A HIGH-BIT-RATE OPTICAL TESTBED FOR MULTI-BEAM ULTRAFAST SWITCHING USING AN ASYMMETRIC FABRY-PÉROT ALL-OPTICAL DEVICE

Transcription

1 A HIGH-BIT-RATE OPTICAL TESTBED FOR MULTI-BEAM ULTRAFAST SWITCHING USING AN ASYMMETRIC FABRY-PÉROT ALL-OPTICAL DEVICE by Darren Tze-Ray Wu A thesis submitted in conformity with the requirements for the degree of Master of Applied Science Graduate Department of Electrical and Computer Engineering University of Toronto c Copyright by Darren Tze-Ray Wu, 2007

2 A High-Bit-Rate Optical Testbed for Multi-Beam Ultrafast Switching Using an Asymmetric Fabry-Pérot All-Optical Device Darren Tze-Ray Wu Master of Applied Science, 2007 Graduate Department of Electrical and Computer Engineering University of Toronto Abstract A multi-beam system was developed for an Asymmetric Fabry-Pérot all-optical switch to demonstrate the multiple channel functionality of the switch and to show its ability to perform in high speed optical networks. Two channels were simultaneously switched using one optical control pulse. Short recovery times as low as 6.8 ps were observed, and channel to channel crosstalk was measured to be less than -50 db. The bandwidth is large - from 10 nm to >40 nm depending on the application. The design and construction of an 80 Gbit/s optical communication testbed is also presented. The testbed provides a high energy, short pulse width optical clock which gives it the capability of performing system demonstrations of ultrafast all-optical switches. A 10 Gbit/s wavelength conversion demonstration is performed on the Asymmetric Fabry-Perot all-optical switch using the testbed; the first high speed demonstration of this particular all-optical switch. ii

3 iii

4 Acknowledgments The work presented in this thesis would not have been possible without the valuable help and guidance of friends, family, and teachers. First of all, thanks to my supervisors Professor Peter W.E. Smith and Professor Li Qian for their advice, direction, and encouragement. Their continuous mentorship and thorough consideration of my work is greatly appreciated. Special thanks to Professor Li Qian for helping me in the laboratory at all hours of the day. I would also like to acknowledge and give thanks to Professor Eric G. Johnson for allowing me to use his cleanroom facilities, and to his student Pradeep Srinivasan who worked tirelessly with me to fabricate the collimating lens array. I am grateful to my colleagues Eric Lau and Phillip Tong for their hard work and perseverance in the design and building of the custom EDFA used in the Optical Control Pulse block of the testbed. Thanks also goes out to Aaron Zilkie for his insights in experimental methods and other useful tidbits. Professor Amr Helmy has also been very generous with his loan to me of the Agilent 86100C DCA-J for which I am very thankful. I owe a big debt of gratitude to Dr. Waleed Mohammed for his invaluable help and friendship throughout my thesis. Without his assistance, the collimating lens array might never have been made. His expertise in optics and modelling has been very helpful, especially in the analysis of the Fresnel lens and in the production of the amplitude masks for the additive lithography process. To my family, I thank them for always being there for me, and always looking out for me no matter the circumstance. To Jen for her support, her patience, and her love. iv

5 v

6 vi

7 Table of Contents List of Figures List of Tables List of Acronyms xv xvii xviii 1 Introduction Types of All-Optical Switches Fiber All-Optical Switches Semiconductor Optical Amplifier All-Optical Switches Reflective All-Optical Switches Asymmetric Fabry-Pérot Switch Proposed Uses of the AFP Switch Testbed for All-Optical Switches Thesis Outline Multi-Beam Delivery System for AFP Switch Design of the Multi-Beam System Configuration of the Optics Specifications of the Focusing Lens Specifications of the Collimating Lens Array Specifications of the Fiber Array Fabrication of Collimating Lens Array Introduction to Additive Lithography Process Efficiency of a Quantized Fresnel Lens vii

8 viii Table of Contents Fabrication Step One: Characterizing the Photoresist Fabrication Step Two: Patterning the Photoresist Fabrication Step Three: Transfer the Pattern from Photoresist to Substrate Characterization of the Lens Array and Integration into Multi-Beam System Characterizing the Fresnel lenses in the array Aligning the Fiber Array and the Lens Array Alignment of the Focusing Lens and Multi-Beam System Performance Demonstration of the Multi-Beam System with the AFP Switch Multi-channel operation of the AFP switch Crosstalk Measurements Wavelength Conversion Demonstration Testbed for High Speed All-Optical Switches Testbed Design Criteria and Outline Testbed Design Criteria Overview of the Testbed Optical Data Channel Generation Design of the Optical Data Channel Block Signal Quality in the Optical Data Channel Block Optical Control Pulse Generation Energy Requirements for AFP Switch Creating a Burst of Pulses High Power EDFA Wavelength Conversion Demonstration Conclusions Significance and Contribution Future Work A Structure of the Asymmetric Fabry-Pérot Switch 95

9 Table of Contents ix A.1 Comparison With Other AFP Switches A.1.1 Optical Nonlinearity of the Nonlinear Medium A.1.2 Contrast Ratio, Carrier Recovery Time, and Bandwidth of the Switches. 97 A.1.3 Polarization Dependence B Components and Layout of Testbed 101 B.1 Wiring of electronic components B.2 List of Components References 105

10 x Table of Contents

11 List of Figures 1.1 Evolution of network bit-rates. Horizontal dotted grey lines at 10 Gbps and 40 Gbps represent capacity achieved by TDM alone. Above these dashed lines, WDM and TDM are needed together to provide the necessary capacity. ( After Ref.[7] ) Schematic illustrating the operation of an all-optical switch Estimated evolution of optical networks from static point-to-point links into a dynamically reconfigurable network. As point-to-point connections (Lambda Timescales) can be established and torn down faster, network capacity is estimated to increase accordingly. Such networks will require more dynamic add/drop multiplexers as well as more flexible and automated routing capabilities. (After Ref.[8]) Example of an all-optical switch based on a Mach-Zehnder interferometer using SOAs as the nonlinear medium. (After Ref.[21]) Schematic illustrating the operation of a reflective all-optical switch Diagram of Asymmetric Fabry Perot all-optical switch For low intensity light, the AFP switch is in an OFF state For high intensity light the AFP is in an ON state The AFP switch can be used to switch multiple signals all at once The AFP switch can be used at many different wavelengths. (After Ref.[33]) The AFP switch as a multiple wavelength converter The AFP switch can be used as optical boolean logic gates. The truth table for each type of gate is shown next to the physical setup needed to realize to the gate Schematic of testbed, and all-optical switch under test A honeycomb or hexagonal arrangement packs the most beams into a fixed space General configuration of the multi-beam system xi

12 xii List of Figures 2.3 With the multi-beam system, the AFP switch can be easily reconfigured by changing the focusing lens(es) Adding circulators can double the number of signal inputs capable of being simultaneously switched by one control pulse. Circulators are represented by circles with arrows inside them. Port numbers are labeled on the top most circulator. The thick black lines represent optical fiber, and the arrows indicate the direction of propagation Multi-beam system configuration that was built To focus a larger number of collimated beams of diameter D onto the AFP switch, a bigger lens must be used. For the same f-number, a larger clear aperture implies a larger focal length, and hence a larger focal spot size Example of replication method used to imprint lenses on a substrate. After [40] Basic concept of a Fresnel Lens Pitch of collimating lens array Photo of fiber array from Moritex Corp. A ribbon of single mode optical fibers is glued inside a quartz chip with v-grooves. After [41] Detailed drawing of the 32 fiber array from Moritex Corporation. All dimensions are in mm. After [41] With the additive lithography technique, an 8-level pattern can be created with 3 masks exposed one after another Analysis of Fresnel lens diffraction efficiency in cylindrical coordinates Electric field magnitude at the focal plane of the Fresnel lens A lens array was made taking up an area roughly 2 cm by 4 cm in the middle of the wafer Plot of resist removed after being developed vs. exposure time to mercury lamp Schematic of direct wafer stepper patterning chamber Determining the exposure time for the three masks Depth profile of Fresnel lens pattern in photoresist measured with a Zygo optical profiler

13 List of Figures xiii 2.20 Schematic of ICP RIE chamber. The electric field RF 1 creates a plasma of reactive ions used in the etching, while the electric field RF 2 biases the substrate. After [45] Depth profile of Fresnel lens in the quartz substrate measured with a Zygo optical profiler Microscope images of a Fresnel lens in the collimating lens array Setup to measure transmittance of the lens array Measured transmittance of the lens array Experimental setup to measure focal length of Fresnel lenses Positioning of fiber array and lens array Vidicon images combined to show collimated beam propagation Integration of the focusing lens with the fiber and collimating lens array completes the multi-beam system Beam profile of Fiber A, Fiber B and Control at the focal plane of the focusing lens Measured insertion loss of the multi-beam system Laser system used to demonstrate AFP switch Experimental setup for multiple channel demonstration of AFP switch. Circulator ports are numbered Pump probe trace for Data Channel 1 and Data Channel 2. The contrast ratio (CR) and recovery time of the AFP switch are indicated in the legend Pump probe trace of Data Channel 1 at a signal wavelength of 1576 nm Contrast ratio of AFP switch measured at different signal wavelengths. Pump wavelength varies from 1536 nm to 1540 nm for a signal wavelength that varies from 1562 nm to 1587 nm as shown Spectral profile of pulses at the output port of Data Channel 1. Signal pulses are at 1578 nm wavelength Example of simultaneous multiple channel wavelength conversion using the multibeam system with AFP switch Experimental setup for single channel wavelength conversion

14 xiv List of Figures 3.9 Results of wavelength conversion experiment using multi-beam system. In all plots, the horizontal scale is 50 ps/div Outline of testbed functions The testbed can be used to demonstrate demultiplexing The testbed configuration for the AFP switch and multi-beam system. Solid lines indicate single mode optical fiber, while dashed lines indicate electronic coaxial cable. Similarly solid boxes indicate components with optical input/output while dashed boxes indicate components with electronic input/output. This convention will be used in all further figures showing the testbed layout Schematic of the testbed with all components shown. Darkly shaded boxes indicate components used in the Optical Control Pulse block. Lightly shaded boxes indicate components used in the Optical Data Channel block Components used in Optical Data Channel block of Figure 4.3 are shaded in grey. Bit rates at various points in the block are indicated A MZM is used to encode electronic data onto a 10 GHz optical pulse train Schematic of a 2X optical clock multiplier Optical multiplication and preservation of the PRBS pattern length using a pattern as an example Gbit/s, length, electronic output of PRBS generator Optical output of MZM as it modulates a CW laser Optical output of MZM, i.e. 10 Gbit/s in Figure Pulses from the MLFL are convolved with the frequency response of the DCA sampling module to yield 7.6 ps FWHM pulses. Vertical scale is 5.00 mw/div. Horizontal scale is 5.0 ps/div Eye diagram of the 40 Gbit/s optical output of the Optical Clock Multiplier Eye diagram of the 80 Gbit/s point in the Optical Data Channel block Components used in Optical Control Pulse block of Figure 4.3 are shaded in grey. 82

