1. INTRODUCTION

Hybrid silicon integrated optics, also known as heterogeneous silicon integration, has gotten great attention since its invention a decade ago [1,2]. It has progressed tremendously in both advanced research and commercial optical communication systems with its core capability as a high-performance on-chip laser source on silicon [2–4]. The hybrid silicon integration platform combines traditional III–V photonic integration with silicon photonic integration technologies. It inherits high index-contrast material and device libraries from silicon photonics and the advanced complementary metal–oxide–semiconductor (CMOS) process [5,6], while it is open to many materials, including a number of III–V compounds and other electro- and magneto-optic materials. These materials enable complementary functionalities on silicon, including but not limited to efficient light emission and amplification [7–12]. A selective die-to-wafer bonding is normally preferred to transfer pricey III–V materials to the silicon-on-insulator (SOI) substrate as a cost-effective way [13].

For the high-performance computing (HPC) we are particularly interested in, energy-efficient data connectivity within tens of meters reach is required to handle dynamic bandwidth scenarios including node-to-switch hub traffic [tens to a few hundred gigabytes per second (Gb/s)] and switch hub to switch hub traffic [terabytes per second (Tb/s) level]. For node-to-switch application, we have reported hybrid quantum-well (QW) microring lasers on silicon with low threshold current and high operation temperature; microring lasers integrated with a novel hybrid metal–oxide–semiconductor (MOS) tuner for ultra-efficient wavelength tuning and high-speed direct modulation [14–17]. On the same platform, we also reported a fully integrated $5\times 12.5\mathrm{Gb}/\mathrm{s}$ wavelength division multiplexing (WDM) transmitter based on hybrid QW microring lasers [17]. Similar compact hybrid micro-disk lasers with even smaller dimensions and impressive thresholds have also been demonstrated by benzocyclobutene (BCB) bonding techniques as well [18–20]. More recently, we further expanded the device portfolio to develop a high-performance hybrid comb laser based on InAs/GaAs quantum-dot (QD) gain materials [9,21] for dense WDM optical links in switch hub to switch hub applications [22,23].

It is well-known that QD materials have low transparency current densities that favor low-power operation [24]. Its optical gain has excellent thermal stability and wide spectral bandwidth, which are helpful for high-temperature operation [25,26] and for WDM or widely tunable laser applications [24], respectively. Lasers made using InAs/GaAs QD material have shown low linewidth enhancement factors and are thus less sensitive to external optical feedback [27]. This is critical for stable on-chip laser operation, as an optical isolator is generally difficult to realize on-chip. More importantly, compared with QW, the high-quality QD material with dense dot distribution is more tolerant to materials and fabrication imperfections [28,29], and high-performance and reliable monolithic QD lasers have been demonstrated on silicon recently [30–32]. A large wafer-scale heteroepitaxy growth of QD material on silicon may significantly reduce the unit price of the gain material and, therefore, the total cost of photonic transceivers. However, there are two remaining key challenges before heteroepitaxy becomes commercially viable. First, the threading dislocation density needs to be reduced in order to realize a high reliability for commercial applications. Second, the buffer layer between the silicon and QD active region needs to be reduced in order to achieve efficient optical coupling between the QD active region and the silicon photonic circuit. Both challenges conveniently do not exist in the hybrid integration approach, making it a more practical solution to realize superior QD-containing devices on silicon photonics. This has allowed hybrid QD lasers to be demonstrated by other research groups as well [22,23].

In this paper, we report the first low-threshold and high-speed QD microring lasers that were hybrid-integrated on silicon. Unlike straight-cavity hybrid QD lasers [9,21–23], the microring (with a diameter of 50 μm in this work) is designed in such a way that both the laser cavity and silicon hybrid mode convertors are defined in the same step as the ring–bus waveguide (WG) coupling. In this paper, the design, fabrication, and testing results will be discussed. We report, for the first time, to the best of our knowledge, direct modulation of hybrid QD microring lasers with error-free communication up to 15 Gb/s data rate.

