In this paper, we report the results of the efforts to extend our previous work through the packaging and redesign of a heterogeneously integrated silicon-photonic circuit for use in a modulation side-band injection-locked optical RF generation system. Towards that effort, we attempted to improve the RF spectrum coverage of our design by decreasing the laser cavity length. Despite the unintended formation of an additional parasitic cavity in that device, we demonstrated increased spectrum coverage between 5 and 50 GHz in a packaged module with an ∼1-Hz linewidth.
© 2014 Optical Society of America
As engineers, we try to accommodate the continually increasing demand for high-bandwidth communications into our already crowded spectrum  by extending our usage to higher frequencies, all the way up to 110 GHz  and beyond. Optical RF generation is a promising alternative for synthesizing spectrally pure RF with the capability of covering the full spectrum from 1 to 110 GHz. Optical RF generation inherently offers extreme bandwidth because very small shifts in wavelength correspond to very large changes in frequency. For example, 250 GHz of RF bandwidth can be covered with a change in wavelength of just 2 nm when operating near 1550 nm .
Optical RF is produced by heterodyning on a photodiode and can be implemented in a variety of ways. Integrated systems have been created based on techniques such as direct mixing, direct mixing of modulation sidebands, multi-wavelength lasers [5–8], optical phase locked loops (OPLL) [9–11], and various types of injection locking schemes [12–19].
Here we focus on the packaging and design improvements of a dual laser chip in a hybrid silicon platform for use in a widely tunable and spectrally pure optical RF source, utilizing a modulation side-band injection-locking scheme . Modulation side-band injection-locking is advantageous because it offers continuous tunability and harmonic suppression without sacrificing power, while avoiding complex feedback mechanisms and the need for ultra-narrow filters.
2. Silicon-photonic RF chip design
Two variations of the hybrid silicon-photonic RF chip are shown in Fig. 1. In Fig. 1(a), we see the original design , while Fig. 1(b) shows the new design. Both designs start with two 1-mm-long InP chips (Covega SAF1128C) that provide the gain in the laser cavities. The interface facets of the gain chips have an antireflective coating while the back facets of the gain chips have a 90%-reflective metallic coating, which acts as a broadband reflector. In both designs, the gain chips are coupled to a silicon chip that has waveguides, Bragg gratings, and optional multi-mode interference (MMI) couplers. The Bragg gratings act as wavelength-selective mirrors and output couplers for the lasers, while the optional 50/50 MMIs can be used to split and combine the lasers into individual ports and a combined output port on the chip. Those ports are then accessed using a V-Groove fiber array connected to an external injection-locking system that primarily consists of a lithium niobate modulator and a filter. The new design further extends the integration level of the orignal design by providing placeholders for optional on-chip thermistors and thermally-tunable filters. On top of the silicon chip is a thin layer of silicon dioxide used for electrical isolation. NiCr heaters, which directly overlap the Bragg gratings, are deposited on top of the silicon dioxide and used to tune the spectral response of the gratings. Finally, gold electrical contact pads are used to supply power to the heaters and gain chips.
In the original design the laser cavity length was approximately 4.5 mm (1 mm of gain chip plus 3.5 mm of silicon waveguide). In the new design the laser cavity length has been shortened to approximately 1.25 mm (1 mm of gain chip plus 0.25 mm of silicon waveguide). This design change was undertaken in order to increase the free spectral range (FSR) of the laser cavity and promote single-mode operation. The free spectral range of a cavity is given by ,Equation (2) extends Eq. (1) to account for the two parts of our laser cavity Eq. (1) and Eq. (2), c = 299, 792,458 the speed of light in a vacuum, ngc = 3.22 the modal index of the gain chip, Lgc = 1030 μm the length of gain chip, nSi = 3.439 the modal index of the silicon waveguide, and LSi is the length of the silicon waveguide in μm. By decreasing LSi from 4,500 μm down to 250 μm the FSR of the cavity was increased from 10 GHz to 36 GHz. The gain-chip modal index was provided by the manufacturer, while the silicon waveguide modal index was calculated using Lumerical™  simulation software.