15 List of Figures xv 4.16 A slow rise/fall time in the burst control signal to MZM control results in pulses at the beginning and end of the burst being attenuated. Grey triangles represent optical pulses from the MLFL. Dashed lines indicate the burst voltage Electronic output of Burst Generator The burst of 10 GHz pulses from MZM The output power of the high power EDFA for a CW input at 1550 nm at various input powers The last 8 pulses of a 32 pulse burst at the EDFA output. Vertical scale is 5.00 mw/div. Horizontal scale is ps/div Experimental setup for 10 Gbit/s wavelength conversion demonstration. The Optical Data Channel block of the testbed is not needed in this experiment Experimental setup for 10 Gbit/s wavelength conversion demonstration. Data at a carrier wavelength of 1550 nm is converted to data at 1572 nm. The Optical Data Channel block of the testbed is not needed in this experiment. Vertical scale is 665 µw/div. Horizontal scale is 50.0 ps/div A.1 Physical structure of the AFP switch. Drawing is not to scale, and not all the layers in the Bragg reflection stack are shown B.1 Arrangement of electronic components in the testbed

16 xvi List of Figures

17 List of Tables 2.1 Focusing lens specifications Fiber array specifications Etch Recipe Crosstalk Measurements B.1 Components in Optical Data Channel block B.2 Components in Optical Control Pulse block B.3 Components in rest of testbed xvii

18 List of Acronyms AFP CW DCA EDFA FC/APC FWHM MLFL MZM ODL OOK OTDM PRBS TDM TLS VOA Asymmetric Fabry-Perot Continuous Wave Digital Communications Analyzer Erbium-Doped Fiber Amplifier Ferrule Ceramic / Angled Physical Contact Full Width at Half Maximum Mode-Locked Fiber Laser Mach-Zehnder Modulator Optical Delay Line On-Off Keying Optical Time-Division Multiplexing Pseudo-Random Bit Sequence Time-Division Multiplexing Tunable Laser Source Variable Optical Attenuator xviii

19 Chapter 1 Introduction Fiber-optic communications technology has enabled us to build high speed and long distance networks that have greatly enriched our lives. These networks have been embraced by the global community and usage continues to grow year after year as more users are added and more services are offered. As a rough measure of the global appetite for network bandwidth, Internet backbone traffic was measured to almost double every year in the period from 1997 to 2004 [1, 2]. To meet future demands for ever higher bit-rate networks, we are interested in technologies that will allow us to transmit more data over our current network infrastructure, thereby making more efficient use of the available bandwidth of optical fiber. One solution that leads to higher network data rates is time-division multiplexing (TDM) where low bitrate channels are aggregated together in different time slots to form a single high bit-rate channel. Another solution is wavelength-division multiplexing (WDM) where many low bitrates channels are transmitted simultaneously by giving each data channel its own carrier wavelength. In practice, service providers have found that a combination of both TDM and WDM techniques provide the optimum solution for increasing network capacity. Current commercial backbone networks employ TDM to combine multiple data channels from access networks, typically into a single serial channel at 2.5 or 10 Gbit/s. A number of 2.5 or 10 1

20 2 Chapter 1. Introduction Gbit/s channels are then multiplexed using WDM and then transmitted through an optical fiber. There is a clear trend that the industry is moving to higher TDM bit-rates. Commercial serial 40 Gbit/s systems have been on the market since 2002 and are mostly deployed in backbone networks although development is underway to make them affordable for metro networks as well [3, 4]. This trend to higher TDM bit-rates is also evident in research systems. Early research systems capable of very high bit-rate transmission often used low channel bit-rates, and dense WDM such as the demonstration in [5] where 373 channels each at 10 Gbit/s were used. The latest research systems tend to use higher bit-rate channels such as the recent demonstration of 25.6 Tbit/s transmission of 160 WDM channels with each containing a pair of 85.4 Gbit/s signals [6]. Although much higher serial bit-rates than 85.4 Gbit/s have been achieved, the increasing effect of dispersion at such high speeds tends to make these systems unmanageable and the general trend is to keep channel rates below 160 Gbit/s [3]. As an illustration, Figure 1.1 shows the growth in the bit-rate capable of being transmitted through a fiber as a result of increasing serial channel bit-rates through TDM, and through combination of multiple serial channels through WDM. Ultimately, network designers want to maximize network bandwidth. One way to achieve this is to use TDM and WDM techniques to make more efficient use of fiber bandwidth between connection points in a network. In conjunction with TDM and WDM, making more efficient use of network connections can increase network bandwidth. In other words, a network can transmit more data by making sure that all the connections in the network are transmitting data all the time, and that none of the connections are laying idle. Network connections can be utilized more efficiently by using dynamic network architectures. Current networks are largely circuit switched with the links often being static, meaning that point-to-point connections are setup for long periods of time and can take hours or even days to reconfigure [7, 8]. At various nodes in these static circuits, packet routers are used to route traffic to their end destination and to ensure efficient network utilization. These packet routers are opaque meaning that they convert the packets they

21 3 Figure 1.1: Evolution of network bit-rates. Horizontal dotted grey lines at 10 Gbps and 40 Gbps represent capacity achieved by TDM alone. Above these dashed lines, WDM and TDM are needed together to provide the necessary capacity. ( After Ref.[7] ) receive from the optical to electrical domain in order to perform their routing, and then convert back into the optical domain for transmission. Based on the past growth rates of doubling Internet traffic every year, there is concern that packet routers will not be able to scale in a cost effective manner to meet the increasing volumes of traffic [9, 10]. In search for more scalable, high bit-rate technologies, dynamic network architectures (also known as reconfigurable or automatically switched network architectures) have been proposed as an alternative to the continued scaling of static networks. There are three types of dynamic network architectures. They are listed in order of how quickly network connections can be reconfigured, from fastest to slowest. The more quickly a network can be reconfigured, the more data it can transmit. 1. Optical packet switching networks 2. Optical burst switching networks 3. Wavelength selective routing networks

22 4 Chapter 1. Introduction All-optical switches will be an important component in future dynamic networks, especially in networks where connections must be reconfigured quickly such as in optical packet switching networks. This is because they can perform a number of needed functions in dynamic networks such as wavelength conversion, demultiplexing, packet label processing, and logic operations at high speeds. Because ultrafast all-optical switches have the capability to perform these functions at the serial channel bit rate, they enable future dynamic networks to reconfigure connections quickly, and thus route traffic more efficiently through the network. An all-optical switch is a switch where one beam of light controls the transmission of another beam of light. This is shown schematically in Figure 1.2 where a Control beam of light turns an all-optical switch ON and OFF. This controls which Data pulses are transmitted through the switch, and which ones are blocked by the switch. All-optical switches can be operated at very high speeds because they utilize ultrafast nonlinear effects as a switching mechanism. Figure 1.2: Schematic illustrating the operation of an all-optical switch. Networks utilizing all-optical switches may have advantages such as lower latency, due to fewer conversions between the optical and electronic domain, which would be especially important in real-time applications. All-optically switched networks could also have simpler routing tables since they can be developed to take advantage of modern routing protocols and packet-based switching. An estimated prediction of the evolution of future metropolitan optical networks into optically packet switched networks is shown in Figure 1.3. As optical networks mature and develop into optical packet switched networks, all-optical switches will play an increasingly important and essential role.

23 1.1. Types of All-Optical Switches 5 Figure 1.3: Estimated evolution of optical networks from static point-to-point links into a dynamically reconfigurable network. As point-to-point connections (Lambda Timescales) can be established and torn down faster, network capacity is estimated to increase accordingly. Such networks will require more dynamic add/drop multiplexers as well as more flexible and automated routing capabilities. (After Ref.[8]) The development of fast all-optical switches will be an important part of realizing future dynamic networks. In addition to being very fast, all-optical switches should also operate over large wavelength ranges so that they can be used in networks utilizing WDM Types of All-Optical Switches All-optical switches can be grouped by the nonlinear material that they use as their switching medium. They can be further classified by the physical structure of the switch Fiber All-Optical Switches Optical fiber itself can be a nonlinear medium, and all-optical switches using regular fiber can offer subpicosecond recovery times. Unfortunately, the nonlinearity in optical fiber is very

24 6 Chapter 1. Introduction weak, and fiber based all-optical switches require large lengths of fiber and high switching powers to be effective. By using highly nonlinear fibers and modifying the physical structure to increase the peak intensity, shorter fibers can be used, but are still typically longer than 1 m [11, 12, 13] Semiconductor Optical Amplifier All-Optical Switches To make more compact devices, semiconductor based all-optical switches were developed because of their high nonlinearity. Many of these switches use semiconductor optical amplifiers (SOA) as their nonlinear medium [14, 15, 16, 17]. Using a material with a higher nonlinearity has the advantage of requiring less optical power for the control pulse. All-optical switches using SOAs are usually interferometers, such as the Mach-Zehnder interferometer (MZI) shown in Figure 1.4 or Sagnac interferometers [18, 19, 20]. Figure 1.4: Example of an all-optical switch based on a Mach-Zehnder interferometer using SOAs as the nonlinear medium. (After Ref.[21]) In general, SOA-based all-optical switches need to have an interferometer structure since the recovery time of the SOA can be long. Without the interferometric structure, the long carrier recovery time of the SOA limits the speed at which the all-optical switch can be

25 1.1. Types of All-Optical Switches 7 turned off. Using an interferometric structure allows the all-optical switch to be turned off very quickly, but can limit the repetition rate at which an SOA-based all-optical switch can repeat the switching operation. The limited repetition rate of SOA-based all-optical switches can be overcome using novel techniques. For instance, in [22] a polarization switching scheme using a nonlinear birefringence is demonstrated, instead of a more conventional nonlinear phase modulation as the switching mechanism. Using polarization switching, recovery times as short as 580 fs were achieved which would allow the SOA-based switch to operate at high repetition rates. However, this increase in speed comes at the cost of a low switching contrast ratio of 5.2 db, and the use of many optics to properly control the polarization of the beams in the demonstration Reflective All-Optical Switches Another type of switch that can be used is a reflective switch which uses one of the following as the nonlinear medium: 1. Reflective switches using a bulk GaAs material as the nonlinear medium [23]. The bulk GaAs is grown in a low-temperature MBE process which results in a reduced carrier recovery time [24, 25, 26]. This nonlinear medium will herein be referred to as low-temperature GaAs. 2. Reflective switches using a multiple quantum well (MQW) structure made of lowtemperature MBE grown III-V semiconductor [27, 28, 29, 30, 31, 32] as the nonlinear medium. This nonlinear medium will herein be referred to as low-temperature MQW. 3. Reflective switches using a bulk InGaAsP nonlinear medium [33], grown using helium plasma assisted molecular beam epitaxy (MBE) [34, 35, 36]. This material will herein be referred to as helium plasma InGaAsP. Reflective switches are slightly different than conventional all-optical switches in that the switched output beam is a reflection from the switch, rather than a transmission through the switch. This principle is shown in Figure 1.5

26 8 Chapter 1. Introduction Figure 1.5: Schematic illustrating the operation of a reflective all-optical switch. In low-temperature GaAs, low-temperature MQW, and helium plasma InGaAsP, picosecond recovery lifetimes of the carriers are achieved by introducing defects into the semiconductor material through the growing and doping process. These carrier recovery lifetimes are much shorter than in conventionally grown III-V materials. This fast carrier recovery allows these materials to be used in all-optical switches operating at high repetition rates. All the aforementioned ultrafast materials also exhibit large nonlinearities relative to optical fiber which makes them suitable for compact all-optical switching devices. Reflective all-optical switches incorporating low-temperature GaAs were developed and carrier recovery times as short as 3 ps were demonstrated. However, these switches normally operate at wavelengths around 850 nm, and cannot be used at telecom wavelengths around 1550 nm. Reflective all-optical switches using low-temperature MQWs exhibit polarization dependence that is not as evident as in bulk material [37]. This polarization dependence can be used to achieve fast switching with impressive results demonstrated in [38]. However, use of the polarization dependence comes at a high cost; complex optics are needed for the delivery of control and data pulses to the switch since their polarization must be set to a certain state. To develop a simpler and more robust reflective switch with equivalent performance, work was done by Qian et al. [33] to demonstrate an all-optical switch using helium plasma InGaAsP as its nonlinear medium. The reflective switch developed by Qian et al. uses an

27 1.1. Types of All-Optical Switches 9 asymmetric Fabry-Pérot structure, and will herein be referred to as the AFP (Asymmetric Fabry-Pérot) switch. The use of InGaAsP as the nonlinear medium in the AFP switch allows the device to operate at wavelengths around 1550 nm. Because the AFP switch uses bulk semiconductor instead of low temperature MQW as its nonlinear medium, it exhibits less polarization dependence. As a result, fewer optics are needed to deliver beams to and from the AFP switch since the polarization of the beams does not need to be controlled. This results in a simpler and more robust switching package. Furthermore, it is of practical interest to avoid polarization dependent devices in optical networks since it is very costly to control the polarization of light as it propagates through optical fiber Asymmetric Fabry-Pérot Switch The AFP switch developed by Qian et al. is shown in Figure 1.6. The structure of the entire AFP switch is very simple; it is an asymmetric Fabry-Pérot etalon composed of a low reflectance top layer and a high reflectance bottom layer. Sandwiched in the etalon is the helium plasma InGaAsP which is used as a saturable absorber. Helium plasma InGaAsP has a carrier recovery time of a few picoseconds and this enables the AFP switch to be switched ON and OFF at repetition rates in the tens of GHz range. The saturable absorber in the AFP structure will herein be referred to as the active layer. Figure 1.6: Diagram of Asymmetric Fabry Perot all-optical switch.