2. DEVICE DESIGN AND FABRICATION

As schematically shown in Fig. 1, we designed hybrid silicon ring lasers with QD gain material that contain eight layers of InAs/GaAs QDs with the photoluminescence wavelength around 1300 nm. The hybrid microring consists of two elements: a III–V ring with a 5 μm wide mesa and a 25 μm outer ring radius and a concentric passive ring on the silicon layer with a WG width of 1.5 μm. There is an intentional WG offset between the silicon ring and III–V ring, which is designed to optimize the coupling between the hybrid ring and the silicon bus WGs. A positive offset means that the outer radius of the III–V ring is larger than that of the silicon. The electrodes of the diode sit on the $n$-GaAs layer and $p$-GaAs on top of the III–V mesa, respectively. The fundamental lasing mode overlaps with the III–V and the silicon and then to the bus waveguide. The silicon bus WG is curved along with the hybrid ring with a 200 nm or 250 nm gap.

 

Fig. 1. Schematic structure of the hybrid QD microring laser on silicon and its fabrication process: (a) passive components fabrication on SOI, (b) GaAs substrate bonding to transfer QD gain layers to SOI, (c) $p$-metal deposition and ring mesa formation, (d) $n$-metal deposition and residual III–V removal, (e) passivation and probe metal deposition, (f) schematic cross-section of the microring mesa, and (g) its SEM image.

Download Full Size | PDF

Figures 1(a)–1(e) highlight the main fabrication steps to build these hybrid QD microring lasers. First, the silicon WG and grating couplers were patterned on the 400 nm thick device layer on an SOI substrate [1 μm thick buried oxide (BOX) layer] with a 248 nm deep UV (DUV) process and a ${\mathrm{Cl}}_2$-based reactive-ion etching (RIE) dry-etch tool [Fig. 1(a)]. Then, the GaAs-based laser epitaxial material with eight QD layers was transferred to the patterned SOI substrate through molecular bonding. This step does not require a precise alignment between the two samples. The bonded sample was annealed at 200°C because of the large thermal expansion coefficient mismatch between GaAs and silicon. The GaAs substrate was selectively removed in a solution of ${\mathrm{H}}_2{\mathrm{O}}_2\text{:}{\mathrm{NH}}_4\mathrm{OH}=30:1$, leaving the 2 μm thick III–V layer stack on silicon [Fig. 1(b)]. The III–V ring mesa was then formed by inductively coupled plasma dry etch with a gas mixture of ${\mathrm{BCl}}_3/{\mathrm{Cl}}_2/\mathrm{Ar}$ to expose the bottom $n$-type contact layer. The Ni/Pd/Au-based $p$- and Ge/Au/Ni/Au-based $n$-type metallization were deposited on top of the mesa and on the $n$ layer, respectively [Figs. 1(c)–1(d)]. The devices were passivated with thick dielectric layers to reduce the parasitic capacitance. Finally, contact vias were etched, and a thick metal pad was evaporated for device probing [Fig. 1(e)]. More device fabrication details can be found in [9]. Figure 1(f) shows the schematic device cross-section and the corresponding scanning electron microscopy (SEM) image in Fig. 1(g). It is noted that the imperfect mesa etch step led to a “skirt” above the $n$ layer, which could contribute to a higher cavity loss and impact coupling to the bus WG.

The devices described in this paper were made possible by the high alignment accuracy of our 193 nm DUV stepper, which was not available to us when we reported on self-aligned hybrid microring lasers [14]. The design allows us to position the $n$-metal contact outside of the ring similar to hybrid micro-disk lasers [20]. This increases the $n$-type contact area, therefore improving the injection efficiency because of better overlap between the carrier path with the fundamental mode in the ring. However, unlike the vertical optical coupling design in micro-disk lasers [18–20], lateral optical coupling is still preferred for the convenient integration of a MOS capacitor in the future. After the fabrication, the SOI wafer with active devices was diced and mounted on the heat sink with temperature control for device characterization. The output light was collected with a cleaved single-mode fiber (SMF) through grating couplers in the silicon device layer.

The laser’s performance is highly dependent on the coupling between the hybrid ring and bus WG, which determines the mirror loss of the laser cavity, in both clockwise and counterclockwise directions. Due to the large propagation constant difference between the hybrid ring WG ($n_{\mathrm{TE}0\_\mathrm{ring}}=3.587$) and the bus WG ($n_{\mathrm{TE}0\_\mathrm{bus}}=3.501$), only a small fraction of power can be coupled to the straight-bus WG across a 200 nm wide coupling gap, which is the limit of our DUV lithography system. So, a curved bus WG structure was implemented to extend the coupling interaction length and therefore enhance the coupling strength [33]. A finite-difference time-domain (FDTD) simulation was conducted to evaluate the coupling between the ring and the bus WG. The source is situated in the hybrid ring and is injecting the fundamental TE mode of the hybrid structure, as shown in Fig. 2(b). This mode corresponds to the one with the highest confinement factor in the active area, leading to the highest modal gain during electrical injection. Two monitors are placed on the other end of the ring: one in the ring, one in the bus WG. A study has been made by sweeping the angle of curved coupling section $\theta$, shown in Fig. 1(a), and the corresponding coupling length.

