BEAM: Design and characterization of a 10 Gb/s broadband electroabsorption modulator

Transcription

1 BEAM: Design and characterization of a 1 Gb/s broadband electroabsorption modulator S.D. McDougall, B.C. Qui, G. Ternent, D.A. Yanson, V. Loyo-Maldonado, J.H. Marsh Intense Photonics Ltd., 4 Stanley Boulevard, Hamilton International Technology Park, Blantyre, Glasgow G72 BN Scotland, UK. ABSTRACT The market for data modulators at 1 Gb/s is currently dominated by Mach-Zehnder phase modulators fabricated in LiNbO 3 (LN). However they are relatively expensive to manufacture and large compared to semiconductor devices. InP based electroabsorption modulators (EAMs), are more compact; however they have a limited bandwidth (5-8 nm) over which chirp is in the correct range to allow >8 km reach. This paper reports the broadband electroabsorption modulator (BEAM) concept in which reach performance in line with LN modulators can be achieved using integrated InP components. The BEAM consists of a series of EAMs, each one tuned to give the correct chirp over a certain wavelength range. The bandwidth of the BEAM can be extended to cover the C-band (1535nm 1565nm). In addition, a semiconductor optical amplifier (SOA) is serially integrated in order to recover the total insertion loss. Details of the design, fabrication and testing of prototype BEAM chips operating at 1 Gb/s are reported. Quantum well intermixing technology is employed to realize the multiple bandgaps required for the prototype chips which are fabricated on semiinsulating InP substrates. Highlights of the operational characteristics of the BEAM chips include extinction ratios of up to 12 db at 1 Gb/s and SOA gains of 2 db. Keywords: Electroabsorption Modulator, Bandwidth, Quantum well intermixing, Photonic Integration 1. INTRODUCTION In order to realize photonic integrated circuits on InP, active functional building blocks such as semiconductor optical amplifiers (SOAs), electroabsorption modulators (EAMs), and p-i-n photodiodes must be monolithically integrated on a single substrate along with low-loss waveguide connectors. There are three main methods that can be employed to integrate multiple active components on a single InP substrate: etch and re-growth; selective area growth; and quantum well intermixing. The former two processes can be used independently or in conjunction with each other to give effective results 1. However both of these processes rely on complex wafer patterning and processing within the epitaxial growth stage and may require multiple growth runs to complete the device. This can lead to high cost and low yield in manufacturing. In addition a major issue associated with the performance of integrated devices made by re-growth methods is the misalignment of optical waveguides within the device, which leads to unwanted internal reflections. There can also be reliability issues associated with defects at growth run interfaces. Quantum well intermixing (QWI) is a powerful technique that can be used to spatially alter the active region properties of multi-quantum well InP-based epitaxial structures 2. One of the key advantages of QWI over the other integration methods is its simplicity and low cost, as all of the processing is carried out post epitaxial growth. This leads to the intrinsic suitability of QWI processes for high yield manufacture of photonic integrated circuits. In QWI techniques, thermal processing is used to diffuse point defects through the epitaxial layers, leading to inter-diffusion of atomic species between the layers. However only the average composition of very thin layers is significantly altered during this process. When quantum wells are present, with thicknesses between 5 nm and 1 nm, QWI can produce a change in their composition and width, which has an associated change in the emission absorption or refractive index properties. These properties lead to the other key attribute QWI processing can provide to photonic integrated circuit manufacture: that the optical waveguide is perfectly aligned throughout, with only a very small change in the effective index due to

