Vertical-cavity multiple quantum well electroabsorption modulators (EAM) offer gigahertz modulation speeds, insensitivity to light polarisation and can be integrated into large arrays. They are therefore good candidates for efficient parallel signal processing architectures. We present high-performance 2×128 and 2×64 EAM arrays that were fabricated at 4” wafer-scale by using optimised fabrication and hybridisation processes. The arrays exhibit contrast ratios of 20:1 with a voltage swing of 1 V, a maximum contrast ratio of 335:1 for a 10 V bias and a modulation frequency in excess of 15 MHz.
© 2005 Optical Society of America
Processing broadband signals in the optical domain may become a key technology for next-generation mobile communication systems, future optical networks and radar systems. While electronic processors are mainstream today, optoelectronic signal processing techniques have long been expected to offer unique performance, and thereby to overcome limitations of electronic processors in terms of speed, power dissipation and cost issues. During the 1990s, large research programmes have been conducted to develop competitive solutions based in particular on ferroelectrical liquid crystal modulator arrays, and have clearly demonstrated advantages of optoelectronic solutions compared to their electronic counterparts in terms of computation power for a given power consumption level, weight or volume of the processor . However, to compete with state-of-the-art electronics, it has often been acknowledged that even larger improvements in parallel processing and integration of optoelectronic solutions have to be achieved. Integrated into small form-factor processors, arrays of vertical-cavity multiple quantum well (MQW) modulators based on the quantum confined stark effect (QCSE)  are good candidates to efficiently manipulate broadband signals and may offer new opportunities to implement novel signal processing operations in the optical domain . Owing to their unique speed advantage, arrays of surface-normal MQW modulators are very well suited for fast and parallel signal processing, and can be developed for both digital and analogue signals.
Several prototype MQW modulator arrays have been reported in the literature for different applications such as automatic pattern recognition (optical correlators) [4–5], optical vector-matrix multiplicators , optical inteconnect  and free-space optical communication [8–9]. In this paper, we present the fabrication and characterisation of EAM arrays for the parallel processing of broadband signals. 1-D arrays can be implemented in different optical signal processing architectures, including direct processing of independent parallel channels, and processing of transformed signals (spectral filtering or signal processing based on time-to-space conversion). Applications range from microwave signal processing (transversal filtering, correlation) for adaptive radar or wireless receivers to parallel signal processing (channel equalisation, amplification, regeneration and conversion; header recognition for optical packet switching) in optical networks.
To satisfy the requirements of most optical signal processing systems, high contrast ratios and response times in the range 10 ns-10 µs are needed . Low power consumption and good uniformity over the modulator array are also criteria for any practical implementation of the technology in real-life scenarios. In order to meet the contrast ratio and low driving voltage requirements, we designed the MQW modulator as an asymmetric Fabry-Perot (FP) modulator. We report on two types of vertical cavity EAM arrays with 2 x 64 and 2 x 128 pixels, respectively, fabricated using a 4” wafer-scale process. Each processed wafer includes 160 array devices with high yield. The arrays exhibit uniform electrical and optical performance across the array active area (5 mm x 5.12 mm), a peak contrast ratio of 335:1 for a 10 V bias, and a response time below 100 ns characterised by static and temporal optical response measurements.
2. The modulator structure design and MOVPE growth
The modulators presented in this work consist of undoped GaAs/AlGaAs quantum wells embedded in an asymmetric FP resonance cavity. The front mirror and back mirror of the FP cavity are n- and p-doped Distributed Bragg Reflector, respectively. When a reverse bias is applied to the modulator, its reflectance is modified by an electrical field induced absorption change based on the QCSE. To achieve high contrast ratios at low driving biases the distance between quantum well exciton and the FP cavity has been optimised to 10 nm by adjusting the cavity length. In our designed modulator structure, there is 5 pairs of n-doped λ/4 Al0.22Ga0.78As/Al0.77Ga0.23As front mirror, a 60 period of 70Å GaAs and 60Å Al0.3Ga0.7As quantum wells and 25 pairs of Al0.2Ga0.8As/Al0.69Ga0.31As p-doped back mirror. The simulation spectrum without bias shows the FP cavity resonance peak at 856 nm. The reflectance at 856 nm decreases with increased applied bias due to the red-sift of QW exciton peak. When the absorption inside the cavity increases, the total reflectance of the modulator drops to zero, producing very high contrast ratios. However, if the bias is increased even higher, the reflectance at operating wavelength increases again, thereby lowering the contrast ratio.