28 10 Chapter 1. Introduction For low intensity incident light, the AFP switch has been designed such that the output reflections off the top low reflectance layer and the bottom high reflectance layer will interfere destructively as shown in Figure 1.7. The amplitudes of the two reflections will be equal due to the absorption of the active layer. The reflection off the bottom layer will be out of phase due to the extra path length traveled. For low intensity light, the switch is an OFF state. Figure 1.7: For low intensity light, the AFP switch is in an OFF state. For high intensity incident light, the absorption can be saturated in the active layer as shown in Figure 1.8. In this case, the AFP switch is in an ON state. With the absorption of the active layer reduced, the reflection off the bottom layer of the etalon will be much stronger than the reflection of the top layer, and the net reflection from the switch s surface will be much higher compared to the OFF state. Figure 1.8: For high intensity light the AFP is in an ON state. The absorption in the active layer can be saturated by the incident beam of light itself, or a separate beam can be used to saturate the absorption. By using a beam of high intensity pulses as the control pulses, and a beam of low intensity pulses as the signal pulses, an all-

29 1.1. Types of All-Optical Switches 11 optical switch can be realized. It is necessary to keep the signal pulses at a low intensity to prevent the signal pulses themselves from turning the AFP switch ON (self-switching). It is possible to configure the all-optical switch to simultaneously perform switching on multiple signal beams as shown in Figure 1.9. This is accomplished by spatially separating the signal pulses and varying their angle of incidence to the switch. Figure 1.9: The AFP switch can be used to switch multiple signals all at once. More information about the AFP switch used for these studies can be found in Appendix A. Previous demonstration of the AFP switch by Qian et al. showed a fast recovery from ON to OFF state. A 5 ps recovery time was measured showing that the switch can operate at repetition rates as high as 80 GHz (recovery time measured at 1/e of peak). Previous experiments also demonstrated the large wavelength range over which the AFP switch can be used as shown in Figure 1.10, illustrating its usefulness in networks utilizing WDM.

30 12 Chapter 1. Introduction Figure 1.10: The AFP switch can be used at many different wavelengths. (After Ref.[33]) 1.2. Proposed Uses of the AFP Switch One of the main advantages of the AFP switch is its ability to simultaneously process multiple inputs and produce multiple outputs. For instance, in Figure 1.9 the AFP switch could be used in this configuration as a multiple channel demultiplexer, where Input Signals 1-3 are high speed TDM channels in an optical network, and Outputs 1-3 are lower speed channels that have been demultiplexed out. Different Input beams could even have the same wavelength if the user desired since the Input and Output signals are spatially separated. This ability to simultaneously process multiple signals in multiple physical channels is not found in many other waveguide switches. Because of this unique feature, reflective switches can also be used as all-optical serial to parallel converters which would be useful in packet switched networks for header processing, or other high speed all-optical signal processing requirements. Wavelength conversion is a necessary function for routing in optical networks, and can also be used for contention resolution of packets. The AFP switch can be used to simultaneously convert a high speed data channel at one wavelength into many other wavelengths as shown in Figure 1.11.

31 1.2. Proposed Uses of the AFP Switch 13 Figure 1.11: The AFP switch as a multiple wavelength converter. The AFP switch can also function as an optical logic gate for optical computing and signal processing purposes. When used as a demultiplexer, the AFP switch is acting as an AND boolean logic gate. Two incident beams to the AFP switch act as the inputs to the AND gate as shown in Figure 1.12(a). The output from the AND gate is the reflected beam from the AFP switch. A logic level of 0 corresponds to zero optical power in the beam. A logic level of 1 corresponds to a non-zero optical power level. In this configuration, a 1 in the reflected beam B 1 will only occur if both A 1 and A 2 are also 1. A 2 must be a high intensity beam in order to turn the AFP switch ON. A 1 should be of low enough intensity that it does not cause self-switching. (a) AFP as an AND gate (b) AFP as an OR gate Figure 1.12: The AFP switch can be used as optical boolean logic gates. The truth table for each type of gate is shown next to the physical setup needed to realize to the gate.

32 14 Chapter 1. Introduction The AFP switch can also be configured as an OR logic gate as demonstrated in Figure 1.12(b). In this case the beams A 1 and A 2 are of high enough intensity that each one alone can turn the AFP switch ON. A continuous wave (CW) laser, B i is incident on the device. When either A 1 or A 2 or both are set to 1, B 1 is also set to 1. In all cases, the actual area on the AFP switch where switching occurs can be quite small; essentially switching occurs only in the control beam diameter. In addition, the AFP switch is a bulk device without any fine structure and it can be grown to quite large dimensions. Given the large surface area that is available, multiple spots on a single device could be used. For optical logic gates, this could mean that one physical device could be used to realize multiple logic gates. In order to exploit the multiple input and multiple output functionality of the AFP switch, a system needs to be designed to deliver multiple beams to the switch, and collect multiple reflections from the switch. With a multiple beam delivery system, the AFP switch could be used for simultaneous demultiplexing of multiple channels, or for multiple channel wavelength conversion. The design of such a system will be discussed in Chapter Testbed for All-Optical Switches There is a large interest in all-optical switches and devices within the Photonics Group here at the University of Toronto, and a high bit-rate testbed that can be used to characterize a variety of different devices would be a useful long term experimental setup. Having a permanent testbed facility allows for rapid prototyping of all-optical devices since high speed characterization can be performed immediately after fabrication. Device behaviours that may only be evident at high speeds can be discovered early in the design process. With the testbed, all-optical switches can be demonstrated under experimental conditions that directly show how they might perform in an optical network. In anticipation of the current industry trend to higher bit-rates, a research testbed for future optical communication technologies should operate from 40 Gbit/s up to 160 Gbit/s. The testbed should also be able to deliver sufficiently high optical intensities for testing of all-optical devices. Finally, in the interest of making the testbed as a general facility for

33 1.3. Testbed for All-Optical Switches 15 testing of all-optical devices, it should be adaptable to different experiments and should be easy for new users to operate. A high level diagram of the testbed that will be designed for this thesis is shown in Figure The testbed will have two outputs. One output will be an Optical Control that is used to control the ON and OFF operation of an all-optical switch under test. The other output is an Optical Data Channel which is meant to simulate the optical data signals that are transmitted in commercial optical networks. The output from the all-optical switch, which is the switched Optical Data Channel, will be collected by the testbed. The testbed will have instruments to measure the signal quality of the output from the all-optical switch under test. Figure 1.13: Schematic of testbed, and all-optical switch under test.

34 16 Chapter 1. Introduction 1.4. Thesis Outline In chapter 2, we will show the design, fabrication, and characterization of a multiple beam delivery system for the AFP switch. In chapter 3, we describe the integration of the multiple beam delivery system with the AFP switch to form a multiple channel all-optical switching device. We present experimental results measuring the switching recovery time and bandwidth of operation of this integrated device. In chapter 4, we present the testbed design and characterization. The testbed was initially assembled into a configuration for testing of the integrated device made up of the multiple beam delivery system and the AFP switch. To show the capabilities of the testbed, we performed a high speed wavelength conversion with the integrated device. In chapter 5, we will conclude the thesis and the summarize the important results. We also discuss the significance of the work that was presented and suggest future directions. In this chapter, I will also describe my contribution to the work that was presented here.

35 Chapter 2 Multi-Beam Delivery System for AFP Switch The integration of a multi-beam delivery system with the AFP switch is a novel enhancement that increases functionality, and shows how the AFP switch can be packaged for commercial development. The following are the design criteria for the overall multi-beam system: 1. The multi-beam system should be flexible enough that the input and output configuration to the AFP switch can be changed with a minimal effort. 2. Commercially available components should be used to minimize the cost, complexity, and research into custom fabrication processes. Such commercially available, off-theshelf components will herein be referred to as stock components. 3. The least number of optical components should be used to achieve a robust package. 4. The multi-beam system should have fiber terminations so that it can be integrated into an optical network. In the first section of this chapter we describe the design of the multi-beam system for the AFP switch, including design criteria for the individual components of the multi-beam system. In the second section we explain the fabrication process for one of the components in the multi-beam system. In the third section we show our results from the optical characterization of the multi-beam system. 17

36 18 Chapter 2. Multi-Beam Delivery System for AFP Switch 2.1. Design of the Multi-Beam System Configuration of the Optics The main purpose of the multi-beam system is to deliver many beams to a single AFP switch to achieve many parallel switching operations. A honeycomb arrangement for the beams as shown in Figure 2.1 results in the highest beam density in the space around the AFP switch, thus making the multi-beam system as compact as possible. Figure 2.1: A honeycomb or hexagonal arrangement packs the most beams into a fixed space. In addition, the multi-beam system should be terminated with single mode optical fibers for integration into an optical network. The reflective AFP switch also requires that all beams be focused onto the same spot on the switch. The configuration of the multi-beam system was designed to meet these criteria. A sample configuration is shown in Figure 2.2.

37 2.1. Design of the Multi-Beam System 19 Figure 2.2: General configuration of the multi-beam system. Minor changes can be made to the setup shown in Figure 2.2 to expand or alter the functionality of the multi-beam system. More single mode fibers and collimating lenses can be added in the honeycomb arrangement to increase the number of channels delivered to and from the AFP switch. By changing the focusing lens, different functionality can be achieved. An example of this is shown in Figure 2.3. By switching from a larger focusing lens to two smaller ones, the system can be changed from a 6 beam system into two 3 beam systems. The beams are shaded in grey in Figure 2.3 to help illustrate this idea. Only a single row of fibers is shown in Figure 2.3 rather than the honeycomb arrangement to make the drawing more clear. (a) Configuration 1 (b) Configuration 2 Figure 2.3: With the multi-beam system, the AFP switch can be easily reconfigured by changing the focusing lens(es).