2 the slight difference in quantum well composition. The key to the fabrication of integrated circuits is then the ability to alter selectively the atomic inter-diffusion rate across a wafer, leading to a multiple bandgap QWI process 3. This paper presents details of the design, fabrication and testing of a BEAM chip fabricated by a multiple bandgap QWI process in InP. The BEAM (broadband electroabsorption modulator) concept is shown schematically in Figure 1. Depending on their bias condition and operation wavelength, EAMs give either a positive or negative chirp on the optical data pulses they produce 4,5. The reach in fiber of these data is then limited by group velocity dispersion. In order to achieve a reach of over 8 km in fiber, the chirp parameter α of a 1 Gb/s EAM has to be controlled to a value between -.5 to -1. Conventionally in InP EAMs this can only be achieved within a wavelength range of around 5 nm to 8 nm. By using QWI the centre wavelength of the absorption spectrum of an EAM, and hence the spectrum over which the chirp can be suitable for > 8 km reach, can be tuned to a specific value. In the BEAM chip several of these EAMs are serially integrated along an optical ridge waveguide, each one tuned via QWI to cover a separate part of the C-band with suitable chirp. In this way any input signal within the C-band input to the BEAM can be modulated by the appropriate EAM, while the others are made transparent either by grounding or low forward bias. In order to recover any excess insertion loss, an SOA section is included at the input of the BEAM chip. In the schematic example BEAM chip layout shown in Figure 1, four 1 Gb/s EAMs are integrated on a semi-insulating substrate with an SOA section. SOA EAM1 EAM2 EAM3 EAM4 λ = 155 nm nm nm nm nm Figure 1: Schematic layout of BEAM chip To integrate a single EAM and SOA on an InP chip requires at least three distinct semiconductor bandgaps. This is illustrated in Figure 2. The SOA section bandgap is that of the as-grown quantum well material. In order to operate efficiently within the SOA gain bandwidth, the EAM absorption peak must be widened by around 1 mev so that absorption is minimized at V bias and maximized at the operating bias (-Von) for optimal extinction ratio. The precise extent of the bandgap widening is determined by the wavelength range over which the chirp is to be controlled. For electrical isolation purposes, a gap between the SOA and the EAM is required. To minimize the total loss of the chip, the isolation region waveguide should be rendered transparent by widening the bandgap as much possible. Absorption -Von EAM V EAM Passive waveguide gain SOA gain Energy (mev) Figure 2: Spectra of multiple bandgap sections in SOA integrated EAM chip

3 For each additional EAM, a further bandgap is required on the chip, with its bandgap widened using the multiple bandgap QWI process. This paper describes the development of demonstrator BEAM chips which consist of an SOA integrated with two separate bandgap EAMs. Section 2 discusses some of the design issues involved in the development of the demonstrator chip, including design of an epitaxial structure that optimizes both SOA gain and EAM extinction ratio and determination of the bandgap detuning required for each modulator. Section 3 describes some of the details of the fabrication of the demonstrator devices, including the QWI process. The performance of the BEAM chips is reported in Section 4, including characterization of the individual SOA and EAM components. 2. DESIGN 2.1. BEAM demonstrator chip It was decided that the most appropriate chip to demonstrate the BEAM concept was a monolithic chip that integrated a SOA and two EAMs, each EAM detuned to a different extent from the as-grown SOA wavelength via QWI. This chip layout represents the simplest possible device that still demonstrates the 3 three essential elements of the BEAM concept: Multi-bandgap monolithic integration process via QWI Broadband operation via multiple EAM integration Insertion loss reduction via SOA integration The following sections summarize the work carried out to design the demonstrator BEAM chip Initial design considerations In order to design an epitaxial wafer structure suitable for the BEAM chip, the following considerations were made. The basic structure had to be able to provide the gain characteristics required by the SOA section, have electroabsorption properties suitable for producing EAMs with high enough extinction ratio at a 1 GHz bandwidth, and be suitable for processing with QWI. Parameter Min Max Wavelength Output Power (dbm) 5 - Reach (km) 8-3dB bandwidth (GHz) 1 - Extinction Ratio (db) 13 - Power consumption (W) Table 1: BEAM demonstrator cardinal point specification Table 1 shows the cardinal point specification used as a basis for the BEAM chip design. An integrated suite of custom design software was developed in order to carry out the complex design of the BEAM chip. This model included an electroabsorption model for QWI processed multi-quantum well material, including the ability to assess insertion loss and chirp spectra across the C-band for multiple bandgap sections, and an SOA gain saturation model. The design work

