Open Access
Add to BibSonomy
Abstract
We developed short-active-length distributed Bragg reflector (DBR) lasers to reduce the power consumption of chip-to-chip optical interconnects. These lasers have buried bulk InGaAsP waveguides to increase the coupling efficiency between the active region and DBR to 99.79% from the 98.14% of our previous DBR lasers that had InP channel waveguides. We achieved continuous wave operation of 5- to 80-µm active-length DBR lasers and the 5-µm-long laser consumed 24 fJ/bit with a 10-Gbps NRZ signal. The threshold current of the 5-µm laser was 51 µA, which compares favorably to our previous 10-µm DBR lasers with a threshold current of 170 µA.
© 2024 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
The power consumption used for data transmissions in data centers is increasing due to the rapid growth of Internet traffic [1]. Increasing the transmission capacity and reducing the power consumption of the transceivers would enable the use of short optical interconnections over distances as small as those of chip-to-chip interconnects [2,3,4]. To employ optical technologies at such ultra-short distances, it is important to further reduce the power consumption of optical transmitters, and in the case of chip-to-chip interconnects, it is necessary to have an operating energy of less than 34 fJ/bit [3]. A candidate technology is the vertical-cavity surface-emitting laser (VCSEL), which is widely used for data transmissions from a few meters to about 100 m [5,6,7]. By reducing the active region to 5 µm, VCSEL would consume only about 48 fJ/bit in operation. However, it is challenging to reduce the power consumption any further by reducing the aperture diameter.
We have proposed a membrane laser in which a thin III-V layer is sandwiched by low refractive index materials such as air and SiO2. Since the membrane laser has a larger optical confinement factor than that of conventional lasers, we expect it would provide an increase in modulation efficiency. In addition, the lasing wavelength can be precisely controlled by changing the grating period. This makes the membrane laser suitable for wavelength division multiplexing (WDM), which is an important technology that will increase the transmission capacity in a single waveguide when optical interconnects are introduced to chip-to-chip interconnections. In addition, WDM will reduce the energy cost of CPUs by using optical filtering and switching [8,9]. For extremely short-distance optical interconnects such as on-chip interconnects, we have studied lambda scale embedded active-region photonic-crystal (LEAP) lasers [10,11,12] and achieved an energy cost of 4.4 fJ/bit by using 2.4-µm-long active region [10]. However, the maximum output power was limited to ∼40 µW when we used a 2.5-µm-long active region with six quantum wells (QWs).
For chip-to-chip optical interconnects, we believe that a membrane laser on SiO2/Si may be able to achieve appropriate output power and operating energy levels. Thus, we are trying to reduce the active volume of membrane distributed Bragg reflector (DBR) lasers [13] and distributed reflector (DR) lasers [14] on SiO2/Si substrate. When we used nine QWs and reduced the active length of a DBR laser to 10 µm, the device exhibited a 170-µA threshold current, 150-µW maximum output power, and direct modulation capability at a bit rate of 25.8 Gbit/s [13]. The active region of this device consisted of InGaAsP multiple quantum wells (MQWs) that were buried in an InP slab. On the other hand, the DBRs were composed of InP channel waveguides. To suppress the optical mode field mismatch between the channel waveguides and the active region, we used a 1.5-µm-wide InP waveguide. A concern is that DBRs with such a wide waveguide will increase the optical loss due to the generation of higher-order optical modes. In addition, the channel waveguide has a higher loss than that of the buried waveguide. Thus, to further improve the device characteristics, it is necessary to reduce the optical loss of the DBRs.
In this paper, we report on lasers whose DBRs are fabricated with buried bulk InGaAsP waveguides to suppress optical loss. Depending on the transmission distance, lasers with different active lengths are required; therefore, we fabricated lasers with active lengths ranging from 5 to 100 µm. In the case of the laser whose active length was 5 µm, we reduced the threshold current to 51 µA compared with the 170 µA of the previous laser and achieved a maximum output power of 136 µW with single-mode lasing. The fiber coupling loss was 2.75 dB thanks to the use of a spot-size convertor (SSC) consisting of an InGaAsP/InP buried waveguide and an SiOx waveguide. The very short active region realized an energy cost of 24 fJ/bit in the case of direct modulation of an NRZ signal at 10 Gbit/s.