 

Fig. 2. (a) Simulated coupling efficiency at different offsets as a function of angle $\theta$ of the curved coupling section for fundamental TE mode injection in the ring and corresponding top-view electrical field profiles for (b) $\theta =75$ and (c) $\theta =105$ under zero offset. Light injection location is highlighted by the yellow arrow in (b).

Download Full Size | PDF

We then extract the percentage of the optical power in the ring and in the bus WG and normalize it to the injected power at the source. Several simulations were performed for different angles of the curved coupling section. Figure 2 shows the power coupling coefficient as a function of angle $\theta$ for a ring with a diameter of 50 μm, 200 nm coupling gap, and varied offset between silicon and III–V ring WG. The simulation shows that the coupling coefficient oscillates, indicating optical power shifting back and forth between the hybrid ring resonator and curved silicon bus WG when the curved coupling section is longer than 40 μm. A maximum coupling of 23% can be obtained for 100 and 200 nm offset when the coupling section is 33 μm long. This behavior is indeed like a bent directional coupler and gives us another degree of freedom to control the coupling coefficient besides a narrow coupling gap [33]. We also note that a positive offset does not significantly affect the coupling efficiency. However, a negative offset will diminish the coupling efficiency because the negative offset further increases the modal index of the hybrid ring WG, e.g., $n_{\mathrm{TE}0\_\mathrm{ring}}=3.595$ at $-100\mathrm{nm}$offset, and subsequently increases the phase mismatch with the bus WG.

The thermal performance of the hybrid QD ring lasers is also investigated. Micro-lasers are known for their high thermal impedance, which easily makes them subject to joule heating. One of the major obstacles for micro-lasers on the SOI substrate is the thermal barrier of the thick BOX layer in the SOI substrate, which blocks the heat transmission from the active device to the substrate [15,34]. We simulated the temperature distribution of the QD microring laser on silicon with a finite element method. Figure 3(a) shows the temperature distribution at the device cross-section of the hybrid QD microring laser with 50 μm diameter and 5 μm mesa width, and the SOI substrate with 1 μm thick BOX and 600 μm thick silicon substrate whose temperature was fixed at 20°C in the simulation. When 50 mW thermal power is dissipating in the QD active region, the maximum temperature at the active region rises to about 29°C. We compared the thermal impedance of the microring lasers with different mesa widths and BOX thicknesses, as shown in Fig. 3(b). We choose the 5 μm mesa as a compromise between thermal impedance, optical mode confinement, and injection efficiency, which is defined by the overlap between the electrical carriers and the lasing mode. A 1 μm thick BOX was used in this work in order to minimize the thermal impedance while still having sufficient mode confinement in micro silicon WGs. Unfortunately, the 1 μm thick BOX layer leads to a poor coupling efficiency between the grating couplers and the optical fiber. The predicted thermal impedance of the GaAs-based QD microring laser is about 180 K/W with 50 μm diameter and 5 μm mesa width. The silicon-based QD laser has potentially better thermal performance compared with GaAs native substrate, as silicon has much higher thermal conductivity than GaAs. The impact of the BOX layer on the thermal characteristics can be largely eliminated with a special thermal shunt design, as we have shown previously [15]. By integrating thermal shunts in future designs, the BOX thickness can be increased, which will improve the coupling efficiency between the grating coupler and the fiber without having a negative impact on the thermal performance of the device.

 

Fig. 3. (a) Temperature distribution at the cross-section of the hybrid QD microring laser with 50 μm diameter, 5 μm mesa width, and 1 μm BOX under 100 mW dissipation power; (b) simulated thermal impedance of hybrid QD microring laser with 50 μm diameter and varying mesa width and BOX thickness.