Two etch-stop layers with high and low aluminium content, respectively, are needed to be grown before the n-i-p diode structure growth to allow easily removal of the GaAs substrate and the first etch-stop layer by selective wet chemical etching.
The n-doped front mirror was grown on top of the two etch-stop layers, followed by intrinsic QWs and the p-doped back mirror by MOVPE. To characterise the quality of the structure from its front mirror side after the growth, one of the grown wafers was bonded to a SI-GaAs substrate, and the original GaAs substrate and the first etching stop layer were removed by lapping and selective chemical etching. Reflectance spectra at different positions from centre to the edge on the 4” wafer were measured with a spectrophotometer, and the results demonstrate the good agreement between the optical modelling and the measurement results. The FP cavity position varies by 1.2% across the wafer, from 857 nm at the wafer centre to 867 nm close to the edge (4.5 cm from the centre) due to radial non-uniformity of the epitaxy growth. This is considered as a good uniformity for a 4” wafer grown by MOVPE .
The doping concentration in p- and n-doped material has been measured by Hall effect to be 6×1017 cm-3 and 1×1019 cm-3, respectively. High doping levels offer a good possibility to make Ohmic contacts with low resistance. In fact, a highly p-doped back mirror is paramount to the overall electrical quality of the device because p-contacts are processed on the surface of the back mirror, and the electrical field crosses the heterojunctions of the whole back mirror (3.3 µm).
3. The EAM array fabrication and hybridisation to its electrical connector
The dimension of both 2×128 and 2×64 EAM arrays is 6.0 mm×6.2 mm with an active area of 5 mm×5.12 mm. The pixel pitch in the array direction is 40 µm and 80 µm in the 2×128 and 2×64 arrays, respectively. Each pixel is 2 mm long, and the modulator rows in the arrays are separated by a 1 mm gap, as illustrated in Fig. 1(a). A transmission line model (TLM) test structure is arranged in each individual chip, enabling inspection of n- and p-contact quality during the process, as shown in Fig. 1(b). A mesa height test structure is also placed in the chip bottom-left corner. Each chip has alignment marks for different process steps, dicing the wafer into chips and flip-chip bonding. The EAM arrays were fabricated on the 4” wafer-scale, whereby creating 160 arrays from one wafer in same process run.
The EAM array fabrication process is schematically illustrated in Fig. 2(a). The first step was to fabricate an individual p-contact on each pixel by standard optical lithography, metal evaporation and lift-off. The pixels were formed by etching mesas 4.5 µm deep through the back mirror and QW region using wet etching. A 90% fill factor for the 2×64 array and 80% for the 2 x 128 array, respectively, were achieved after etching process optimisation. A common n-contact frame around the active area was produced using similar processes for fabrication of the p-contact, then both p- and n-contact were alloyed at high temperature to ensure good Ohmic behaviour of the contacts. The surface was passivated and via holes for metal bumps were made to enable flip-chip mounting of the EAM array to its electrical connector (EC). The metal bumps were formed by optical lithography, metal evaporation and lift-off processes, as shown in Fig. 2(b). The final step was to dice the wafers into chips by mechanical sawing.
The EC was designed to bring all electrical contacts to the periphery of a 17 mm×17 mm chip, allowing easily connect the contacts to an external driver. The EC has 2×128 and 2×64 metal tracks with the same pitch as their associated EAM arrays, respectively, and each metal track is expanded to a wire-bonding pad. Two wide common contacts are arranged at corresponding common contact positions of the EAM arrays and connected to a metal frame around the EC with six via openings.