38 20 Chapter 2. Multi-Beam Delivery System for AFP Switch It is possible to further double the number of beams that can be delivered to the AFP switch with the multi-beam system by incorporating circulators into the fiber array as shown in Figure 2.4. By adding circulators, each fiber in the array can transmit one signal input to the AFP switch in the forward direction (left to right in the figure) and receive one output reflection from the switch in the backwards direction. This configuration of multi-beam system with circulators will herein be referred to as bi-directional operation. It is easy to add and remove circulators from the multi-beam system since they are discrete components that are fiber terminated. Figure 2.4: Adding circulators can double the number of signal inputs capable of being simultaneously switched by one control pulse. Circulators are represented by circles with arrows inside them. Port numbers are labeled on the top most circulator. The thick black lines represent optical fiber, and the arrows indicate the direction of propagation. The multi-beam system that we developed for this thesis was designed to facilitate the proof-of-concept demonstration of the AFP switch s ability to simultaneously switch multiple signals and to operate bi-directionally. The configuration for the multi-beam system was designed to be as simple as possible for this proof-of-concept demonstration and is shown in Figure 2.5. The multi-beam system will deliver two signal pulses, one through Fiber A and one through Fiber B, to the AFP switch and their reflections will be collected for measurement.

39 2.1. Design of the Multi-Beam System 21 Figure 2.5: Multi-beam system configuration that was built. In a network, optical data that is to be switched would be input into Fibers A and B. The central fiber will be used to deliver the control pulse to the AFP switch. A Crosstalk channel will be used to measure the crosstalk between fibers in the multi-beam system Specifications of the Focusing Lens To determine the optics needed for the multi-beam system, the power requirements of the AFP switch were examined. The main switching mechanism in the AFP switch is absorption saturation, and high contrast ratios between the ON and OFF state of the switch can be obtained by using a high intensity control pulse. High intensities can be obtained by using high energy control pulses, or by focusing the control pulse to a small spot on the AFP switch. To reduce the power consumption of the AFP switch, the focused spot size of the control pulse was minimized. The focused spot size on the AFP switch was calculated assuming a collimated beam is input to the focusing lens using the Gaussian beam formula [39]: 2W o = 4λf πd (2.1) Where D is the collimated beam diameter before the focusing lens, f is the focal length of the lens, λ is the wavelength, and 2W o is the focused beam diameter. Since a small focused spot size is desired, the focal length should be short and the

40 22 Chapter 2. Multi-Beam Delivery System for AFP Switch collimated beam diameter should be as large as possible. On the other hand, to achieve multiple channel operation, many collimated beams must be focused by the focusing lens. The more collimated beams that must fit into the clear aperture of the focusing lens, the smaller their diameter must be. Thus, filling the focusing lens with more beams results in a larger focused spot as shown in Figure 2.6. Figure 2.6: To focus a larger number of collimated beams of diameter D onto the AFP switch, a bigger lens must be used. For the same f-number, a larger clear aperture implies a larger focal length, and hence a larger focal spot size. To optimize the trade off between the number of beams, and the focused spot size, a focusing lens with a low f-number should be chosen (f-number, or F # = f ). Microscope D objectives and aspheric lenses typically have low F # and are suitable candidates for the focusing lens. In addition, these optics have been designed to correct for spherical aberration which will help minimize the focused spot size. Of the various stock lenses and microscope objectives that were available, an aspheric lens from Newport Corporation (KPA013) was selected as the optimal focusing lens for the multi-beam system primarily for its low F #. The specifications for this lens are shown in Table 2.1. In addition to its very low f-number, we chose this lens for its small diameter and long working distance. Preliminary research on lens arrays showed that most stock arrays had small lens pitches. A small diameter focusing lens was needed to match the lens pitch of these arrays, and the rationale for this will be discussed in more detail in the next subsection. A long working distance was needed for integration of the multi-beam system with the AFP switch. A long working distance in the focusing lens gives more flexibility in the choice

41 2.1. Design of the Multi-Beam System 23 Table 2.1: Focusing lens specifications Specification Focal Length (mm) 6.78 Clear Aperture (mm) 7.2 Diameter (mm) 9.0 Working Distance (mm) 4.08 F # 0.75 of opto-mechanics that can be used to mount the AFP switch, which must be placed at the focal plane of the focusing lens. For instance, we chose to mount the AFP switch in a gimbal mount since its angle relative to the direction of incident light is very important for good coupling of the reflection back into the fiber array. Since gimbal mounts often hold the optic in a recessed position, a long working distance helps to ensure that the AFP switch can be placed at the focal plane of the focusing lens Specifications of the Collimating Lens Array After the focusing lens was chosen, the specifications of the collimating lens array were determined. In order to obtain the smallest focal spot possible, it is necessary to collimate the output of the optical fiber into a beam to completely fill the clear aperture of the focusing lens. For the multi-beam system with three fibers that is being built for this thesis, the largest diameter that the collimated beam can be is 2.4 mm using the selected focusing lens. A survey of commercially available, stock lens arrays showed that most arrays designed for the 1550 nm wavelength had lens diameters under 500 µm and were unsuitable for the multibeam system. These lens arrays are typically used for optical interconnect applications and MEMS based optical switches. For these applications, compactness is an important factor which is why the individual lenses have such a small diameter. Unable to find a suitable stock lens array, custom lens arrays from commercial companies were considered. One method that is used to make lens arrays is the thermal reflow method. This is a type of lithography where a cylinder of photoresist is melted under a highly con-

42 24 Chapter 2. Multi-Beam Delivery System for AFP Switch trolled process into the shape of the lens that is to be etched into the substrate. This process is effective at making arrays with many refractive lenses. However, it is difficult to make lenses with a short enough focal length using this method because it is hard to achieve a large enough sag. Custom lens arrays from commercial companies are also made using the replication method. In this method, a liquid polymer is placed on a substrate, and a mold with the lens shape is lowered onto the polymer as shown in Figure 2.7. Once the polymer has been molded to the correct shape, UV light is used to harden the polymer to set it. Lens arrays using this method are commercially available and can produce lenses with the specifications required for the multi-beam system. However, making of the mold is a custom job that is quite expensive and beyond the budget of the thesis. Figure 2.7: Example of replication method used to imprint lenses on a substrate. After [40] Since a commercially available lens array could not be found that met the required lens diameter, f-number, and cost constraints of the the thesis, we decided to fabricate a custom lens array in collaboration with the University of Florida s College of Optics and Photonics (CREOL&FPCE). The lens array that was made was an array of Fresnel lenses. A Fresnel lens was chosen for it s low sag even for large diameter and short focal length lenses. The fabrication process used to make the lens array is known as additive lithography and will be discussed in detail in the next section. A Fresnel lens is a type of diffractive lens that mimics the beam shaping properties of a refractive lens. Neglecting the losses due to surface reflections, a refractive lens in Gaussian

43 2.1. Design of the Multi-Beam System 25 beam optics essentially just affects the phase of an incident beam of light. The amount of phase change depends on the thickness of the lens that the incident light encounters at that particular point on the lens. By changing the phase relationship as the incident wavefront passes through the lens, the shape of the exiting wavefront is changed. Since a refractive lens only changes the phase of the incident wavefront, portions of the lens that cause a 2π phase change can theoretically be removed without affecting the output wavefront from the lens. For a plane wave normally incident to the Fresnel lens at wavelength λ, the lens can be made thinner by: m λ (2.2) n lens 1 where m is an integer and n lens is the refractive index of the lens material. This concept is shown in Figure 2.8. Figure 2.8: Basic concept of a Fresnel Lens Due to the high cost of making a pattern mask for the additive lithography process, an existing pattern mask was reused. The Fresnel lens produced with this mask is 2 mm in diameter with a focal length of 6-7 mm. When used to collimate the beam from a single mode fiber, the collimated beam diameter is 0.95 mm at 13.5% of the peak intensity and 1.45 mm at 1%. Although a 0.95 mm beam diameter will not clearly fill the aperture of the focusing

44 26 Chapter 2. Multi-Beam Delivery System for AFP Switch lens, the diameter is large enough to yield a small spot when focused with the focusing lens - roughly 14 µm according to formula 2.1, which is acceptable. In addition, using a smaller beam diameter has the advantage of lower crosstalk in the multi-beam system. This is because there is very little overlap in space between adjacent collimated beams. One of the advantages of the additive lithography process is that Fresnel lenses can be fabricated immediately next to one another, in any arbitrary pattern. Lenses were arranged in a honeycomb pattern in the array, as shown in Figure 2.9 for compactness. The pitch of the lenses in the array is 2 mm. Figure 2.9: Pitch of collimating lens array. Although only a linear array of four lenses (indicated in dark grey in Figure 2.9) will be used in the proof-of-concept multi-beam system, a larger lens array was fabricated. This is so that the components in the multi-beam system would be ready to handle more channels for future experiments Specifications of the Fiber Array One of the most common methods to produce stock fiber arrays is to simply etch a series of v-grooves into a silicon or quartz substrate and then place the fibers in the v-grooves. After the fibers are laid in the v-grooves, a UV curable epoxy is applied. A cover plate is then placed on top of the fibers to hold them in place, and the epoxy is then cured. This method creates a single row of fibers. A stock fiber array was bought from Moritex Corporation for the multi-beam system that was made using this method. A picture of the fiber array is shown in Figure The fiber array has 32 single mode fibers in a single row. Fibers are bundled in groups

45 2.1. Design of the Multi-Beam System 27 Figure 2.10: Photo of fiber array from Moritex Corp. A ribbon of single mode optical fibers is glued inside a quartz chip with v-grooves. After [41]. of 8. Within each group of 8, the fiber pitch is 0.25 mm. Between groups of 8 fibers, there is a gap of 0.25 mm. The output facets of the single mode fibers are polished to an 8 angle. These specifications for the fiber array are summarized in Table 2.2. A detailed drawing of the fiber array is shown in Figure Table 2.2: Fiber array specifications Specification Fiber type SMF-28 Number of fibers in the array 32 Fiber pitch 250 µm Output facet angle 8 The lenses in the collimating lens array have a 2 mm pitch and the fibers in the fiber array must have the same pitch. A 2 mm pitch was achieved by using every 7th fiber in the array (because of the 0.25 mm gap between each group of 8 fibers). Four fibers can be used in the 32 channel fiber array. The 32 fibers in the array are bundled in a ribbon, and left unterminated by the manufac-

46 28 Chapter 2. Multi-Beam Delivery System for AFP Switch (a) Output facet of fiber array v-groove chip. Fibers are spaced 0.25 mm apart. The v-groove labeled D in the far left of the figure is magnified and displayed in (b). (b) Dimensions of V-groove. (c) Side view of fiber array chip showing 8 circ angled output facet. Figure 2.11: Detailed drawing of the 32 fiber array from Moritex Corporation. All dimensions are in mm. After [41]. turer. We spliced FC/APC connectors to the fibers in the array before integrating the fiber array into the multi-beam system. FC/APC connectors allow the user of the multi-beam system to quickly change inputs and outputs with commercially available, connectorized optical components.

