method for membrane lasers and, therefore, we did not integrate the laser with a Si output waveguide. A microscope image of the fabricated device is shown in Fig. 14(e), in which the coupon was put on the edge of the target substrate to obtain optical coupling to a single mode fiber. The laser coupon was 160 × 700 µm 2 and the coupons are bonded to the substrate by van der Waals forces [20]. We designed a distributed reflector (DR) laser consisting of a 140-µm-long distributed feedback (DFB) laser section and a 60-µm-long distributed Bragg reflector (DBR) section. The coupling coefficients of the grating were 550 cm −1 for the DFB section and 250 cm −1 for the DBR section. An output waveguide was fabricated using an InP ridge waveguide structure. Fig. 15(a) shows I-L-V characteristics for room temperature continuous-wave operation. The threshold current was 0.8 mA. Due to the reflection at the coupon facet, mode hopping occurred when the bias current was larger than 5 mA because there was no AR coating at the facet. Fig. 15(b) shows a lasing spectrum with a bias current of 5 mA. Under this condition, a longer sideband for the DFB laser section with uniform grating was successfully selected for the DBR section. As shown, interference between the coupon facet and the laser was observed.
We have successfully demonstrated a transfer printing method for the membrane laser. Integration with a Si waveguide is already demonstrated, and results are described in [34].

V. PHOTONIC AND ELECTRONIC CO-INTEGRATION
Co-integration of photonics and electronic chips is important for fabricating high-throughput and low-power transmitters [36], [37], [38]. For this purpose, flip-chip bonding is a promising technology. In addition, the flip-chip bonding interconnection technique minimizes the parasitic inductance between the photonic device and driver IC. Therefore, it suppresses the degradation in the group delay characteristic due to bonding interconnections. Furthermore, it is important to design a driver IC including a laser as an electronic component.
In this section, we describe a 2-channel 4-level pulse amplitude modulation (PAM4) transmitter front-end consisting of a 2-channel PAM4 shunt laser driver and 2-channel O-band directly modulated membrane lasers, where an electronic chip and a phonic chip are integrated by using the flip-chip bonding technique [39]. Fig. 16(a) shows a block diagram of the 2channel PAM4 transmitter front-end. To decrease the power consumption, the front-end uses a shunt laser driver incorporating a digital-to-analog converter (DAC). The driver also has equalizing functions to achieve clear eye-openings for the laser outputs. Since the shunt driver is placed in parallel with the lasers without impedance matching, this sort of front-end is a low-power architecture. Therefore, flip-chip-bonding interconnections are important to minimize the multiple reflections due to impedance mismatch. Fig. 16(b) and (c) show simulated results for the simultaneous operation of 53-Gbit/s PAM4 signals without and with equalizing parts, respectively. We configured equivalent circuit models of the lasers and rate equations in our HSPICE environment to simulate the electric-optic conversion behavior [8]. One equalizing part consisting of a RC filter suppresses the overshooting of the PAM4 optical waveforms, and the other equalizing part, consisting of a capacitor, decreases the fall time of the PAM4 optical waveforms. As a result, we can achieve clear eye-openings when the lasers operate simultaneously.
Microphotograph images of the flip-chip bonded 2-ch PAM4 transmitter front-end, 2-channel membrane laser array, and 2channel PAM4 driver IC are shown in Fig. 17. The ground pads of the 2-channel driver IC were connected to the ground terminal through GND posts by wire bonding. LSB1 and MSB1 were input terminals for channel 1; LSB2 and MSB2 were those for channel 2. The driver IC chip was fabricated in 65-nm CMOS technology. Dummy pads, which the dashed line outlines, were used to prevent the 2-channel laser array from leaning when the laser array was flip-chip bonded to the driver IC. The face-down laser array was stably connected to the driver IC by flip-chip bonding through Au bumps.    The total power consumption of the 2-channel PAM4 transmitter front-end, including the 2-channel PAM4 driver and laser array, was only 60.7 mW, resulting in 0.57 mW/Gbps.

VI. CONCLUSION
We demonstrated heterogeneous integration of membrane photonic devices on the Si photonics platform by using waferand chip-scale fabrication. The thin membrane structure enables us to grow III-V compound semiconductors on a thin directly bonded InP template on Si substrates. Therefore, we used epitaxial regrowth to fabricate BH and other bandgap InP-based materials. As a result, we demonstrated the heterogeneous integration of membrane lasers with Si-based or InGaAsP-based MZM with wafer-scale fabrication. We fabricated the lasers and integrated them with MZMs using both Si-and InGaAsP-based phase modulators. Chip-scale fabrication using micro-transfer printing was also successfully demonstrated, showing the same device performance as wafer-scale fabrication devices. We described the importance of designing the electric IC including the laser as an electrical component, which results in increasing the modulation speed with low power consumption. Flip-chip bonding is important to make transmitters compact and to suppress unwanted electrical reflections between electronics and photonics chips. In 2014, he joined NTT Photonics Laboratories, Atsugi, Japan. His research interests include heterogeneously integrated III-V semiconductor photodiodes, modulators, and laser diodes on Si photonics circuits.
He is currently a Member of the Institute of Electronics, Information, and Communication Engineers, and the Japan Society of Applied Physics. He is a Member of IEICE, JSAP, and the Physical Society of Japan.