The ECs were fabricated on a 4” Si substrate with SiO2 insulating layer. The tracks and pads with thickness of 1 µm were formed by metal evaporation and wet etching. To passivate the EC surface an insulating layer was deposited, and the openings for the contacts were created by optical lithography and dry etching. The metal bumps were then fabricated on the openings at corresponding positions to the EAM array, as shown in Fig. 2(c). The metal tracks were connected to each pixel of the EAM array using the metal bumps by flip-chip bonding. Then underfiller was introduced into the gap between the EAM array and the EC, as shown in Fig. 2(d). After the underfiller was cured, the GaAs substrate of the optical chip was removed by combining mechanical lapping and selective chemical etching, as shown in Fig. 2(e).
Figure 3 shows fabricated 2×64 and 2×128 arrays that flip-chip bonded on their ECs after the thinning process. Hybridised chips were then mounted and wire-bonded to a flexible cable including individual tracks for each modulator and zero-insertion force connectors to connect the array to a PCB drive electronics board.
By comparison, only a few EAM arrays could be produced if the EC were formed on the EAM epi wafer directly due to the large dimension of the EC, which would significantly increase device cost. Using the hybridisation of the EAM arrays to the EC is also an efficient way to eliminate reflectance background from the contacts, thereby enhancing the signal to noise ratio. Moreover the EC serves as a mechanical support to the array after removal of the GaAs substrate, which enhances the mechanical stability of the device.
4. Characterisations of the EAM arrays
4.1 Electrical properties
Before the hybridised EAM arrays were mounted on their flexible cables, I–V characteristic of the EAM arrays were measured by using a HP 4155A semiconductor parameter analyser and a probe station with two probes that contacted on common contact and each p contact pad. In such a measurement the electrical characteristic uniformity of the EAM arrays was also obtained, and imperfect pixels could be mapped and excluded in the following wire bonding process.
Figure 4 shows two typical I-V curves of the pixels in 2×128 and 2×64 arrays. The insert of the Fig. 4 shows breakdown voltages (Vbr) of all pixels in one row in a 2×64 EAM array and one row in a 2×128 EAM array, which indicates a good electrical uniformity of the EAM arrays.
The experimental results show a high breakdown voltage of about −30 V with about 0.3 pA/µm2 leakage current density for both type of arrays. This low leakage current causes negligible heating, which ensures a high thermal stability of the modulators. A maximum electrical field of about 3.7×105V/cm can be derived from the following formula :
where W=0.8 µm is the thickness of the intrinsic region in our EAM structure. Such maximum electrical field for GaAs based material reveals not only a good quality of the intrinsic region, but also a successful surface passivation avoiding surface recombination states.
4.2 Reflectance spectra
The reflectance spectra of hybridised EAM arrays were measured using a pigtailed superluminescent laser diode with an emission in the 820–865 nm range as the optical source. Fiber collimators were used to illuminate the array with a beam spot of 1mm in diameter and to couple the light back to a fiber and send it to an optical spectrum analyser for detection.
Several 2×64 and 2×128 arrays were selected randomly at different positions including close to the wafer edge to compare their optical properties. The reflectance spectra were measured for the arrays, and a typical spectrum is shown in Fig. 5 where the FP cavity is located at 856 nm and QW exciton is located at 847 without bias. The QW exicton wavelength was the same for all measured arrays, however their FP dip wavelength varied slightly for different chips, and across single chip due to the FP cavity resonance sensitivity to the epitaxy growth accuracy. To investigate the uniformity of the FP in the arrays, the reflectance measurements were carried out carefully at positions along or perpendicular to the pixel row with a 0.5 mm step across the active region of the arrays (5 mm) with the beam spot size of 0.5 mm. The insert of the Fig. 5 shows the FP uniformity across the 2×128 (A) and 2×64 (B1 and B2) EAM arrays. The results illustrate that the variation of the FP wavelength across the chip is less than 1.5 nm for all measured arrays in both parallel and perpendicular directions. Such good uniformity ensures uniform optical performance when the array is implemented into the optical system. In particular, the good uniformity even for the chips from the wafer edge is a indication that this technology is suitable for volume production.