4 was undertaken with the following general iterative cycle: a wafer structure was proposed in order to give suitable EAM performance; this structure was then simulated with the QWI process model to check that sufficient bandgap shifts could be obtained; the structure was then run through the SOA model to verify suitable gain characteristics; appropriate changes are then made to epi-design. The final epi-structure used for the demonstrator was based on the InAlGaAs/InGaAsP/InP waveguide system with the active region consisting of 8 compressively strained InAlGaAs QWs with InAlGaAs tensile stained barriers EAM design The length and QWI wavelength detuning of the EAMs are determined by the extinction ratio, chirp parameters, insertion loss, and RF power required to achieve the desired specification. From theoretical considerations, the quantum confined Stark effect (QCSE) 6, in which the absorption edge shifts toward longer wavelength when an electrical field is applied perpendicularly to a quantum well, is proportional to the quantum well width to the fourth power. However the oscillator strength, which is determined via the wave function overlap, is inversely proportional to quantum well thickness. In order to determine the exact quantum well parameters for optimum performance, the width and number of quantum wells were varied to achieve the highest extinction ratio, keeping the thickness for intrinsic region constant. It was found that the thickness of the quantum well should be around 8.7 nm. The cavity length and intrinsic region thickness were designed to enable a 1 GHz bandwidth. Figure 3 shows DC extinction ratio curves for the chosen 4 µm long EAM device. 16 Extinction ratio (db/cm) DC bias=.6 V =1. V =1.4 V Wavelength (nm) Figure 3: Simulated extinction ratio spectra at various DC bias levels for a 4 µm EAM The required wavelength detuning between the EAM absorption edge and the operating wavelength is determined by both chip parameter and extinction ratio. The chirp parameter is defined as α =4π n/ λ α (1) where n and α are the changes in refractive index and absorption coefficient respectively due to the change of the voltage applied to the modulator, and λ is the operation wavelength. Both theory and experiment 7 show chirp has to be in the range between -.5 and -1, in order to achieve highest bit-rate and transmission distance product in standard single mode fiber which has positive dispersion. For the chirp to be in the desired range, the bandwidth of the EAM is only about 5 nm as is shown from the calculation in Fig. 4. From the model, the amount of bandgap detuning, which is defined as difference between the operation wavelength and the peak photoluminescence (PL) wavelength of the EAM, for the first modulator is about 45 nm, while for the second one it is 55 nm. This was designed to extend the bandwidth of the demonstrator BEAM device to > 1nm with suitable chirp at fixed DC bias. With variable DC bias, this could be extended to cover as much as half of C-band.

5 .5 Chirp parameter Operation wavelength (nm) Figure 4: Typical calculation of chirp parameter from the electroabsorption model 2.4. SOA design In the BEAM, a SOA is integrated to compensate for chip insertion loss, and for coupling loss between the device and a single mode fiber (SMF). Given that the device is mainly targeted as an alternative to LN modulators for long-reach WDM communication systems, it is required that the device has low chirp, low insertion loss, and high power handling 8. Consequently the SOA needs to be at the input end of the device to avoid introducing any excessive chirp. The optimization is therefore made to achieve highest saturation output power rather than low signal gain. Our modeling shows the saturation power which is also dependent on injection current, peaks at an SOA length of about 2 µm, as is shown in Figure 5. However, due to the fact that thermal impact on the gain is more pronounced when cavity is short, a 5 µm was chosen for the length of the SOA section SOA saturation output power (dbm) I=1 ma I=15 ma I=2 ma Cavity Length (µm) Figure 5: SOA gain saturation design curves