3. Silicon chip fabrication procedure
We start our fabrication process with a silicon-on-insulator (SOI) wafer that has a 1.5-μm silicon device layer on top of a 3-μm buried oxide (BOX) layer. The sample is first cleaned by ultrasound in deionized water. This is followed by a second cleaning procedure (repeated 3 times) in which the sample is first placed in a solution of 3:1 sulfuric acid:hydrogen peroxide (piranha etch) and then briefly dipped in buffered oxide etch (BOE). After blow drying with nitrogen, the sample is spin-coated with poly-methyl-methacrylate (PMMA) and the waveguides and reflective gratings are exposed on a Raith e LiNE electron-beam lithography system. The sample is then developed in a 1:3 mixture of methyl iso-butyl ketone:isopropyl alcohol (MIBK:IPA). The silicon is then etched using a SAMCO fluorine-based inductively coupled plasma (ICP) reactive-ion etching system. This is followed by an oxygen ash and another clean in a piranha etch. Next, a 300-nm silicon dioxide layer is deposited using a SAMCO plasma-enhanced chemical vapor depostion (PECVD) system in order to electrically and optically isolate the metal heaters and contacts from the underlying silicon. The sample is then spin-coated with Futurrex NR9-1500PY photoresist and the NiCr heater pattern is exposed by UV lithography. The resist is developed in Futurrex RD6 followed by a 20-second oxygen ash to ensure clean openings. 300-nm of NiCr is sputtered in an Angstrom deposition system onto the sample, followed by an acetone lift-off and another clean in piranha etch. Next, 300-nm gold contacts are evaporated and patterned using the same procedure employed to produce the NiCr heaters. Finally, the silicon is coated with a protective layer of hard-baked MicroChem SU8 photoresist and the edges are polished to produce smooth facets. The sample then undergoes a final clean in piranha etch .
4. Alignment, assembly, and packaging
Assembling the final module requires an active alignment to be completed without the gain chip being permanently attached to anything prior to being epoxied to the silicon chip. Using a custom vacuum chuck with a built-in thermo-electric cooler maintains the gain chip at a constant position and temperature during assembly.
During alignment, the gain chip is mounted on a center stage and electrical probes are brought into contact. A measurement fiber is then aligned to the 90%-reflective facet of the gain chip. The silicon chip, attached to a holder, is mounted on an adjustable stage and pre-aligned to the gain chip. The silicon chip holder is then removed from the alignment setup and optical adhesive Norland NOA61 is applied to the edge of the silicon chip. The chip holder is then replaced on the alignment stage and aligned until the maxicmum power is detected out of the 90% reflective face. An optical spectrum analyzer (OSA) could be used during the alignment to ensure lasing, but we have found that a simple power meter is sufficient because of the large difference in collected power between lasing and non-lasing conditions. The NOA61 is UV-cured to bond the silicon chip and the gain chip together. This procedure is repeated for the second gain chip. The silicon-chip/gain-chip assembly then goes through an aging step in a preheated 50°C oven for at least 12 hours. Next the silicon-chip/gain-chip assembly is placed on another temperature-controlled holder, actively aligned, and bonded to a V-Groove fiber array. The whole module then goes through another aging step .
When the completed module has finished aging it is then mounted on a machined aluminum submount using an electrically and thermally conductive epoxy. The same epoxy is used to secure a 10-kΩ thermistor and ground contact to the submount. At the same time a non-conductive ceramic-like epoxy is used to provide a very secure hold around the fiber array and to attach a power connector board. After curing on a 50°C hotplate for at least 4 hours, the assembled sub-mount is bonded to a TEC, which in turn is also bonded to a copper heatsink with attached fan using the same electrically/thermally conductive epoxy and curing procedure. The completed assembly is then placed in a machined acrylic housing .