4.3 Electro-optical response
The reflectance spectra of both types of arrays were measured under different reverse biases from 0 V to 15 V with a 0.2 V step using the set-up described above. The reflectance spectra were scanned with 0.04 nm steps. The contrast ratio (CR) was obtained from reflectance measurements, when reflectance at zero bias is defined as on-state (Ron) and the reflectance under bias as off-state (Roff) as follows:
Figure 6 shows measured CR results for a 2×64 array. With increasing bias voltage the reflectance at the FP cavity position (856.1 nm) decreased due to the red shift of the exciton of the QWs based on the QCSE, and the CR increased from 87:1 to 130:1 with a voltage swing from 0–4.8 V to 0–5 V. However the CR decreased when the voltage swing increased over 5 V, and a CR of 48:1 was observed at 0–5.2 V. It is worth pointing out that the CR at 0–6 V has a wide spectral bandwidth of 5.5 nm for CR >5:1 as shown in the insert of Fig. 6. Especially when the on-state and off-state were defined at bias of 5 V and 6 V, respectively, CR is 20:1 with only 1 V of voltage swing. The low driving voltage swing enables the device operating with low power consumption and high thermal stability.
Figure 6 also shows CR between 853 nm to 855 nm under a voltage swing from 0–9 V to 0–11 V and a maximum CR of 335:1 at 853.8 nm for a 0–10 V voltage swing. The high contrast ratio appears at a shorter wavelength range than the FP cavity at zero bias due to the blue shift of the FP cavity at high bias (cavity detuning). With further increased electrical field, the absorption of the QWs is red shifted to its absorption edge, which causes refractive index of the QWs to change dramatically and modified the optical cavity length , thereby blue shifting its FP resonance from the position at zero bias. Thus, high contrast ratios were obtained under bias over 6 V at wavelength range 853 to 855 nm.
Figure 7 shows reflectance changes with applied reverse bias at 853.8 nm and 856.1 nm, respectively. It is clear that the reflectance reaches almost zero at 10 V at 853.8 nm and at 5 V at 856.1 nm, where maximum contrast occurs. Desired contrast ratios can be obtained at specific voltage swings. For example, at 856.1 nm the reflectance changes from 21% to 0.2% with increasing bias from 2 V to 5 V, therefore contrast ratio of 105:1 can be obtained with 3 V of voltage swing. In fact, Fig. 7 shows almost linear response between 1 V to 4 V at 856.1 nm and between 2 V to 5 V at 853.8 nm, which is desirable for analogue applications.
The dynamic electro-optic response of both types the EAM arrays was characterised. The measurements were performed at the operating wavelength giving the maximum contrast ratio. The array was reverse biased with a periodic sine wave as electrical input using a signal generator with frequency up to 20 MHz. Figure 8 shows clear the intensity modulation of a 2 x 64 EAM array at 15 MHz. In fact this modulation speed is lower than its electrical cut-off frequency, which is determined by the RC constant of the EAM arrays as follows:
Where R is about 50 Ω that is the device serial resistance, and C is about 20 pF and 10 pF for the modulators in 2×64 array and 2×128 array, respectively. Therefore their electrical cut-off frequency is approximate to 150 MHz and 300 MHz for the two array types. This low optical response is essentially limited by the signal generator used in this experiment.
We have presented high-speed, large arrays of vertical-cavity electroabsorption modulators with high contrast ratios at low driving voltages to independently control the intensity of many parallel incoming signals. The EAM array fabrication, hybridisation to the electronic connector and electrical and optical characterisations have been described. High yield wafer-scale processing has demonstrated a potential of volume production for these high quality EAM arrays.