6 3. CHIP LAYOUT AND FABRICATION 3.1. Demonstrator chip layout The BEAM device is an example of a high speed optoelectronic component, as it includes EAMs which have to operate electrically at frequencies up to 1 GHz. There are two competing technologies for such devices in InP, those involving the use of semi-insulating (SI) substrates 9, and those that involve the use of doped substrates in conjunction with thick insulators 1. SI substrates have several advantages over doped substrates including: providing a platform for extension to >4Gb/s; requirement for only standard surface processing technology; allows easy RF passives integration; compatibility with traveling wave devices; simple serial integration of multiple optoelectronics components. Doped substrate devices have a more complex fabrication technology requirement, and are relatively inflexible for other optoelectronic or RF component integration. A schematic representation of the BEAM device layout is shown in Figure 6. The epitaxial layers are grown on an SI substrate to the design described in the previous Section. As the device is on a SI substrate, for the SOA both n and p- contacts are made to the top surface of the chip. Coplanar waveguide (CPW) feeds sitting directly on the SI substrate are used to supply the 1 Gb/s electrical data to the EAM. The EAM sections are deep etched (through the waveguide core) in order to provide control of the p-n junction area, and hence device capacitance, for 1 GHz operation. However, to obtain high quality SOA performance, the waveguide of the SOA section is shallow etched (ridge etch stopping above the waveguide core). In order to suppress any reflection effects at the shallow/deep ridge waveguide boundary, a tapered coupler section is included. Lengths of passive waveguide between each device are included to ensure sufficient electrical isolation. SOA Coupler EAM1 EAM2 AR - coating Semi - Insulating InP substrate Figure 6: Demonstrator BEAM chip schematic AR - coating 3.2. QWI processing Intense has used a unique proprietary QWI process, based on the technique described by Kowalski et al 11, to achieve the multiple bandgap sections on the single semi-insulating InP chip. The process involves the deposition of a QWIinitiating cap followed by a high-temperature anneal. The amount of bandgap shift obtainable by 2 min anneal cycles as a function of anneal temperature is shown in Figure 6. The multiple-bandgap process used allows an arbitrary number of bandgaps to be created on a single substrate, with a maximum shift limited by the saturation region of Figure 7 (~ 18 nm).

7 Blue shift, nm Temperature, deg.c Figure 7: Bandgap blueshift as a function of anneal temperature for a fixed anneal time (2 min.) The multiple-bandgap process was tuned to obtain two bandgap shifts of 2 and 3 nm in each of the two modulator sections (EAM2 and EAM1, respectively) of the BEAM demonstrator device, with a zero shift in the SOA section. The photoluminescence (PL) spectra showing PL intensity peaks measured in the three device sections are presented in Figure 8. The measured shifts are very close to the target values of 2 and 3 nm, with the as-grown bandgap intact in the SOA region nm PL Intensity, a.u nm Wavelength, nm Suppressed Bandgap 1 Bandgap 2 Figure 8: PL spectra measured in the SOA, EAM2 and EAM1 sections showing shifts of, 19 and 31 nm, respectively. The measurement was performed in liquid nitrogen at 77 K. The QWI process has been shown to have good spatial uniformity and reproducibility. The bar chart in Figure 9 shows a PL scan across a quarter of a 3" wafer, containing repeated patterns of the three bandgap regions (Suppressed for zero shift, Bandgap 1 shifted by 2 nm, and Bandgap 2 shifted by 3 nm). The measurement could only be performed at room temperature, with a low PL signal intensity and broad PL peaks, which resulted in a wavelength measurement error of λ = ± 3 nm. It can be seen that the spatial uniformity is the same or better than the measurement error across the whole quarter-wafer. While the shifts in the Bandgap 1 and 2 regions are almost on target, the suppressed region appears to have shifted by about 5 nm. This is attributed to the poor resolution of the PL probe and the relatively small dimensions of the suppressed regions, which made it difficult to obtain a PL reading in the centre of the region.

8 151 PL wavelength, nm Position across quarter-wafer, mm Suppressed Bandgap 1 Bandgap 2 Figure 9: Room temperature PL scans across a quarter-wafer of BEAM demonstrator devices. The dashed lines show the average readings for each bandgap Device fabrication In conjunction with the multi-bandgap QWI process, standard III-V fabrication technology was employed to manufacture the BEAM demonstrator chips. This included a relatively complex four level vertical dry etch process to enable ridge and contact pad layouts to be achieved: shallow ridge for SOA; deep etch for EAM ridge waveguide; n+ layer etch for SOA n-contact; and final level etch for the EAM CPW line on the SI substrates. Verticality was a critical requirement to ensure an appropriate profile of the deep/shallow ridge waveguide coupler section. All electrical contacts were electroplated with gold. Figure 1 shows a SEM cross section of a completed EAM after fabrication is completed. Figure 1: SEM cross section of BEAM demonstrator chip showing EAM2 ridge waveguide and facet.