2. Design
First, we describe the design of a DBR laser suitable for reducing the active region length. In the case of membrane DBR lasers, the intersection between the active and DBR regions should be carefully designed because coupling loss is to be expected. Figure 1(a) shows schematic diagrams of our previous (left) [13] and the current one (right). The active region of both is composed of a nine-period 1.55-µm InGaAsP MQW with a thickness of 150 nm and width of 0.6 µm and it is buried in InP. The DBRs of the previous device are fabricated with InP channel waveguides because the fabrication process is simple. In this case, to prevent mode mismatch, the refractive index of the waveguide should be close to that of the active region by optimizing its width. A blue line in Fig. 1(b) shows the calculated optical coupling efficiency between the InP waveguide and the buried heterostructure (BH) when the width of the InP waveguide is changed. The maximum coupling efficiency of 98.14% is achieved when the InP channel waveguide is 1.5-µm wide. However, the loss in such a wide waveguide will increase because it becomes a multi-mode waveguide. The coupling efficiency is only about 98.14% at maximum. On the other hand, as shown by the red line in the Fig. 1(b), the calculated coupling efficiency of 99.79% can be achieved using a buried InGaAsP waveguide with a photoluminescence (PL) peak at 1.3 µm. Since the number of reflections increases as the active length becomes shorter, we decided to fabricate DBRs with InGaAsP waveguides to reduce the coupling loss at the joint.
Fig. 1. (a) Schematic diagrams of our previous (left) and current laser (right). (b) Optical coupling efficiency between the InP waveguide and the BH when the width of the InP waveguide is changed.
Figure 2(a) shows the calculation model. At the lasing threshold, the gain in the active region compensates for all the propagation and mirror losses [15]. The threshold gain gth is given by
Fig. 2. (a) Calculation model. (b) Relationship between the grating coupling coefficient κ and gth when Γgth is 40 cm-1 and the active length is 5 µm. (c) Relationship between active length of DBR laser and front DBR length which gives Γgth of 40 cm-1 for various coupling coefficients from 100 to 1200 cm-1. (d) Relationship between active length of DBR laser and threshold current. (e) Slope efficiency for various coupling losses between the MQW and waveguide sections when κ = 800 cm-1.
We define the average internal loss <αi > and the mirror loss αm as
In our calculation model, the mirrors are not lossy, and all mirror loss is delivered to the outside. To consider coupling loss, we multiple C by the lossless mirror intensity reflection R: C is included in internal loss. Among them, we assumed that αi(DBR) is zero in the calculation. As shown by αm, short-cavity lasers need high reflectivity mirrors to keep gth small. In addition, it is important to have a large optical confinement factor that can compensate for the increase in mirror loss. Therefore, our membrane laser, in which a thin InP layer including the active region is sandwiched by low refractive index materials such as SiO2, is suitable for fabricating a short cavity because the membrane structure has three times the optical confinement factor of the conventional laser structure.
The effective length is a critical parameter because the ratio of the DBR section to the effective cavity length increases as the active region of the laser is shortened. Γz ≈ La/ Leff is determined by the effective length Leff = La + Leff.DBR which is given by
Next, we evaluated the effect of κ and the active region length on the threshold current. The threshold current Ith and slope efficiency are given by
As shown in Fig. 2(d), although the threshold current decreases as the active length decreases, it increases as the active length continues to decrease. The active length at which the threshold current begins to rise decreases as the coupling coefficient increases. As a result, the minimum threshold current decreases as the coupling coefficient increases, and one can see that a κ of ∼800 cm-1 seems high enough to achieve a low threshold current with a 5-µm-long laser. Figure 2(e) compares the slope efficiency in the cases of various coupling losses between the MQW and waveguide sections when κ = 800 cm-1. The shorter the active length is, the more affected the coupling efficiency becomes. The slope efficiency obtained with the proposed structure (0.5% coupling loss) is about three times higher than that of the conventional structure (2% loss). These results indicate that the key to realizing high-performance short-cavity DBR lasers is to fabricate a high-coupling coefficient DBR with low coupling loss.