47 2.2. Fabrication of Collimating Lens Array Fabrication of Collimating Lens Array The collimated lens array was fabricated in collaboration with Professor Eric Johnson at the University of Florida s College of Optics and Photonics (CREOL&FPCE). The fabrication technique is known as additive lithography [42, 43], and is suitable for fabricating arrays of Fresnel lenses [44]. The fabrication facilities at the College of Optics and Photonics are capable of making a large array of Fresnel lenses with a high pitch accuracy. Pitch tolerance is estimated to be < 250 µm across the substrate wafer. In addition, it is estimated that the fabrication facilities in combination with the additive lithography technique are capable of creating uniform Fresnel lenses that are consistent throughout the entire array. The next subsections will discuss the different steps of the fabrication process Introduction to Additive Lithography Process Photolithography is a process where light is used to pattern a physical structure onto a substrate. This is done by first coating the substrate with a photoresist, and exposing the resist to varying intensities of light at different points on the substrate. An amplitude mask which acts as an optical filter is used to control how much light strikes the substrate surface. The pattern that is to be transferred to the substrate is first written on the amplitude mask usually using a direct write process. Using one amplitude mask, many copies of this pattern can be transferred to multiple points on the substrate. After the photoresist has been exposed to light according to the pattern on the amplitude mask, the substrate is washed in a developing solution. For the case of a positive photoresist, areas of the resist that have been exposed to light will be washed away by the solution, while the unexposed portions will remain on the substrate. This creates the desired physical pattern in the photoresist. To transfer the pattern from the photoresist into the substrate, the developed substrate is etched using either chemical and/or physical methods to eat away at the substrate and photoresist. Portions of the substrate that are not covered by photoresist will be exposed for a longer period of time during the etching process, and thus eaten away faster by the etching process, thus transferring the physical structure from the photoresist to the substrate.

48 30 Chapter 2. Multi-Beam Delivery System for AFP Switch Additive lithography is a type of photolithgraphy process that uses a series of binary amplitude masks to create a pattern on photoresist. As a result, it creates quantized patterns, and the maximum number of depth levels it can create is 2 M, where M is the number of masks used during patterning. This concept is illustrated in Figure In the figure, the rings comprise the amplitude mask where the shaded rings block the light source from reaching the photoresist. All three masks are exposed, one after the other in one exposure process, to achieve an 8-level pattern. The masks are positioned directly on top of each other. After the exposure process, the wafer is washed once in developing solution, and is then sent for etching to transfer the physical pattern into the substrate. (a) First mask. (b) Second mask. (c) Third mask. Figure 2.12: With the additive lithography technique, an 8-level pattern can be created with 3 masks exposed one after another. The photoresist used is a positive photoresist. This means that the longer the resist has been exposed to light, the more of it will be washed away after patterning. In the figure, the shaded areas indicate photoresist that has been exposed to the light source. Different shades indicate which mask was used to pattern that part of the photoresist. The height of

49 2.2. Fabrication of Collimating Lens Array 31 the shaded areas indicate how long the photoresist was exposed for, and consequently how much will be washed away in the developing solution. The advantages of the additive lithography process are that it can create complex structures such as Fresnel lenses with a simpler process than other 2 M binary lithography techniques. Furthermore, due to the nature of the additive lithography process, the amplitude masks are much more reusable for different structures and different substrate materials compared to photolithography using grayscale masks. Additive lithography is also well suited to making an array of lenses with high accuracy, both on the individual lens and in the pitch between lenses. The disadvantage of using the additive lithography process is that it produces quantized patterns instead of physical structures with continuous features. Quantization of the lens surface results in a decrease in the incident optical power that is focused at the focal spot. The impact of the quantization on the focused power is investigated in the next subsection Efficiency of a Quantized Fresnel Lens For the additive lithography process used to make the Fresnel lens array, 3 amplitude masks were used thus giving the capability of creating an 8-level structure. The total sag of the lens is L = λ, and each step is of uniform height L. The Fresnel lenses in the array were n 1 8 designed for operation at 1550 nm wavelength and assuming a refractive index of 1.5, L = 3100 nm and each step is 388 nm. The efficiency η of the Fresnel lens can be defined as the amount of input power focused into a circular aperture of radius R f in the plane z = z a as shown in Figure R f is a value that represents the radius of this aperture; η is defined by: η = 2π R f 0 U(r a ) 2 r a dr a P in (2.3) where P in is the input power of the incident beam, r a is a parameter that represents the radius from the z-axis in the plane z = z a where the aperture is located, and U(r a ) is proportional to the electric field.

50 32 Chapter 2. Multi-Beam Delivery System for AFP Switch Figure 2.13: Analysis of Fresnel lens diffraction efficiency in cylindrical coordinates. To distinguish between the z = d 0 plane and the z = z a plane in Figure 2.13, r is used for the radial coordinates at the plane z = d 0 where the incident beam meets the Fresnel lens, and r a is used for the radial coordinates at the plane z = z a. To calculate the diffraction efficiency, we first find the phase incurred on a normally incident plane wave passing through the Fresnel lens as shown in Figure The general equation for the field U in (r) after passing through the Fresnel lens at the plane z = 0 is: U z=0 = U in (r) exp (iφ(r)) (2.4) To calculate the phase incurred, Φ(r), we separate the lens into its individual Fresnel zones and sum up their effect on the incident plane wave: Φ(r) = ( N r r h r h 1 2 φ(h) rect r h r h 1 h=1 where N is the number of Fresnel zones and the rect(t) function is defined as: ) (2.5)

51 2.2. Fabrication of Collimating Lens Array 33 0 if t > 1 2 rect(t) = 1 if t = if t < 1 2 (2.6) For the h th Fresnel zone, the phase φ(h) is incurred by the incident plane wave passing through that portion of the lens according to: φ(h) = (n d h + (d 0 d h )) 2π λ (2.7) where n is the refractive index of the lens, d h is the thickness of the h th Fresnel zone, and d 0 is the sag of the lens. Combining equations 2.4 and 2.5, the field at the plane immediately after the Fresnel lens is: by: ( N ( r r h r h 1 2 U z=0 = U in (r) exp φ(h) rect r h r h 1 h=1 )) (2.8) The field at the plane z = z a can be calculated using the Fresnel propagation and is given ( U(r a ) = 2π exp i k ) 0ra 2 2z exp (i k ) 0r 2 2z r=0 J 0 ( k0 r a r z ( N U in (r) exp φ(h) rect h=1 ( r r h r h 1 2 r h r h 1 )) ) rdr (2.9) where J 0 (t) is the zero order Bessel function. Equation 2.9 can be simplified by considering the Fresnel propagation for each Fresnel zone separately. In other words, by taking the integration into the summation, equation 2.9 can be simplified to: ( U(r a ) = 2π exp i k ) 0ra 2 2z N h=1 rh exp(φ(h)) U in (r)exp r h 1 (i k 0r 2 If a Gaussian beam is the input to the lens, equation 2.10 becomes: 2z ) ( ) k0 r a r J 0 rdr (2.10) z ( U(r a ) = 2π exp i k ) 0ra 2 2z N h=1 rh ) ( ) exp(φ(h)) exp ( r2 r h 1 w + ik 0r 2 k0 r a r J 2 0 rdr (2.11) 2z z

52 34 Chapter 2. Multi-Beam Delivery System for AFP Switch A MATLAB simulation was performed with a collimated Gaussian beam as the input to the Fresnel lens. Figure 2.14 shows the field magnitude vs. radius at the focal plane and shows that the beam profile looks very close to a Bessel function despite the quantization of the Fresnel lens. The calculated diffraction efficiency of the Fresnel lens in the array at the design wavelength of 1550 nm is 87%, or 0.6 db transmission loss. The Fresnel lens efficiency will contribute 0.6 db of loss for the transmission of the Control pulse from the input of the fiber array to the AFP switch surface since this beam path passes through one Fresnel lens. For each data channel, two Fresnel lenses are used in the beam path and thus the transmission loss is 1.2 db. This transmission loss is acceptable for the multi-beam system and its later use in the high bit-rate optical testbed. The 0.6 db loss in the beam path for the Control pulses can easily be compensated for using the high gain EDFA of optical testbed as outlined in Chapter 4. The 1.2 db loss in the beam path for the data channels can also be compensated for using an EDFA to amplify the switched data. An EDFA is needed in the data channel even if the Fresnel lens efficiency were higher than 87% since the insertion loss of the AFP switch alone is roughly 10 db. Thus the power degradation due to the Fresnel lens efficiency should not affect the high bit rate experiments we plan to perform later in the thesis.

53 2.2. Fabrication of Collimating Lens Array 35 Figure 2.14: Electric field magnitude at the focal plane of the Fresnel lens Fabrication Step One: Characterizing the Photoresist To make an array of lenses, a 10 cm diameter circular quartz wafer was the substrate used in the additive lithography process. A layer of Shipley 1813 photoresist (a positive photoresist) from Microchem Corp. was spin-coated on the surface of the water using a P6700 Series spin coater from Specialty Coating Systems Inc. The photoresist is comprised of three different liquids: 1. A polymer. 2. A photo initiator that causes polymerization of the polymer when exposed to light. 3. A solvent to dissolve the other two liquids and to ensure the correct viscosity of the resist. We fabricated a array of lenses, comprising an area roughly 2 cm by 4 cm as shown in Figure Since such a large area needs to be patterned, a thin layer of resist 2000 nm thick, was spin-coated on the substrate to ensure good uniformity across the substrate.

54 36 Chapter 2. Multi-Beam Delivery System for AFP Switch After spin coating, the coated wafer is then baked on a hot plate at 115 C for 1 minute and 30 seconds to evaporate the solvent. Figure 2.15: A lens array was made taking up an area roughly 2 cm by 4 cm in the middle of the wafer. After baking, the photoresist needs to be characterized to determine its response to light. A GCA 6300C G-Line 5X direct wafer stepper machine will be used to pattern the photoresist. The light source in the GCA 6300C is a mercury lamp emitting at 436 nm (g-line) and the same light source was used for the photoresist characterization. During the characterization, the lamp output power is kept constant. Small spots on the coated wafer were exposed to varying amounts of light by changing the exposure time to the mercury lamp. The wafer is then washed in developing solution to remove the exposed photoresist. The depth of the removed resist is measured using a Tencor Alphastep 200 Profilometer. The result is plotted in Figure Figure 2.16 shows that the resist exhibits a linear response between exposure times of 0.56 and 0.98 seconds. For exposure times shorter than 0.56 seconds, the resist removed is negligible. For exposure times longer than 0.98 seconds, the resist removed increases nonlinearly with exposure time. The uncertainty in the measurement of the resist depth by the profilometer is ±10 nm according to the manufacturer s specification. The uncertainty in the exposure time to the mercury lamp in the GCA 6300 is ±0.005 s. The error bars in Figure 2.16 are the calculated average difference between the linear fit and the measured data to give some idea of the deviation between a perfectly linear photoresist and the results we measured. The error bars are ±0.012 s in the x-direction and

55 2.2. Fabrication of Collimating Lens Array 37 ±42 nm in the y-direction. Figure 2.16: Plot of resist removed after being developed vs. exposure time to mercury lamp. It is uncertain whether the actual error bars are larger than those shown in Figure although it is likely that they are. Repeated trials of the resist characterization could not be performed due to the time limitations of the cleanroom facility and staff. Measurement uncertainty could be due to a couple of factors. First of all, during the spin coating process, the composition and thickness of the resist may vary slightly at different distances from the centre of the wafer. Since the measurements of the removed resist were taken at various points on the wafer, this could cause some of the deviation from a linear fit. Power fluctuations in the mercury lamp in the GCA 6300C are also a likely culprit of measurement uncertainty. In the next subsection, patterning of the photoresist will be described. To simplify the patterning process, we choose to work in the linear region of the photoresist i.e. exposure times are 0.60 seconds for the thickest Fresnel zones, and 0.95 seconds for the thinnest. This requires that a bias exposure be made to the entire photoresist of 0.60 seconds before patterning begins. To remain in the linear region, patterning exposure times must be shorter than = 0.35 seconds.