9 4. CHIP CHARACTERISATION 4.1. Individual SOA characterization In order to assess the gain characteristics of the SOA contained on the BEAM demonstrator chips, individual SOA sections were cleaved off and AR coated on both facets (AR coating reflectivity 1-4 ). The SOAs were tested using a tunable laser source, DC power supply and optical spectrum analyzer and ultra-stable single-mode fiber coupling rig. Figure 11 shows the measured SOA fiber to fiber gain curves for increasing current, with 8 db gain at -2 dbm input and 7dB gain at -1 dbm input power. The saturation output power was assessed separately at >2 dbm. In this measurement, the fiber coupling losses were estimated at 6 db per facet leading to an on-chip gain of around 2 db. This gain is of a suitable level to recover the excess insertion loss of the two EAM sections (estimated to be 3 db each). The gain spectra shown in Figure 12 show that the SOA sections are suitable for C-band operation Fibre-Fibre Gain (db) Drive Current (ma) -2dBm -1dBm Figure 11: Fiber-fiber gain-current curves of BEAM chip SOA section at two input power levels Fibre-Fibre Gain (db) dbm -15 dbm Wavelength (nm) Figure 12: Gain spectra of BEAM chip SOA section

10 4.2. Individual EAM Characterization Similarly to the SOA testing described in Section 4.1, individual EAM chips were cleaved from fabricated BEAM demonstrator chips for independent assessment. The RF properties of the EAMs were measured using an Agilent 873 Network Analyzer. The EAM chips were mounted on a ceramic substrate and wire bonded to a 5 Ω termination. The simple RF equivalent circuit 12 extracted from the measurements is shown in Figure 13. From this, the small signal performance of the devices can be represented with a contact resistance of 1 Ω and reverse bias diode capacitance of.1pf/µm. The 1 Ω contact resistance modeled in the small signal regime is similar to that measured at DC. Figure 13: Small signal equivalent circuit of EAM including 5 Ohm termination Figure 14 shows the measured and modeled return loss to be < 5 dbm to 6 GHz and <3 dbm from 6 GHz to 1 GHz. The 3 db cut off frequency was obtained from the model to be > 1 GHz for a 4 µm long modulator section. Figure 15 shows S21 simulation from the small signal equivalent circuit. Modelled Modeled Measured Figure 14: Modeled and measured S11 for BEAM chip EAM section. In order to characterize the modulation performance of the EAM section, devices were fiber pigtailed and assembled into a butterfly module (k-connector for RF input). Eye diagrams were taken to confirm 1 Gb/s performance. Figure 16 shows a fully open eye from an EAM with in excess of 1 db extinction ratio measured at 155 nm.

11 Figure 15: Modeled S21 response for BEAM chip EAM derived from equivalent circuit model. Figure 16: Measured 1 Gb/s eye diagram (1.9 db E.R. ) from BEAM chip EAM 4.3. BEAM demonstrator chip characterization To date, only limited characterization of the full BEAM demonstrator device has been completed and is reported here. In order to assess the effectiveness of the QWI process in achieving the specified differential bandgap shift of 1 nm between EAM 1 and EAM2, photocurrent measurements of both absorption edges were measured on the same chip. The plot, shown in Figure 17, clearly shows that, for the same bias conditions, the peaks of the absorption edge for each EAM are 1 nm apart confirming the accuracy of the QWI process. A butterfly module was assembled containing a BEAM chip with EAM1 wired to the input 1 Gb/s data signal and EAM2 connected to ground. Figure 18 shows a photograph of the assembled chip. The SOA was biased in order to give db overall insertion loss for the BEAM chip. An eye diagram was taken (Fig. 19) showing modulation at 1 Gb/s from the BEAM chip. The overshoot in the eye resulted from stray reflections from a poor quality antireflection coating on the output facet.