On the basis of the above calculations, we chose the grating coupling coefficient to be ∼800 cm-1 to suppress the excess loss caused by a deep grating. As shown on the right side of Fig. 1(a), we located DBRs at both the front and rear ends of the active region. The length of the front-side DBRs ranged from 10.2 to 29.7 µm, while the rear-side DBR was 150-µm long in order to provide ∼100% reflection. To reduce the fiber coupling loss, a spot-size convertor was fabricated that consisted of an InGaAsP waveguide and SiOx waveguide. The DBR section was fabricated by etching the surface of InP. To obtain a threshold gain of 40 cm-1, the front DBR length was adjusted. For example, the front DBR length was set to 29.7 µm in the case of the 5-µm-long active region and 10.2 µm in the case of the 80-µm-long active region. When the active length was less than 15 µm long, the threshold gains of the second mode were over two times larger than that of the first mode, which indicated the DBR laser would be single mode.
3. Fabrication details
The active region consisted of a 250-nm-thick InP-based membrane on a 2-µm-thick thermally oxidized Si substrate. The total thickness of a nine-period InGaAsP MQW was 150 nm and the width was 0.6 µm. InGaAsP waveguide sections were buried in a 250-nm-thick InP membrane whose thickness and width were the same as those of the MQW. Figure 3 shows the device fabrication procedure. We employed buried regrowth after direct bonding of the III-V active layers to fabricate the lasers on SiO2/Si substrate [16]. First, InP-based layers including MQWs and an InGaAs etch stop layer were grown on InP substrates by using metal-organic vapor phase epitaxy (MOVPE). Next, the epitaxial wafer was bonded to a thermally oxidized Si substrate by using O2 plasma-assisted direct bonding [Fig. 3(a, b)]. The InP substrate was removed by lapping and wet etching using an InGaAs etch stop layer, which left a thin InP-based membrane including the MQWs on the Si substrate [Fig. 3(c)]. Next, the MQWs were partially removed by dry and wet etching, and the butt-joint regrowth of InGaAsP was carried out by MOVPE [Fig. 3(d)]. Then, we defined a mesa stripe that consisted of the MQW (active region) and InGaAsP (waveguide section) by using the same etching process, and the InP was regrown in order to bury the mesa stripe [Fig. 3(e)]. Then, n-type doping was carried out by silicon ion implantation, and p-type doping was carried out by zinc thermal diffusion to fabricate lateral p-i-n junctions for current injection [Fig. 3(f)]. After that, an SiO2 mask was deposited and surface gratings were formed by electron beam (EB) lithography and dry etching [Fig. 3(g)]. The etching depth was approximately 25 nm in order to obtain a grating coupling coefficient of 800 cm-1. Next, InGaAsP waveguides were made by dry etching and Au-based electrodes were deposited by a lift-off process [Fig. 3(h)]. Finally, spot-size convertors (SSCs) were formed. SiOx cores were formed by photo lithography and dry etching, and the electrode area was opened after depositing the SiO2 cladding layer [Fig. 3(i)]. The InGaAsP waveguide narrowed from 0.6 µm wide to 0.3 µm wide and the InP waveguide narrowed from 7 µm wide to 0.5 µm wide over 300 µm. Beyond the point that the InGaAsP waveguide disappears, the InP waveguide narrowed from 0.5 µm wide to 0.1 µm wide over 100 µm. The thickness of the SiOx core was 3 µm and the width was 3.5 µm.
Fig. 3. Device fabrication procedure. (a) Epitaxial growth of MQWs on InP substrate. (b) O2 plasma-assisted direct bonding. (c) InP substrate and InGaAs etch stop layer removal. (d) Regrowth of InGaAsP. (e) Forming MQW mesa stripes and InGaAsP waveguides. (f) n- and p- type doping. (g) Forming surface grating. (h) Forming InP waveguides and electrode deposition. (i) Forming SSC.