56 38 Chapter 2. Multi-Beam Delivery System for AFP Switch Fabrication Step Two: Patterning the Photoresist After the photoresist has been characterized, a wafer coated with the same thickness of photoresist (2000 nm) is placed inside a GCA 6300C G-Line 5X direct wafer stepper machine. A schematic of the stepper is shown in Figure Figure 2.17: Schematic of direct wafer stepper patterning chamber. The binary amplitude mask in Figure 2.17 contains the pattern to be exposed onto the photoresist. The mask was produced by direct writing using a Leica EBPG Electron Beam System according to the quantized Fresnel zones calculated in subsection. Three amplitude masks were used to produce one entire Fresnel lens pattern. All three masks were written on a single substrate. During patterning, all three masks are exposed sequentially, with the patterns suitably aligned to create the Fresnel lens. The coated wafer is placed on a motorized, high precision x-y translation stage which moves around during the patterning to expose different parts of the wafer to the three amplitude masks. A Fresnel lens is patterned on one spot of the photoresist by exposing to all three masks, and then the wafer is stepped to the next lens center and the process is repeated.

57 2.2. Fabrication of Collimating Lens Array 39 The next step is to calculate how long the exposure time should be for each of the three binary amplitude masks. Since we are working in the linear region of the photoresist, the depth of removed photoresist is directly proportional to the exposure time. Consider the 8- level structure shown in Figure 2.18(a). A bias exposure of 0.60 seconds is first made so that subsequent patterned exposures will be in the linear region of the photoresist. It is desirable to make as deep a structure as possible in the photoresist so that the etching ratio later on will not be too high. This can be done by using the entire linear region of the photoresist response. In order to make a structure with uniform step size, set t 1 + t 2 + t 3 = t 1 + 2t 1 + 4t 1 = 0.35 seconds, and thus t 1 = 0.05 seconds as shown in Figure 2.18(b). (a) The depth of resist to be removed is 7d (b) To make full use of the linear region, set t 1 = 0.05 seconds. to make an 8-level structure. Figure 2.18: Determining the exposure time for the three masks. The three binary amplitude masks are exposed for either t 1, t 2, or t 3 seconds in order to create the desired pattern. After the entire lens array has been exposed, the wafer is removed from the wafer stepper and washed in a developing solution of 5 parts deionized water to 1 part Microposit 351 Developer Solution for one minute to remove the exposed photoresist. Only after development is the physical shape of the Fresnel lens realized in the photoresist. The Fresnel lens patterned in the photoresist is examined using a Zygo NewView 6300 optical profiler as shown in Figure 2.19 to ensure that 8 levels were patterned. The deepest depth of photoresist removed was 1330 nm as marked on Figure 2.19(b) which roughly corresponds

58 40 Chapter 2. Multi-Beam Delivery System for AFP Switch to the photoresist characterization (depth due to 0.95 second exposure - depth due to 0.60 second bias exposure). (a) Profile of the center portion of the Fresnel lens. (b) Cross section of photoresist of the center portion. Figure 2.19: Depth profile of Fresnel lens pattern in photoresist measured with a Zygo optical profiler Fabrication Step Three: Transfer the Pattern from Photoresist to Substrate The next step in the additive lithography process is to transfer the pattern from the photoresist into the substrate. This is done by a reactive ion etching process (RIE) where ions and neutral particles react chemically and physically with the patterned wafer to evenly remove the photoresist and quartz substrate at a controlled rate.

59 2.2. Fabrication of Collimating Lens Array 41 The physical pattern created on the photoresist is 1330 nm at the deepest point. To λ achieve the needed sag of the Fresnel lens of = 3100 nm, the step height between n lens 1 Fresnel zones should be = 388 nm and the deepest point on the substrate that needs to be etched is nm = 2716 nm. To achieve the correct step heights in the quartz substrate, an etching process needs to be used that etches quartz 2.04 times faster than it etches away the photoresist (2716 nm 1330 nm = 2.04). The etcher used was a Unaxis Versaline ROS model which uses the inductively coupled plasma (ICP) technique to generate the ions used for etching as shown in Figure A combination of CHF 3 and Ar gases were used, and the etching recipe is shown in Table 2.3. Figure 2.20: Schematic of ICP RIE chamber. The electric field RF 1 creates a plasma of reactive ions used in the etching, while the electric field RF 2 biases the substrate. After [45].

60 42 Chapter 2. Multi-Beam Delivery System for AFP Switch Table 2.3: Etch Recipe Parameter Value Chamber Pressure 10 mtorr Throttle 31.89% ICP Forward Power 850 W RF Bias Power 75 W RF DC Bias V CHF 3 30 sccm Ar 75 sccm Etching time 14 minutes After the etching was complete, the wafer was cleaned with a mixture of acetone and isopropyl alcohol to remove any remaining photoresist, and again the depth profile was examined with a Zygo NewView 6300 as shown in Figure The figure shows an 8 level structure with a reasonably similar depth (2443 nm vs nm) compared to the ideal value. (a) Profile of the center portion of the Fresnel lens. (b) Cross section of photoresist of the center portion. Figure 2.21: Depth profile of Fresnel lens in the quartz substrate measured with a Zygo optical profiler.

61 2.2. Fabrication of Collimating Lens Array 43 Pictures were also taken of the individual Fresnel lenses using a Nikon Eclipse L150 Microscope to show the entire diameter of the lens, and to get a qualitative feel of surface roughness as shown in Figure The fabricated lens diameter was 2 mm as was designed. (a) An individual Fresnel lens in the array. (b) The central portion of an individual Fresnel lens. Figure 2.22: Microscope images of a Fresnel lens in the collimating lens array.

62 44 Chapter 2. Multi-Beam Delivery System for AFP Switch 2.3. Characterization of the Lens Array and Integration into Multi-Beam System Characterizing the Fresnel lenses in the array We first characterized the individual Fresnel lenses, and then the lens array was assembled into the multi-beam system. The transmittance of the lens array was first measured in the wavelength range from nm using the setup shown in Figure fiber collimator outputs a 500 µm diameter collimated beam. The A continuous wave (CW) tunable laser (TLS) beam was used to sweep the wavelength. The output from the fiber collimator was focused onto the power meter without the lens array, and then the lens array was inserted in between as shown in Figure The transmittance was calculated as power with lens array in the setup and the measured results are shown in Figure power without lens array in the setup Figure 2.23: Setup to measure transmittance of the lens array

63 2.3. Characterization of the Lens Array and Integration into Multi-Beam System 45 Figure 2.24: Measured transmittance of the lens array. Next the focal length of the Fresnel lens was measured using the setup shown in Figure A vidicon camera with a focusing lens is placed on a translation stage moving in the z-direction. The vidicon is first focused on the surface of the Fresnel lens. Then a collimated CW laser beam at 1550 nm is input to the Fresnel lens and the vidicon is then translated until it reaches the focal plane of the Fresnel lens. This appears on the vidicon as the smallest spot in the z-direction since the beam waist is being imaged into the camera. Using this method, the focal length of the Fresnel lens in the array is measured to be 6.3 ± 0.2 mm. Figure 2.25: Experimental setup to measure focal length of Fresnel lenses.

64 46 Chapter 2. Multi-Beam Delivery System for AFP Switch Aligning the Fiber Array and the Lens Array To assemble the multi-beam system, the lens array was first aligned to the 32 channel fiber array that was purchased from Moritex Corporation as shown in Figure of the 32 channels in the fiber array can be used to achieve a 2 mm pitch between fibers. Only three of the channels are needed for the alignment. It is easy to see in Figure 2.26 how a two dimensional fiber array can be added to the multi-beam system to greatly increase the number of channels being delivered to the AFP switch. Figure 2.26: Positioning of fiber array and lens array. The fiber array is mounted on one stage with pitch, roll, and yaw adjustments. The lens array is mounted on another stage with xyz translations, as well as pitch, roll, and yaw adjustments. The lens array is used to collimate the fiber output and so the fiber array is placed one focal length away from the lens array. A BeamScan beam profiler on an xyz translation is placed in the path of the collimated beams to help with the alignment to ensure that all three collimated beams are of uniform diameter, radially symmetric, and have the same beam divergence. Translation in the xy-plane is needed since the beam profiler aperture is only large enough to measure one collimated beam at a time. The beam profiler is then replaced with a vidicon camera and the profile of the three collimated beams was measured every 0.2 mm apart on the z-axis as shown in Figure 2.27.

65 2.3. Characterization of the Lens Array and Integration into Multi-Beam System 47 The vidicon camera images the collimated beam at a distance roughly 10 cm away from the lens array along the z-axis. We can calculate the ideal Rayleigh range z 0 for the collimated beam assuming an ideal thin lens to get a rough idea of whether we are measuring in the depth of focus of the Fresnel lens. Using the formula z 0 = πw 0 after [39], we calculate that λ the Rayleigh range is 72 cm at a wavelength of 1550 nm for a lens which has a focal length of 6.3 mm. In the formula to calculate the Rayleigh range, W 0 is the calculated collimated beam radius at the lens array. Since the z range in Fig is well within the calculated Rayleigh range of the collimated beam, we expect to find highly collimated beams, which was exactly what we have observed, as shown. Figure 2.27: Vidicon images combined to show collimated beam propagation.

66 48 Chapter 2. Multi-Beam Delivery System for AFP Switch Alignment of the Focusing Lens and Multi-Beam System Performance Once the fiber array and lens array are correctly aligned to give collimated beams, the focusing lens is added to complete the multi-beam system. The focusing lens is centered on the optical axis for the multi-beam system as shown in Figure The focusing lens is mounted on an xy translation stage in order to center the lens on the optical axis, and has pitch and roll adjustment to correct the angle of the focusing lens. To align the focusing lens, the beam profiler is again used to make sure the focused beams overlap in the same spot, are radially symmetric, and that the beams from Fiber A and B are symmetrically spaced about the control beam. Figure 2.28: Integration of the focusing lens with the fiber and collimating lens array completes the multi-beam system. The beam overlap at the focal plane of the aligned focusing lens is shown in Figure The beam diameters are also indicated in the legend for each beam at the 1/e 2 power points. The figure shows a relatively circular beam profile in the xy-plane and beam overlap to within 5 µm apart. Once the focusing lens was aligned, the performance of the overall multi-beam system was tested by placing a 99% reflectance mirror in place of the AFP switch to measure the

67 2.3. Characterization of the Lens Array and Integration into Multi-Beam System 49 (a) Y-direction. (b) X-direction. Figure 2.29: Beam profile of Fiber A, Fiber B and Control at the focal plane of the focusing lens. insertion loss of the multi-beam system as shown in Figure 2.30 for transmission from Fiber A, reflected back into Fiber B by the mirror and vice versa. The insertion loss from the FC/APC connector to the mirror is roughly db for each of the channels due to factors such as connector loss, splice loss, reflection loss, diffraction efficiency of Fresnel lens, etc. Any additional loss is likely due to imperfect alignment of the optics including the mirror and can be improved by more meticulous alignment. Figure 2.30: Measured insertion loss of the multi-beam system

68 50 Chapter 2. Multi-Beam Delivery System for AFP Switch In summary, a multi-beam system was designed, fabricated, and characterized. The multi-beam system can deliver four beams to the AFP switch, focused to a small spot size of µm. The beams are focused at the same point on the AFP switch surface. This results in good coupling of reflected beams back into the multi-beam system. In addition, having all of the beams overlap ensures good switching operation when the multi-beam system is used with the AFP switch. This is because the incident signal beams will overlap with the volume of the AFP switch where absorption saturation has occurred due to the control beam.