12 EAM 1 EAM Wavelength, nm - 12 Figure 17: EAM1 and 2 photocurrent spectra from same BEAM chip, showing 1 nm absorption peak separation Figure 18: Close- up photograph of butterfly packaged demonstrator BEAM chip Figure 19: 1 Gb/s eye diagram of operational BEAM demonstrator chip

13 5. CONCLUSION In summary, this paper gives an account of the successful development of demonstrator BEAM chips at Intense. The individual SOA and EAM components performed in line with the target specification. The integrated chip itself has been demonstrated, showing on-chip gain suitable for achieving zero insertion loss performance if optimized fiber coupling was employed. The QWI process has been used to tune the bandgaps of two EAMs on demonstrator chip, each EAM showing 1 Gb/s modulation capability. Full characterization of the BEAM demonstrator chips is now underway, including chirp and bit error rate measured over 8 12 km fiber spans. When fully developed, the small form factor of the BEAM chip when compared to LN modulators could make them ideal candidates for co-packaging in source laser modules, both fixed wavelength and tunable, with the associated benefits of reduced overall cost. 6. ACKNOWEDGEMENTS The authors would like to gratefully acknowledge the assistance of the operations team at Intense, and in particular the following for their support: Gianluca Bacchin for epitaxial wafers, Olek Kowalski for QWI process development and Sebastien Bon for test engineering. REFERENCES 1. J. E. Johnson, L. J-P. Ketelsen, J. A. Grenko, S. K. Sputz, J. Vandeberg, M. W. Focht, D. V. Stampne, L. J. Peticolas, L. E. Smith, K. G. Glogovsky, G. J. Przbylek, S. N. G. Chu, J. L. Lentz, N. N. Tzafara, L. C. Luther, T. L. Pernell, F. S. Walters, D. M. Rmoero, J. M. Freund, C. L. Reynolds, L. A. Gruezke, R. People, M. A. Alam, Monolithically integrated semiconductor optical amplifier and electroabsorption modulator with a dual-waveguide spot-size converter input, IEEE J. Select Topics on Quantum Electronics, 6, pp J.H. Marsh, O.P. Kowalski, S.D. McDougall, B.C. Qiu, A. McKee, C.J. Hamilton, R.M. De La Rue, and A.C. Bryce Quantum well intermixing in material systems for 1.5 µm, J. Vac, Sci. Technol. A, 16, pp X.F. Liu, B.C. Qiu, M.L. Ke, A.C. Bryce, and J.H. Marsh, Control of multiple bandgap shifts in InGaAs-AlInGaAs multiplequantum-well material using different thicknesses of PECVD Si 2 protection layers, IEEE Photonics Technology Letters., 12, pp , 2 4. F.Dorguille and F. Devaux, On the Transmission Performances and the Chirp Parameter of a Multiple-Quantum-Well Electroabsorption Modulator, IEEE Journal of Quantum Electronics, 3, p Yonggyoo Kim, Hanlim Lee, Jaehoon Lee, Jaeho Han, T. W. Oh, Jichai Leong, Chirp Characteristics of 1 Gb/s Electroabsorption Modulator Integrated DFB Lasers, IEEE Journal of Quantum Electronics, 36, p9, D.A.B. Miller, J.S. Weiner, and D.S. Chemla, Electric-Filed Dependence of Linear Optical Properties in Quantum Well Structure: Waveguide Electroabsorption and Sum Rules, IEEE Journal of Quantum Electron., 22, PP , R.A. Salvatore, R.T. Sahara, M.A. Bock, and I. Libenzon, Electroabsorption Modulated Laser for Long Transmission Spans, IEEE Journal of Quantum Electron., 38, PP , Tadashi Saitoh, and Takaaki Mukai, Recent Progress in Semiconductor Laser Amplifiers, J. Lightwave Tech., 6, pp , S.C.Lin, P. K. L. Yu, W. S. C. Chang, Efficient, Low Parasitic 1.3um InGaAsP Electroabsorption Waveguide Modulators on Semi-Insulating Substrate, IEEE Photonics Technology Letters, 1, p27., Isamu Kotaka, Koichi Wakita, Osamu Mitomi, Hiromitsu Asai, Yuichi Kawamura, High-speed InGaAlAs/InAlAs Multiple Quantum Well Optical Modulators with Bandwidths in Excess of 2GHz at 1.55um, IEEE Photonics Technology Letters, 1, p1, O.P. Kowalski et al, A universal damage induced technique for quantum well intermixing, Appl. Phys. Lett. 72, pp , G.L.Li, C. K. Sun, S. A. Pappert, W. X. Chen, P. K. L. Yu, Ultrahigh-Speed Traveling-Wave Electroabsorption Modulator Design and Analysis, IEEE Transactions on Microwave Theory and Techniques, 47, p 1177, 1999.