5. Test setup
The test setup consists of a packaged module powered by two ILX Lightwave 3724B laser drivers, a Keithley 2400 SM DC power supply (integrated heater driver), and a Thorlabs TED 200C temperature controller. The optional on-chip MMIs, filters, thermistors, and combined output port were not present on the devices used in these modules. The external injection-locking components consist of an isolator on the master laser, a circulator on the slave laser, a commercial 40-GHz phase modulator (Covega LN66S), an Agilent PSG-E8267D microwave signal generator capable of providing reference signals up to 44 GHz, a broadband RF amplifier (Picosecond Pulse Labs 5882) that acts as the distortion element for the reference oscillator (to improve harmonic generation), a Yenista Optics XTA-50 tunable optical filter that passes half the sidebands and removes the carrier and other sidebands, an erbium-doped fiber amplifier (IPG Photonics EAD-1K), two semiconductor optical amplifiers (SOA) (Thorlabs S9FC1004P), a u2t XPDV4120R 100-GHz photodiode mixer, and an Agilent PXA-N9030A Signal Analyzer. A detailed layout of the injection locking system is shown in Fig. 2.
6. Results and discussion
Both the original and new designs were fabricated and packaged. Figures 3(a) and 3(b) show the top and side views respectively of the 4.5-mm-cavity packaged module. Through the acrylic housing, we can see the aluminum submount, TEC, copper heatsink and fan. Figure 4 shows a perspective view of the 1.25-mm-cavity packaged module. Figure 5 shows a close up of the surface components in the packaged module. They contain the power connector, thermistor, RF source chip, and V-Groove output. Packaging was primarily pursued to make the devices more robust for testing; however, the packaging process we have developed is scalable to high-volume manufacturing because the materials are producible at high volumes and the processes to attach them, such as wire bonding and epoxy dispensing, are common manufacturing techniques. While an evanescent silicon/III–V coupling scheme like that of Fang et al.  would be preferred when it comes to production volume, edge coupling is still a mainstay of many fiber optic assemblies and the results presented here could be transferred to an evanescent coupling system in the future. Using these packaged integrated modules, we have successfully synthesized tunable narrow-linewidth RF. The 4.5-mm-cavity (original design) packaged module was used in Figs. 6, 7, and 9, where the fourth harmonic of the low-frequency reference oscillator was used to lock the slave laser.
In Fig. 6, we show a comparison between the spectra of the locked and unlocked states near 57.5 GHz. We can see that when the lasers are locked, the frequency separation is exactly the desired RF frequency and that the linewidth narrows significantly compared to the unlocked state. Figure 7 shows the lasers injection locked at 57.5 GHz with a measured linewidth of approximately 1 Hz. The sidebands of the pure tone are most likely due to acoustic vibrations in the injection-locking system fiberoptics. In the future these spurious signals can be overcome by complete integration of the injection locking system. Our proposed design can be seen in Fig. 8. This design overcomes the need to integrate non-reciprocal materials for the construction of isolators and circulators by utilizing wavelength-selective reflectors and a traveling-wave modulator instead. Beginning with the master laser, the light is split by a 50/50, 2X2 MMI and passes through the slave-laser reflector because it is at a different wavelength. The light then travels through a second MMI, completing its coupling to the adjacent waveguide, proceeding to the traveling-wave modulator where sidebands are produced by an RF source, which could be on- or off-chip depending on the integration technology. The modulated master then passes through the next MMI. Here the master-laser reflector allows the sidebands to pass while reflecting the master laser. This causes the master laser to return through the same MMI, completing the coupling to the opposite waveguide, labeled in Fig. 8 as the master-laser output. The sidebands pass through another MMI and couple into the slave laser. The slave-laser light follows a symmetric path opposite that of the master. A traveling-wave modulator is, however, a non-reciprocal device, and is only effective in one direction, allowing the slave laser to pass with minimal modulation. Finally, the two lasers are combined in another 50/50 MMI, giving dual outputs.
Figure 9 shows thermal tuning of the unlocked laser separation around 57.5 GHz. The plot was captured by configuring the signal analyzer to hold the maximum power value while stepping the current to the integrated thermal tuner from 14 mA to 28 mA in 1-mA increments. The changing current to the heater produces a temperature change in the heater, which acts to thermally tune the grating underneath. The actual temperature changes were not directly measured. A 2.5-GHz tuning range is shown with a demonstrated tuning efficiency on the order of 180 MHz/mA. This tuning efficiency can be significantly improved by the addition of thermal isolation trenches around the heated sections of the waveguides . A max-hold plot with such small steps was now aquirable due to the improvements in thermal stability and electrical connection reliability provided by packaging. We use the unlocked spectra to demonstrate tuning coverage because at the span at which Figs. 9 and 10 are taken, the high resolution bandwidth makes for no discernable difference in the display of the locked verus unlocked spectra in terms of the noise floor and visible linewidth. Only a small frequency shift of ∼50 MHz distinguishes the locked and unlocked states, as depicted in Fig. 6. Such a small shift would be imperceptible on the scale of Figs. 9 and 10.