69 Chapter 3 Demonstration of the Multi-Beam System with the AFP Switch With the completion of the multi-beam system, the multi-channel operation of the AFP switch could now be demonstrated. Pump probe measurements were first performed on the AFP switch to determine the recovery time of the switch when operated with multiple signal inputs, and this is discussed in the first section of this chapter. The second section discusses the crosstalk between adjacent fibers when the multi-beam system is used with the AFP switch. In the third section, a wavelength conversion demonstration will be shown. All experiments were performed with the laser system shown in Figure 3.1. The 532 nm CW laser is a Spectra-Physics Millenia Pro 10s. It is used to pump the Ti:Sapphire laser which is a Spectra-Physics Tsunami 3960 LLS which generates laser pulses at a repetition rate of 82 MHz. The pulses from the Tsunami are converted in wavelength using the optical parametric oscillator (OPO) which is a home-built device [33]. The idler tunability of the OPO is in the nm range, and signal tunability in the nm range. Output pulses are transform limited, 2 ps measured at the full-width half-maximum (FWHM) of the peak power. The signal and idler pulses from the OPO are then coupled into the fiber and connected to the multi-beam system using the OPO signal as the input for Fiber A and B, and using the OPO idler as the input for the Control port. 51

70 52 Chapter 3. Demonstration of the Multi-Beam System with the AFP Switch Figure 3.1: Laser system used to demonstrate AFP switch Multi-channel operation of the AFP switch To illustrate the capability of the AFP switch to simultaneously switch multiple inputs at once, a simple two channel demonstration was performed using the experimental setup shown in Figure 3.2. The two channels are: 1. Data channel 1 where Fiber A is the input port and Fiber B is the output port through port 3 of Circulator B. 2. Data channel 2 where Fiber B is the input port and Fiber A is the output port through port 3 of Circulator A.

71 3.1. Multi-channel operation of the AFP switch 53 Figure 3.2: Experimental setup for multiple channel demonstration of AFP switch. Circulator ports are numbered. Multiple channel operation was demonstrated by measuring the recovery time of the AFP switch using the pump probe technique. In the pump probe technique, a strong pump pulse is used to stimulate a nonlinear effect, while a weak probe pulse is used to measure the effect caused by the pump pulse. By varying the time between the pump and probe pulse using a delay stage and measuring the average power of the probe pulse at a particular delay stage position, the temporal profile of the nonlinear effect caused by the pump can be determined. In this experiment, the pump pulse causes absorption saturation in the AFP switch and acts as the Control pulse of the switch - in other words, it turns the switch ON. The pump probe technique is used to measure recovery time of the AFP switch from ON back to OFF state after a Control pulse is absorbed by the switch. The recovery time is defined as the time between the peak and the time where the power has decayed to 1/e of the peak power. The signal output from the OPO is used as the probe pulse. The probe pulse that is sent into Fiber A measures the recovery time for Data Channel 1, while the signal pulse sent into Fiber B measures the recovery time for Data Channel 2. This is done by measuring the power recorded on Power Meter 1 and Power Meter 2 respectively. A sample pump-probe trace is shown in Figure 3.3 which shows the AFP switch recovery time using this method. The Control pulse wavelength was 1536 nm, while the signal pulse wavelength into Fiber A

72 54 Chapter 3. Demonstration of the Multi-Beam System with the AFP Switch and B was 1574 nm. Figure 3.3: Pump probe trace for Data Channel 1 and Data Channel 2. The contrast ratio (CR) and recovery time of the AFP switch are indicated in the legend. The short recovery time of the AFP switch indicates that simultaneous multiple channel operation at 40 Gbit/s should be possible (25 ps bit period), and potentially even operation at 80 Gbit/s. 80 Gbit/s can be achieved by decreasing the recovery time by reducing the control pulse energy, at the cost of a reduced contrast ratio [46]. We also used the pump probe technique to measure the contrast ratio of the AFP switch with the multi-beam system. The contrast ratio for a digital OOK signal is defined as ( ) 10 log 1 level 10. For the case of the AFP switch, the 1 level refers to the reflected 0 level optical power from the AFP switch when the switch is turned ON. The 0 level refers to the reflected power when the AFP switch is turned OFF. We measured the 1 level at the peak of the trace shown in Figure 3.3 which is when the switch is fully turned ON. We measured the 0 level as the power level before the control pulse arrives at the switch. Good contrast ratios of 15.2 db and 16.2 db for Channel 1 and Channel 2 respectively were obtained.

73 3.1. Multi-channel operation of the AFP switch 55 Figure 3.4 shows the pump probe trace for Data Channel 1. The error bars show the scatter when the measurement was repeated four times. The source of most of the scatter is likely to be the output power fluctuation of the laser system due to mode-locking instabilities, and is probably not due to any performance fluctuations in the AFP switch. (a) Pump probe trace showing error bars of measured data. (b) Rising edge of pump probe trace shown in (a). (c) Falling edge of pump probe trace shown in (a). Figure 3.4: Pump probe trace of Data Channel 1 at a signal wavelength of 1576 nm. The experimental error associated with the contrast ratio in Figure 3.4 is ±0.26dB. This error in the contrast ratio is calculated using the following formula for error propagation:

74 56 Chapter 3. Demonstration of the Multi-Beam System with the AFP Switch error in contrast ratio = 10 ln10 (1 level error) 2 + (0 level error) 2 (3.1) where the error in contrast ratio has units of db and the error in the 1 level and 0 level are fractional errors(fractional error defined as absolute error average measured value ). To demonstrate the large wavelength range over which the AFP switch can be used with the multi-beam system, pump probe measurements were made at several different signal wavelengths, and the measured contrast ratios from these experiments are plotted in Figure db or higher contrast ratio is achieved in both channels over a range of approximately 10 nm. It is believed that higher contrast ratios can be achieved by more careful alignment of the multi-beam system. (a) Data Channel 1 (b) Data Channel 2 Figure 3.5: Contrast ratio of AFP switch measured at different signal wavelengths. Pump wavelength varies from 1536 nm to 1540 nm for a signal wavelength that varies from 1562 nm to 1587 nm as shown. In general, the results show that the multi-beam system allows the AFP switch to be used as a multiple channel device, while preserving the large wavelength range and high speed operation of the switch that was previously demonstrated in [33].

75 3.1. Multi-channel operation of the AFP switch Crosstalk Measurements Once multiple channel operation of the AFP switch was demonstrated, the channel-tochannel crosstalk was measured. Using the same setup as that shown in Figure 3.2, the fiber to fiber coupling was measured and is shown in table 3.1. Table 3.1: Crosstalk Measurements Control pulse ON Control pulse OFF From To Crosstalk Fiber A Fiber C -58 db < -60 db Fiber B Fiber C -59 db < -60 db Control Fiber A or B -56 db Not applicable Control Fiber C < -60 db Not applicable The results in table 3.1 are further illustrated by examining the spectral profile of a pulse after being switched and reflected from the AFP switch, shown in Figure 3.6. In this figure, a pump probe experiment was performed with a signal wavelength at 1578 nm and a Control wavelength of 1539 nm. Despite the Control pulses having an average power approximately 23 db higher than the signal pulses, any coupling of the Control pulses into the signal fibers (Fibers A through C in Figure 3.2) have low enough power they are in the noise floor of the optical spectrum analyzer (OSA). The low values for crosstalk that were measured for the AFP switch using the multi-beam system are comparable to commercially available telecom components, thus demonstrating the good channel-to-channel isolation Wavelength Conversion Demonstration One of the basic requirements for dynamic optical networks is wavelength conversion where digital information in the form of bits is transferred from one optical carrier wavelength to another. Because of its multiple channels, the AFP switch with the multi-beam system has the potential to simultaneously convert a data channel at one wavelength into many other wavelengths as shown in Figure 3.7. Unfortunately, our demonstration is limited by the

76 58 Chapter 3. Demonstration of the Multi-Beam System with the AFP Switch Figure 3.6: Spectral profile of pulses at the output port of Data Channel 1. Signal pulses are at 1578 nm wavelength. array size of the commercial 1D fiber array available to us, which can only accommodate a maximum of 4 parallel channels. However, single channel wavelength conversion where one channel of data is transferred to another carrier wavelength can still be demonstrated. This will show that wavelength conversion is possible with the multi-beam system and that with a bigger fiber array, the system can be quickly setup and easily reconfigured as a wavelength conversion device. Figure 3.7: Example of simultaneous multiple channel wavelength conversion using the multibeam system with AFP switch. The experimental setup for the single wavelength conversion demonstration is shown in

77 3.1. Multi-channel operation of the AFP switch 59 Figure 3.8, where a data channel at wavelength λ 2 = 1539 nm is converted to a data channel at wavelength λ 1. The tunable laser source (TLS) wavelength λ 1 was varied to demonstrate conversion across a wide wavelength range. The result of the wavelength conversion was viewed on an Agilent 86100C Infiniium DCA-J wideband oscilloscope with an 86116B detector module that has a 65 GHz optical bandwidth. A photodiode measuring the signal output from the OPO was used as the trigger to the oscilloscope. An erbium-doped fiber amplifier (EDFA) is used to amplify the wavelength converted signal for viewing on the oscilloscope due to the high noise level inherent in the oscilloscope. Figure 3.8: Experimental setup for single channel wavelength conversion. The results of the experiment are shown in Figure 3.9 for three values of λ nm, 1550 nm, and 1572 nm. Qualitatively, all wavelengths appear to have good contrast ratios, >7 db. It is difficult to measure the exact contrast ratio since the noise of the optical channel on the oscilloscope is 400 µw and thus it is difficult to accurately measure the OFF state, or 0 level.

78 60 Chapter 3. Demonstration of the Multi-Beam System with the AFP Switch (a) λ 1 = 1532 nm. Vertical scale is 1.46 mw/div. (b) λ 1 = 1550 nm. Vertical scale is mw/div. (c) λ 1 = 1572 nm. Vertical scale is mw/div. Figure 3.9: Results of wavelength conversion experiment using multi-beam system. In all plots, the horizontal scale is 50 ps/div. In this chapter, we have demonstrated some of the multiple channel functionality of the AFP switch, made possible by the multi-beam system. A simultaneous all-optical switching operation was performed on two bi-directional channels. Good contrast ratios, as high as 16.2 db were obtained, and short recovery times as low as 6.8 ps were observed. Crosstalk measurements between adjacent signal fibers in the multi-beam system showed coupling ratios of <-50 db, demonstrating good isolation between channels. This result shows that adding more channels to the multi-beam system should not be a problem.

79 3.1. Multi-channel operation of the AFP switch 61 Wavelength conversion was also demonstrated with the AFP switch and multi-beam system across a wide wavelength range from 1532 nm to 1572 nm. The results from these experiments show that the wavelength converted pulses have very short pulse widths - at most 8 ps at the FWHM which is the measurement limitation of our instruments. The results from these experiments strongly suggest that the range of wavelengths that can be converted is much wider, especially at wavelengths longer than 1572 nm where the absorption of the AFP switch is quite low. Simultaneous conversion of multiple channels at different wavelengths should also be achievable based on the results obtained in this chapter.