Download Full Size | PDF

3. RESULTS AND DISCUSSION

First, the static direct current (DC) performance of the hybrid QD lasers was characterized at various stage temperatures. The light–current–voltage (LIV) curve of 50 μm diameter devices is shown in Fig. 4. The output power shown here is a normalized value found by deducting the fiber coupling loss (around 10 dB) from one output of the laser. The QD microring laser shown in Fig. 4(a) has a 2.4 mA threshold current at 20°C, corresponding to a threshold current density of about $340\mathrm{A}/{\mathrm{cm}}^2$. Some devices showed threshold currents as low as 0.7 mA; however, their optical output power was below 1 μW. The microring laser can work under continuous wave (CW) operation of up to a 70°C stage temperature, as shown in Figs. 4(b) and 4(c). A threshold current of 6.5 mA and a maximum output power of 130 μW are measured at 70°C. The maximum operating temperature of the microring laser is lower than that of a Fabry–Perot laser with the same III–V epitaxial stack on Si, where operation up to 100°C was demonstrated [9]. The limited working temperature is believed to be due to an imperfect $n$-type contact resistance that results in extra joule heating and excess cavity loss from the “skirt” in the microring mesa. The $n$-type specific resistances are measured to be $1\times {10}^{-4}\mathrm{\Omega }/{\mathrm{cm}}^2$, two orders of magnitude higher than our typical results on previous InP-on-silicon devices. The large contact resistance is likely due to the low-contact anneal temperature of 300°C. While higher anneal temperatures should result in better contacts, the large thermal mismatch between GaAs and silicon will likely result in more defects at those temperatures. Additional process optimization is required. Kinks in the light–current (LI) curves are due to mode hopping between the resonant modes and the lasing direction, switching between clockwise and counterclockwise modes. Both could be due to the $-10\mathrm{dB}$ reflection from grating couplers. Because of the broad bandwidth of the optical gain from the QD material, usually multiple longitude lasing modes can be observed, as shown in Fig. 4(d). But, single-spectral-mode operation with an up to 45 dB side-mode suppression ratio was still observed at certain bias points when one cavity mode dominates the lasing to deplete most of the carriers. The free spectral range (FSR) of the ring is about 2.96 nm, indicating that the group index of the fundamental mode in the ring $n_g$ is 3.69. As the stage temperature is raised, the lasing wavelength drifts to longer wavelengths at about 0.08 nm/°C. More stable single-spectral-mode operation can be achieved by reducing the dimension to increase the FSR, reducing the inhomogeneity in the QD active region growth to minimize spectral gain bandwidth, adding a grating structure in the ring cavity to assist lasing mode selection [35], and using optical injection locking to stabilize the single-mode operation [36].

 

Fig. 4. (a) LIV curve of a typical hybrid QD microring laser at 20°C, (b) temperature-dependent LI curves, and (c) the function of threshold current and maximum output power with different stage temperatures, (d) the output lasing spectrum at 20°C and 50°C, respectively, (e) the summary of threshold current, and (f) maximum output power distributions of 50 μm diameter hybrid QD ring lasers with varied coupling gaps, coupling angles, and WG offsets.

Download Full Size | PDF

A series of hybrid microring lasers with different designs in coupling angle $\theta$ and WG offset were fabricated on the same die. Figures 4(e) and 4(f) summarize the threshold current and maximum output power of the lasers with 200 nm and 250 nm coupling gaps, respectively. The curved coupling angle ranges from 0 to 120 deg, and the WG offset ranges from −100 nm to 200 nm. As shown in Fig. 4(e), the majority of the devices exhibit threshold currents between 1.5 and 3 mA at a 20°C stage temperature. For both 200 nm and 250 nm gap designs, a small bent coupling angle is desired for low-threshold current, when the equivalent mirror loss is smaller. The optimum WG offset is between 0 and 100 nm. On the other hand, slightly stronger coupling strength between the hybrid ring and bus WG is desired to improve the laser slope efficiency, i.e., the output power, as shown in Fig. 4(f). It shows that a WG offset larger than 100 nm is optimum for higher output power by enhancing the overlap of the mode in the hybrid ring with the bus WG. Those results are in reasonably good agreement with the simulation in Fig. 2.