While we have measured an ∼1-Hz linewidth at every frequency we have injection locked to, our goal of completely continuous tuning from 1–100 GHz was impeded by a mode-hopping issue. For example, attempts to tune the laser separation beyond the span seen in Fig. 9 caused the lasers to mode hop to a significantly different RF separation. To improve on this issue, we attempted to shorten the silicon laser cavity to increase the cavity free spectral range (FSR) as described in Section 2.
Using the reduced-cavity-length device our frequency coverage over the full tuning range was improved. As seen in the unlocked tuning spectra of Fig. 10, we were able to cover around 2 GHz continuously in every 5 GHz span from 5 to 50 GHz. The data shown here was captured by holding the master-integrated heater at different currents, indicated by the different colors, and tuning the master-laser-gain-chip current from 150 to 322 mA. The variations in power are caused by varying optical powers at different laser currents, as well as higher RF loss and lower conversion efficiency of the photodiode at higher frequencies. By design, we expect our gratings to be around 30% reflective with a 3-dB bandwidth of ∼90 GHz . Considering that Fabry-Perot modes of the 4.5-mm-cavity-length device should be spaced at 10 GHz, it is likely that the single-mode operation is determined by fairly small differences in the gain threshold . In explaining the output response of the 1.25-mm-cavity-length device, we know that there is a laser cavity with cavity-mode spacing of 36 GHz. The collected spectra shows that there is an additional cavity influencing the output. The frequency spacing of the modes is ∼5.4 GHz. Utilizing Eq. (2) from Section 2, we find that this corresponds to a silicon path length of roughly 7.1 mm, which corresponds to a cavity formed by the back mirror of the gain chip and the start of the heater for the second filter section in Fig. 1(b). Based on the designed FSR and reflector strengths, the designed cavity has a 3-dB bandwidth of 7 GHz . Since this is larger than the FSR of the parasitic cavity, it is favorable for the laser to mode hop as the shorter cavity device is tuned through the parasitic cavity’s modes. The coverage went up because, despite the presence of more modes, the tuning range around those modes was not diminished significantly because the bandwidth of the parasitic cavity was similar to or greater than that of the 4.5-mm-cavity-length device. This is because the parasitic cavity was long with a weak reflector of at most 8%, based on the FSR and covered bandwidth . One reason for the variation in bandwidth covered at each mode is because the lengths of both cavities change simultaneously when the gain chip current is tuned. Additionally, the gain chip and silicon chip are in intimate contact and the temperature change in the gain chip as the current is tuned may also be causing a temperature change in the grating.
In Fig. 11, we show the single-sideband phase noise of an optically generated 11.4-GHz signal produced from a 11.4-GHz reference (solid curve), and phase noise of the 11.4-GHz reference itself (dashed curve). We demonstrated a phase noise of −107.5 dBc/Hz at a 10 kHz offset, an improvement of 25 dB over our previously reported results . We attribute this improvement to the increased thermal stability and electrical connection reliability provided by packaging. Below 20 kHz, mechanical vibrations of the loose fiber in the injection-locking system cause the large phase noise difference between the optically generated signal and the reference . At this time we believe that the injection power was not sufficient to maximize the noise correlation between the lasers, resulting in the excess phase noise above 20 kHz. Going forward the injection strength could be increased by an improved coupling design between the silicon waveguides and the optical fibers, something that was not optimized on this device. Despite being higher than that of the reference, the root mean square (RMS) phase error (100 Hz to 1 MHz) of the optically generated signal is still about 3°, which is suitable for 8 bit-Quadrature Amplitude Modulation (QAM) encoded digital microwave links with bit error rates less than 10−6 .