4. Device characteristics
Here, we discuss the laser characteristics of the DBR lasers on SiO2/Si substrate. After the fabrication process, the substrates were lapped and cleaved into bars. The devices were mounted on a temperature-controlled stage, and all the measurements were carried out at a temperature of 25°C. Figure 4(a) shows the output power versus injected current (L-I characteristics) of DBR lasers with active lengths of from 5 to 80 µm. Threshold current and slope efficiency are shown in Fig. 4(b), (c). We observed an almost monotonic decrease in threshold current and maximum output power as the active region length was shortened. Next, we measured the relative intensity noise (RIN) spectra and extracted the relaxation oscillation frequencies (fr). When the output power was not large enough, we used an erbium doped fiber amplifier (EDFA) and an optical bandpass filter. Figure 4(d) plots fr versus the square root of the injected current minus the threshold current of the 5 µm-long DBR laser. fr was determined by fitting the RIN spectrum with a numerical model [15]. The fr slopes (D factors) of the lasers with various active region lengths are summarized in Fig. 4(e). The highest D factor (18.2 GHz) is that of the shortest-cavity (5 µm) laser, which shows the effect of increasing the modulation efficiency by shortening the active length. The fitting curves are shown in Fig. 4(b), (c) and (d). The internal quantum efficiency (ηi), internal loss of the active region (αi(a)) and carrier lifetime (τn) are estimated to 0.75, 17 cm-1 and 1.07 ns, respectively.
Fig. 4. (a) Output power versus injected current (L-I characteristics) of DBR lasers with active lengths of 5 to 80 µm. (b) Threshold current. (c) Slope efficiency. (d) fr versus the square root of the injected current minus the threshold current of 5 µm-long DBR laser. (e) D factor of lasers with various active region lengths.
Figure 5(a) shows the relationship between power density and wavelength shift Δλ. The calculated thermal resistance using the slope of Δλ / ΔT of 0.013 nm/K [17] is shown in Fig. 5(b). Thermal resistance is usually constant, which is the case when the active length is long enough, but it becomes smaller as the active length decreases. This is due to heat escaping in the front-back directions of the active layer.
To determine appropriate cavity lengths for low-energy-cost optical communications, we estimated the required energy cost of lasers with various active region lengths in operation at different modulation speeds. Here, we assumed that the maximum modulation speed is given by 1.3 × 1.55 × fr [15] and the energy cost is the product of the bias current and applied voltage divided by the maximum modulation speed. fr was evaluated from the RIN measurement. The relationship between energy cost and estimated bit rate is shown in Fig. 6(a). Thanks to the minimum active region volume and successful reduction in coupling loss, the laser with the active length of 5 µm can be expected to have the lowest energy cost, 12.7 fJ/bit, when the modulation speed is ∼10 Gbps.
Finally, to demonstrate low-energy-cost direct modulation, we measured the eye diagram for the 5-µm-long active-region DBR laser. Figures 6(b) and (c) show the eye patterns of the laser modulated with 10-Gbit/s and 25.8-Gbit/ NRZ pseudo-random bit sequence (PRBS) signals with a length of 231-1. In the case of the 10-Gbit/s modulation, the bias current was 0.2 mA and the bias voltage was 1.2 V, which corresponds to an energy cost of 24 fJ/bit. The energy cost is higher than the value expected from fr because the signal-noise-ratio is lower even though an EDFA is used. In the case of the 25.8-Gbit/s modulation, the bias current was 0.6 mA and the bias voltage was 1.63 V, which corresponds to an energy cost of 38 fJ/bit.
Fig. 6. Relationship between energy cost and estimated bit rate. (b) Eye pattern of laser modulated with a 10-Gbit/s signal. (c) Eye pattern of laser modulated with a 25.8-Gbit/s signal.
5. Conclusion
We developed a DBR laser using an InGaAsP waveguide to suppress the optical loss and thereby to reduce the power consumption of chip-to-chip optical interconnects. In the case of membrane DBR lasers, the intersection between the active and DBR regions should be carefully designed to suppress the coupling loss, which is particularly important for short-active-length lasers as they are more affected by coupling loss. The previous device used InP channel waveguides because their simple fabrication process. However, to prevent mode mismatch, wide (1.5 µm) waveguides had to be used and the coupling efficiency was only about 98.14% at maximum. By comparison, the current device’s use of a buried InGaAsP waveguide increases the coupling efficiency to 99.79% and its slope efficiency (0.5% coupling loss) is about three times higher than that of the conventional structure (2% loss). The fabricated 5- to 80-µm-long active-length DBR lasers consumed a small amount of power in continuous wave operation. The threshold current of the 5-µm active-length laser was 51 µA and the energy cost was 24 fJ/bit with a 10-Gbps NRZ signal, so we believe that these “ultrashort” cavity membrane DBR lasers are promising for short optical links that require low power consumption.