The dynamic performance of the hybrid QD microring lasers were characterized with a 20 GHz light wave component analyzer. The normalized electrical–optical (EO) response of the hybrid QD microring laser under small signal current modulation is shown in Fig. 5(a). The 3 dB bandwidth of the laser is about 3.5 GHz at 5 mA injection current at 20°C. A maximum 3 dB bandwidth of 7.5 GHz was achieved with 22 mA driving current. We note that InAs/GaAs QD lasers typically show direct modulation bandwidths around 10 GHz or less due to strong gain compression and low saturated gain [37,38]. In the data transmission test, we modulated the same hybrid QD microring laser with a $2^7–1$ non-return-to-zero (NRZ) pseudorandom binary sequence (PRBS) signal with the testing setup shown in Fig. 5(b). An O-band optical amplifier (OA) was used to compensate for the optical loss from the grating couplers and boost up the power entering the detector. A tunable filter (TF) filtered out all but one resonant lasing mode to suppress the noise from other modes and amplified spontaneous emission (ASE) noise from the OA. The resulting signal was sent to a photodetector (PD) with high responsivity. The differentiated outputs from the PD connected to a digital communication analyzer (DCA) and an error detector (ED) for data analysis.

 

Fig. 5. (a) Small signal response of the hybrid QD microring laser under varied bias current, (b) schematic of the laser modulation experiment, (c) NRZ eye diagrams at 12.5 Gb/s and 15 Gb/s data rate, and (d) corresponding bathtub plot with data error rate.

Download Full Size | PDF

Two communication configurations were used to test the direct modulation of the hybrid QD microring lasers. In the back-to-back (B2B) configuration, the output power of the laser was directly input into the testing equipment. Then, a 4.2 km long SMF-28 fiber link was also used. In both configurations, open eye diagrams are observed at a 12.5 Gb/s data rate direct modulation at a stage temperature of 20°C, as shown in Fig. 5(c). For both link configurations, the bias and swing voltage are 1.7 V and 0.8 V, respectively. The optimum signal-to-noise ratio (SNR) is 11.1 dB for the B2B link and 11.2 dB for the 4.2 km SMF link. The same test was also performed at a stage temperature of 50°C. As shown in Fig. 5(c), the B2B communication at 12.5 Gb/s data rate shows an open eye with more noise than at 20°C. The measured SNR is 9.4 dB at the bias point and the voltage swing, which are 1.27 V and 0.65 V, respectively. A clear eye opening was observed up to 15 Gb/s under the B2B configuration, with slightly more noise visible in the eye diagram. It translates to a high-energy efficiency of 1.2 pJ/bit. MOS capacitor-induced photon lifetime modulation with higher intrinsic bandwidth than conventional current modulation will be implemented in future devices to overcome limited direct modulation bandwidth in QD lasers [17,39].

Figure 5(d) shows the bit error rate (BER) measurement under the same condition. Error-free operation was obtained at 20°C with the BER below ${10}^{-12}$ in both the B2B and SMF links. The bathtub curves in Fig. 5(d) show an eye opening of 0.33 unit interval (UI), or 26.4 ps for the B2B configuration; and it is 0.47 UI, or 37.6 ps after the 4.2 km SMF fiber. Because of the ultra-low dispersion of the SMF fiber around the 1300 nm wavelength, no BER deterioration was observed after the SMF transmission. The B2B link shows higher BER of $3\times {10}^{-6}$ at 50°C, which is still well above the threshold for error-free communications with an error correction algorithm. We expect similar BER measurement results at a 15 Gb/s modulation rate. An experimental confirmation will be conducted when we acquire a new BER tester with higher-speed capacity.

4. CONCLUSION

In this work, we developed InAs/GaAs QD microring lasers on a silicon substrate with a hybrid integration approach. These hybrid QD microring lasers on silicon with 50 μm diameter have shown both low-threshold current and high direct modulation bandwidths. We report up to 15 Gb/s NRZ data communication with direct modulation, a record for O-band QD lasers on silicon, to the best of our knowledge. Error-free communications at 12.5 Gb/s for both B2B and 4.2 km SMF data links were measured. When we include the AC and DC bias of the microring laser, the calculated energy efficiency is about 1.2 pJ/bit at a 15 Gb/s data rate at 20°C, more than eight-fold energy efficiency improvement over our previous design with QW active region [17]. The hybrid integration of QD materials further extends our device library of silicon photonics. It provides energy and cost-efficient solutions, not only for light sources in high-speed interconnects in future datacenters and supercomputers, but also for applications in optical switching and memory [40–42].