In summary, we described the design, fabrication, and packaging of an integrated heterogeneous module. We have demonstrated improved frequency coverage and stability with our newly designed and packaged modules and used them to show tunable RF generation with an ∼1-Hz linewidth. In the future, thermal stops will also be introduced into our design to improve the thermal tuning efficiency. We plan on further integrating the injection-locking system by incorporating filters and modulators, either on-chip , such as presented in Fig. 8, or heterogeneously , and anticipate eventually incorporating the reference oscillator and control electronics on the same platform.
The authors acknowledge the support of R. Nelson from the US Air Force Research Laboratory and G. Pomrenke from the Air Force Office of Scientific Research, as well as other US Government agencies.
References and links
1. “US Frequency Allocation Chart,” (Nat. Telecommun. Inf. Admin., Washington DC, USA, 2011) (2011).
2. J. Macario, P. Yao, S. Shi, A. Zablocki, C. Harrity, R. D. Martin, C. A. Schuetz, and D. W. Prather, “Full spectrum millimeter-wave modulation,” Opt. Express 20, 23623–23629 (2012). [CrossRef] [PubMed]
3. D. Grund, G. Ejzak, G. Schneider, J. Murakowski, and D. Prather, “Heterogenous integrated silicon-photonic module for producing widely tunable narrow linewidth RF,” in Microwave Photonics (MWP), 2013 International Topical Meeting on, (2013), pp. 100–103. [CrossRef]
4. R. K. Price, V. B. Verma, K. E. Tobin, V. C. Elarde, and J. J. Coleman, “Y-branch surface-etched distributed bragg reflector lasers at 850 nm for optical heterodyning,” IEEE Photon. Technol. Lett. 19, 1610–1612 (2007). [CrossRef]
5. C. Cheng, Z. Ling-Juan, Q. Ji-Fiang, L. Yang, W. Wei, and L. Cai-Yun, “Dual-wavelength distributed bragg reflector semiconductor laser based on a composite resonant cavity,” Chinese Phys. B 21, 094208 (2012). [CrossRef]
6. S. D. Roh, T. Yeoh, R. B. Swint, A. E. Huber, C. Y. Woo, J. S. Hughes, and J. Coleman, “Dual-wavelength InGaAs-GaAs ridge waveguide distributed bragg reflector lasers with tunable mode separation,” IEEE Photon. Technol. Lett. 12, 1307–1309 (2000). [CrossRef]
8. A. R. Criado, C. de Dios, P. Acedo, G. Carpintero, and K. Yvind, “Comparison of monolithic optical frequency comb generators based on passively mode-locked lasers for continuous wave mm-wave and sub-thz generation,” J. Lightwave Technol. 30, 3133–3141 (2012). [CrossRef]
9. S. Ristic, A. Bhardwaj, M. J. Rodwell, L. A. Coldren, and L. A. Johansson, “An optical phase-locked loop photonic integrated circuit,” J. Lightwave Technol. 28, 526–538 (2010). [CrossRef]
10. L. Ponnampalam, M. J. Fice, F. Pozzi, C. C. Renaud, D. C. Rogers, I. F. Lealman, D. G. Moodie, P. J. Cannard, C. Lynch, L. Johnston, M. J. Robertson, R. Cronin, L. Pavlovic, L. Naglic, M. Vidmar, and A. J. Seeds, “Monolithically integrated photonic heterodyne system,” J. Lightwave Technol. 29, 2229–2234 (2011). [CrossRef]
11. M. Lu, H. Park, E. Bloch, A. Sivananthan, A. Bhardwaj, Z. Griffith, L. A. Johansson, M. J. Rodwell, and L. A. Coldren, “Highly integrated optical heterodyne phase-locked loop with phase/frequency detection,” Opt. Express 20, 9736–9741 (2012). [CrossRef] [PubMed]
12. A. Criado, C. de Dios, E. Prior, G. Dohler, S. Preu, S. Malzer, h. Lu, A. Gossard, and P. Acedo, “Continuous-wave sub-thz photonic generation with ultra-narrow linewidth, ultra-high resolution, full frequency range coverage and high long-term frequency stability,” IEEE Trans. THz Sci. Technol. 3, 461–471 (2013). [CrossRef]