This work was partly financed by the EU IST-2001-37435 project. The authors would like to thank the Acreo MOVPE group for the epitaxy growth, as well as Sylvie Tonda and Stéphane Formont from Thales for fruitful discussions on the parallel broadband signal processing system.
References and links
1. D. T. Carrott, G. L. Mallaley, R. B. Dydyk, and S. A. Mills, “Third generation miniature ruggedized optical correlators (MROC) module,” in Optical Pattern Recognition IX, David P. Casasent and Tien-Hsin Chao, eds., Proc. SPIE 3386, 38–44 (1998). [CrossRef]
2. D. A. B. Miller, D. S. Chemla, T. C. Damen, A. C. Gossard, W. Wiegmann, T. H. Wood, and C. A. Burrus, “Bandedge electro-absorption in quantum well structures: the quantum confined Stark effect,” Phys. Rev. Lett , 53, 2173–2177 (1984). [CrossRef]
3. U. Efron and G. Liverscu, “Multiple Quantum Well Spatial Light Modulators,” in Spatial Light Modulator Technology: materials, devices and applications, Uzi Efron, ed. (Marcel Dekker, 1995), pp. 217–286.
5. Stéphane Junique, Qin Wang, Susanne Almqvist, JuanHua Guo, Henk Martijn, Bertrand Noharet, and Jan Y. Andersson, “GaAs-based multiple quantum well spatial light modulators fabricated by a wafer-scale process,” Appl. Opt. 44, No.9, 1635–1641 (2005). [CrossRef] [PubMed]
6. Uriel Arad, Eddie Redmard, Moshe Shamay, Arkadi Averboukh, Shimon Levit, and Uzi Efron, “Development of a large high-performance 2-D array of GaAs/AlGaAs multiple quantum-well modulators,” IEEE Photon. Technol. Lett. 15, 1531–1533 (2003). [CrossRef]
7. Hong Liu, Chien-Chung Lin, and James S. Harris Jr., “High-speed, dual-function vertical cavity multiple quantum well modulators and photodetectors for optical interconnects,” Opt. Eng. 40, 1186–1191 (2001). [CrossRef]
8. G. C. Gilbreath, W. S Rabinovich, T. J. Meehan, M. J. Vilcheck, R. Mahon, R. Burris, M. Ferraro, I. Solkolsky, J. A. Vasquez, C. S. Bovais, K. Cochrell, K. C. Goins, R. Barbehenn, D. S. Katzer, K. Ikossi-Anastasiou, and M. J. Montes, “Large-aperture multiple quantum well modulating retroreflector for free-space optical data transfer on unmanned aerial vehicles,” Opt. Eng. 40, 1348–1356 (2001). [CrossRef]
9. Qin Wang, Stéphane Junique, Daniel Ågren, Bertrand Noharet, and Jan Y. Andersson, “Fabry-Perot electroabsorption modulators for high-speed free-space optical communication,” IEEE Photon. Technol. Lett. 16, 1471–1473 (2004). [CrossRef]
10. D. Dolfi, J. Tabourel, Q. Durand, V. Laude, J. P. Huignard, and J. Chazelas, “Optical architectures for programmable filtering and correlation of microwave signals,” IEEE Trans. Microwave Theory Tech. 45, 1467–1471 (1997). [CrossRef]
11. H. Malm, C. Asplund, S. Becanovic, J. Borgling, A. Parekh, and B. Hirschauer, “Advanced process control for high quality R&D and production of MOVPE material by RealTemp,” J. Cryst. Growth 248, 229–234 (2003). [CrossRef]
12. G. E. Ponchak, “PIN diodes,” in GaAs MMIC Reliability Assurance Guideline for Space Applications, chapter 6 (1996), http://parts.jpl.nasa.gov/mmic/contents.htm.
13. J. A. Trezza and J. S. Harris Jr., “Creation and optimization of vertical cavity phase flip modulators,” J. Appl. Phys. 75, 4878–4884 (1994) [CrossRef]