Acknowledgments
We thank K. Ishibashi, Y. Shouji, Y. Yokoyama, and J. Asaoka for assistance in fabricating the devices.
Disclosures
The authors declare no conflicts of interest.
Data availability
Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.
References
1. Cisco, “Cisco Annual Internet Report (2018-2023),” Cisco Public, (2020), https://www.cisco.com/c/en/us/solutions/collateral/executive-perspectives/annual-internet-report/white-paper-c11-741490.html.
2. M. Taubenblatt, “Optical interconnects for high-performance computing: How Do We Get Beyond the Cost & Power Wall?” in Proceedings of Optical Fiber Communication Conference and Exposition (2021), paper Th4 l.4.
3. D. A. B. Miller, “Device requirements for optical interconnects to silicon chips,” Proc. IEEE 97(7), 1166–1185 (2009). [CrossRef]
4. Y. Ajima, “Optical connection of top-level supercomputers: current status and future expectations,” in Proceedings of Optical Interconnects XXII (2022), paper 1200704.
5. G. Cardinali, F. Hjort, N. Prokop, et al., “Low-threshold AlGaN-based UVB VCSELs enabled by post-growth cavity detuning,” Appl. Phys. Lett. 121(10), 103501 (2022). [CrossRef]
6. C. Paranthoen, C. Levallois, G. Brévalle, et al., “Low Threshold 1550-nm Emitting QD Optically Pumped VCSEL,” IEEE Photonics Technol. Lett. 33(2), 69–72 (2021). [CrossRef]
7. H. R Ibrahim, Ahmed M. A. Hassan, Xiodong Gu, et al., “1060 nm Single-mode Metal-aperture VCSEL Array with Transverse Resonance and Low Power Consumption below 50 fJ/bit,” in Proceedings of European Conference on Optical Communication (2021), paper 21433378.
8. A. Shacham, K. Bergman, and L. P. Carloni, “Photonic networks-on-chip for future generations of chip multiprocessors,” IEEE Trans. Comput. 57(9), 1246–1260 (2008). [CrossRef]
9. S. Rumley, D. Nikolova, R. Hendry, et al., “Silicon photonics for exascale systems,” J. Lightwave Technol. 33(3), 547–562 (2015). [CrossRef]
10. K. Takeda, T. Sato, A. Shinya, et al., “Few-fJ/bit data transmissions using directly modulated lambda-scale embedded active region photonic-crystal lasers,” Nat. Photonics 7(7), 569–575 (2013). [CrossRef]
11. S. Matsuo and K. Takeda, “λ-Scale Embedded Active Region Photonic Crystal (LEAP) Lasers for Optical Interconnects,” Photonics 6(3), 82 (2019). [CrossRef]
12. K. Takeda, T. Tsurugaya, T. Fujii, et al., “Optical links on silicon photonic chips using ultralow-power consumption photonic-crystal lasers,” Opt. Express 29(16), 26082–26092 (2021). [CrossRef]
13. K. Takeda, E. Kanno, T. Fujii, et al., “Continuous-wave Operation of Ultra-short Cavity Distributed Bragg Reflector Lasers on Si Substrates,” in Proceedings of Compound Semiconductor Week (2016), paper ThD1-2.
14. E. Kanno, K. Takeda, T. Fujii, et al., “Twin-mirror membrane distributed-reflector lasers using 20-(m-long active region on Si substrates,” Opt. Express 26(2), 1268–1277 (2018). [CrossRef]
15. L. A. Coldren and S. W. Corzine, Diode Lasers and Photonic Integrated Circuits (Wiley-Interscience Publication, 1995).
16. T. Fujii, T. Sato, K. Takeda, et al., “Epitaxial growth of InP to bury directly bonded thin active layer on SiO2/Si substrate for fabricating distributed feedback lasers on silicon,” IET Optoelectron. 9(4), 151–157 (2015). [CrossRef]
17. T. Fujii, K. Takeda, K. Hasebe, et al., “Heterogeneously Integrated Membrane Lasers on Si Substrate for Low Operating Energy Optical Links,” IEEE J. Sel. Top. Quantum Electron. 24(1), 1–8 (2018). [CrossRef]