13. H. L. Stover and W. H. Steier, “Locking of laser oscillators by light injection,” Appl. Phys. Lett. 8, 91–93 (1966). [CrossRef]
14. L. Goldberg, H. F. Taylor, and J. F. Weller, “FM sideband injection locking of diode lasers,” Electron. Lett. 18, 1019–1020 (1982). [CrossRef]
15. L. Goldberg, H. F. Taylor, J. F. Weller, and D. M. Bloom, “Microwave signal generation with injection-locked laser diodes,” Electron. Lett. 19, 491–493 (1983). [CrossRef]
16. C. Laperle, M. Svilans, M. Porer, and M. Tetu, “Frequency multiplication of microwave signals by sideband optical injection locking using a monolithic dual-wavelength DFB laser device,” Opt. Express 17, 20727–20734 (2009). [CrossRef]
17. J. Huang, C. Sun, B. Xiong, and Y. Luo, “Y-branch integrated dual wavelength laser diode for microwave generation by sideband injection locking,” Opt. Express 17, 20727–20734 (2009). [CrossRef] [PubMed]
18. C. Sun, J. Huang, B. Xiong, and Y. Luo, “Low phase noise millimeter-wave generation by integrated dual wavelength laser diode,” in OFC/NFOEC, (2010).
19. L. Ponnampalam, C. Renaud, I. Lealman, L. Rivers, P. Cannard, M. Robertson, M. Moodie, F. van Dijk, A. Enard, F. Blache, M. Goix, F. Mallecot, and A. J. Seeds, “Injection-locked integrated twin DBR lasers for mm-wave generation,” in Eur. Workshop Photon. Solutions for Wireless, Access, and In House Networks, (2009).
20. D. W. Grund, G. A. Ejzak, G. J. Schneider, J. Murakowski, and D. W. Prather, “A widely tunable narrow linewidth RF source integrated in a heterogeneous photonic module,” J. Lightwave Technol. 32, 1363–1369 (2014). [CrossRef]
21. G. J. Schneider, J. A. Murakowksi, S. Shi, and D. W. Prather, “Radiofrequency signal-generation system with over seven octaves of continuous tuning,” Nat. Photonics 7, 118–122 (2013). [CrossRef]
22. A. E. Siegman, Lasers (University Science Books, 1986), chap. 11, p. 432.
23. “Lumerical design software,” (Lumerical Solutions, Inc., Vancouver, Canada).
24. D. W. Grund Jr., G. A. Ejzak, G. J. Schneider, J. Murakowski, S. Shi, and D. W. Prather, “Integrated silicon-photonic module for generating widely tunable, narrow-line RF using injection-locked lasers,” in Proc. SPIE8259, (2012), paper 825906. [CrossRef]
25. A. W. Fang, B. R. Koch, R. Jones, E. Lively, D. Liang, Y.-H. Kuo, and J. E. Bowers, “A distributed bragg reflector silicon evanescent laser,” IEEE Photon. Technol. Lett 20, 1667–1669 (2008). [CrossRef]
26. T. Erdogan, “Fiber grating spectra,” J. Lightwave Technol. 15, 1277–1294 (1997). [CrossRef]
27. A. M. Sarangan, M. W. Wright, J. R. Marciante, and D. J. Bossert, “Spectral properties of angled-grating hig-power semiconductor lasers,” IEEE J. Quantum Electron. 35, 1220–1230 (1999). [CrossRef]
28. B. E. A. Saleh and M. C. Teich, Fundamentals of Photonics (Wiley-Interscience, 1991). [CrossRef]
29. W. Reuter, “Source and synthesizer phase noise requirements for qam radio applications,” Microwave Product Digest pp. 36–57 (1999).
30. T. Baehr-Jones, R. Ding, Y. Liu, A. Ayazi, T. Pinguet, N. C. Harris, M. Streshinsky, P. Lee, Y. Zhang, A. E.-J. Lim, T.-Y. Liow, S. H.-G. Teo, G.-Q. Lo, and M. Hochberg, “Ultralow drive voltage silicon traveling-wave modulator,” Opt. Express 20, 12014–12020 (2012). [CrossRef] [PubMed]