80 62 Chapter 3. Demonstration of the Multi-Beam System with the AFP Switch

81 Chapter 4 Testbed for High Speed All-Optical Switches High speed all-optical switches are a promising candidate for future reconfigurable optical networks and the significant effort that has been put toward developing these devices is an indication of this belief. Characterization of all-optical switches at high bit rates is important in determining how well a device will perform under realistic conditions, and can expose device behaviours that are not noticeable at low bit rates such as pattern dependent effects. A high bit rate testbed is needed to test all-optical switches, and other telecommunication devices to perform a proper high speed characterization. Due to the difficulty in setting up such a testbed, the establishment of such a facility here at the University of Toronto should encourage collaboration with other researchers working on telecommunication devices who wish to perform high speed characterization. Within the Photonics group at the University of Toronto, several different high speed alloptical switching devices are currently under development and hopefully will soon be ready for experiments demonstrating the functions that they offer, one such device being the AFP switch. To meet future demands, we designed and built a high serial bit rate testbed that could be used to test a variety of all-optical switches. 63

82 64 Chapter 4. Testbed for High Speed All-Optical Switches 4.1. Testbed Design Criteria and Outline Testbed Design Criteria All-optical switches are being developed for future optical networks which will most likely have higher serial bit rates than are currently implemented to make better use of existing fiber bandwidth. 10 Gbit/s has become an industry standard for telecom (e.g. OC-192, STM-64, 10 gigabit Ethernet), and future bit rates will likely be some multiple of this. Therefore, we designed the testbed to operate at speeds of 10 Gbit/s and higher. One of the main benefits of all-optical switching is the use of an optical control pulse to turn the switch ON and OFF. By using an optical control signal instead of an electronic control signal, it is hoped that speed constraints associated with parasitic capacitances in electronics will be avoided, and thus allow faster operation. Most all-optical switches are based on an ultrafast, intensity-dependent nonlinear effect, which enables them to quickly turn ON and OFF. As a result, most all-optical switches require high peak intensities to create a large nonlinearity and thus achieve a good switching contrast ratio. High peak intensities can be achieved by using short optical control pulses. Taking these considerations into mind, the design criteria for the testbed are: 1. Bit-rates should be adjustable to meet the needs of a variety of telecommunication devices being tested, including all-optical switches. 2. The testbed should provide a synchronized train of optical control pulses. 3. The optical control pulses in the testbed should: (a) Have a high peak power to create large nonlinear effects. (b) Have a repetition rate of 10 GHz. (c) Have a short duration so that the control pulses can be used in high bit rate experiments. 4. The generated optical data pattern should be at a bit rate that is a multiple of 10 Gbit/s. 40 Gbit/s is the minimum bit rate based on current industry development.

83 4.1. Testbed Design Criteria and Outline The testbed should be fiber based. This allows for easy reconfiguration of the testbed since there are no free space optics to align. 6. The testbed should make the best use of existing lab equipment. A schematic showing the proposed functions of the testbed is shown in Figure 4.1. Figure 4.1: Outline of testbed functions Overview of the Testbed The testbed we built was first configured to perform high speed experiments on the AFP switch with the multi-beam system. It can be reconfigured later for other all-optical switches, or other optical telecommunication devices. For demultiplexing, the testbed was initially setup for a 10 GHz demultiplexing rate by setting the control pulse repetition rate to 10 GHz. An example of this is shown in Figure 4.2 where an all-optical switch operates on a 40 Gbit/s data channel comprised of four 10 Gbit/s channels that have been multiplexed together using TDM. Depending on the repetition rate of the optical control pulses, in this case 10 GHz, the all-optical switch in Figure 4.2 selects one of the four channels to demultiplex. The triangles in Figure 4.2 represent a train of repeated pulses. Rectangles represent bits encoded as pulses that are either ON for 1 or OFF for 0 (OOK) in a return-to-zero (RZ) format. This convention will be used for future diagrams as well.

84 66 Chapter 4. Testbed for High Speed All-Optical Switches Figure 4.2: The testbed can be used to demonstrate demultiplexing. The schematic of the testbed that was built to demonstrate the AFP switch is shown in Figure 4.3. The testbed generates two outputs - an Optical Data Channel, and an Optical Control Pulse. The Optical Data Channel block generates a high bit-rate serial channel of data for the AFP switch to operate on. The Optical Control Pulse provides a high powered optical clock to control the operation of the AFP switch. The RF synthesizer in Figure 4.3 outputs a 10 GHz sine wave that acts as the system clock to make sure all components are operating at the same frequency. A mode-locked fiber laser from PriTel Inc. was used to generate a train of short pulses at the repetition rate set by the RF synthesizer. This pulse train is split into two fibers using a 50/50 coupler, one train being used for the optical data channel, and the other for the control pulse generation or optical clock generation. The switched output from the AFP switch is measured using the same digital communications analyzer (DCA) that was used in the previous chapter, an Agilent 86100C Infiniium DCA-J wideband oscilloscope with an 86116B detector module

85 4.1. Testbed Design Criteria and Outline 67 and 86107A Precision timebase for low jitter. The 86116B sampling module on the DCA has two input ports - a 1.85 mm port for electronic signals, and an FC/PC port for optical fiber (there is no need for an external photodiode). Figure 4.3: The testbed configuration for the AFP switch and multi-beam system. Solid lines indicate single mode optical fiber, while dashed lines indicate electronic coaxial cable. Similarly solid boxes indicate components with optical input/output while dashed boxes indicate components with electronic input/output. This convention will be used in all further figures showing the testbed layout. A more detailed schematic of the testbed is shown in Figure 4.4. In Figure 4.4 the components used in the Optical Data Channel and Optical Control Pulse blocks are shown that were previously abstracted in Figure 4.3.

86 68 Chapter 4. Testbed for High Speed All-Optical Switches Figure 4.4: Schematic of the testbed with all components shown. Darkly shaded boxes indicate components used in the Optical Control Pulse block. Lightly shaded boxes indicate components used in the Optical Data Channel block.

87 4.2. Optical Data Channel Generation 69 The next sections will go into more detail on the Optical Data Channel block and the Optical Control Pulse block of Figure 4.3. The details of the testbed components such as manufacturer and model number can be found in appendix B together with the layout of the electronic components. This will be helpful to users who will be working with the testbed Optical Data Channel Generation Design of the Optical Data Channel Block The Optical Data Channel block of Figure 4.3 has the function of creating a high serial bit rate channel of optical data from the train of pulses output by the mode-locked fiber laser. The components used to do this in the Optical Data Channel block are shown in Figure 4.5 and are indicated by the shaded boxes. The Optical Data Channel block simulates data traffic that would be the input to an all-optical switch in an optical network. The following is an explanation of the acronyms used in Figure 4.5. The presence of a 1 suffix refers to the fact that the component is used for the Optical Data block. A 2 suffix will be used later in the thesis to indicate components that are used in the Optical Control Pulse block. MZM 1 - The MZM stands for Mach-Zehnder Modulator. MD 1 - The MD stands for Modulator Driver. This is an electronic amplifier that amplifies the voltage of the input signal to high enough levels to drive the MZM, roughly Vpp. ODL - Optical Delay Line P RBS - Pseudo-Random Bit Sequence V OA - Variable Optical Attenuator The data channel begins with the mode-locked fiber laser (MLFL) that outputs a train of transform-limited Gaussian pulses at a repetition rate of 10 GHz. The wavelength range of the MLFL is from nm. The MLFL pulse width can be adjusted to be between

88 70 Chapter 4. Testbed for High Speed All-Optical Switches Figure 4.5: Components used in Optical Data Channel block of Figure 4.3 are shaded in grey. Bit rates at various points in the block are indicated. < ps FWHM by adjusting the bandwidth of the passband filter in the laser cavity. Since the output of the MLFL will also be used to generate the control pulse, it is desirable to keep the pulse as short as possible to ensure the AFP switch recovery time is short. A gain filter was selected that resulted in an output pulse width of ps. Next, a 10 Gbit/s bit sequence is encoded onto the pulse train from the MLFL using a MZM as shown in Figure 4.6. The PRBS generator outputs a bit long PRBS pattern in a non-return-to-zero (NRZ) format. An example bit sequence is shown in Figure 4.6. This electronic data is transferred to the optical pulse train by switching the MZM ON and OFF for 1 and 0 bits respectively. Since the optical pulses ( ps) are much shorter than the bit period at 10 Gbit/s (100 ps), the optical data is in a return-to-zero (RZ) format. A polarization controller is used at the MZM input to minimize insertion loss of the MZM. The phase shifter, which will be represented in later diagrams by just the symbol φ, is used to control the timing delay between the pulses from the MLFL and the 10 Gbit/s bit sequence from the PRBS generator.

89 4.2. Optical Data Channel Generation 71 Figure 4.6: A MZM is used to encode electronic data onto a 10 GHz optical pulse train. The 10 Gbit/s RZ optical data channel is then multiplied into a 40 Gbit/s RZ channel using a device known as an Optical Clock Multiplier, from PriTel Inc. An example of a 2X multiplier is shown in Figure 4.7 which uses a Mach-Zehnder configuration. A 50/50 coupler is used to split the data into two copies, one in each fiber. One fiber has an ODL which delays the data in one of the fibers by half the bit period ( T b 2 ). The VOA in the other fiber ensures that the peak powers are the same in both fibers. The two channels are then combined back together using another 50/50 coupler. Two of these multipliers are cascaded in series in the Optical Clock Multiplier. Figure 4.7: Schematic of a 2X optical clock multiplier.

90 72 Chapter 4. Testbed for High Speed All-Optical Switches In addition to multiplying the bit rate, a 2 n 1 PRBS pattern length can be maintained if the relative delay between the two arms of the Mach-Zehnder interferometer used to multiply the bit rate is set to half the pattern length. An example is shown in Figure 4.8 for a bit pattern [47]. Figure 4.8: Optical multiplication and preservation of the PRBS pattern length using a pattern as an example. Optical clock multipliers can be cascaded in series to double the bit rate, where the cascadability is limited by the insertion loss they introduce and the temporal width of the pulses to avoid intersymbol interference. The testbed has been setup for an 80 Gbit/s data channel using three 2X optical clock multipliers. Two are found in the Optical Clock Multiplier bought from PriTel Inc. and one is comprised of discrete stock laboratory components (VOA, ODL, 50/50 couplers). Given the short pulse width of the MLFL, 2X multipliers can be cascaded to achieve a bit rate of 160 Gbit/s (6.7 ps bit period) with less than 2% overlap between pulses. Doubling the bitrate to 320 Gbit/s or higher requires pulse compression to reduce the pulse width as was demonstrated in [48]. Each 2X multiplier results in a 5-6 db insertion loss. A minimum of 3 db loss occurs from recombining the two bit streams together with a 50/50 coupler, plus insertion losses associated with the VOA, ODL, and connector and splice losses. Since the maximum average output power of the MLFL is 10 mw, an EDFA is needed to view the Optical Data Channel pulses on the DCA. An ODL (ODL 1B) is also needed after the EDFA to adjust the timing

91 4.2. Optical Data Channel Generation 73 of the Optical Data Channel pulses so that they overlap with the pulses from the Optical Control Pulse block. The design of the Optical Data Channel block provides a high serial bit rate RZ data channel with bit rates adjustable from 10 Gbit/s up to 80 Gbit/s. The channel rate can easily be doubled to 160 Gbit/s with the addition of another 2X multiplier. The optical data channel is generated in optical fiber, and copies of this channel can easily be created with a directional coupler for multiple channel demonstrations with the AFP switch and multi-beam system Signal Quality in the Optical Data Channel Block In this section, eye diagrams showing the signal quality at various points in the Optical Data Channel block are presented. All plots are taken with the DCA. The electronic output of the PRBS generator is plotted in Figure 4.9 which shows both the eye diagram and a sample of a bit pattern at 10 Gbit/s. After the PRBS generator output is amplified by the modulator driver (MD 1), it is used as the electronic control signal for the MZM (MZM 1). To measure the transmission of the MZM output as it is being controlled by the PRBS generator, the experimental setup shown in Figure 4.10(a) can be used. The eye diagram for this setup is shown in Figure 4.10(b). The signal-to-noise ratio in the eye diagram of Figure 4.10(b) cannot be compared to the eye diagram of Figure 4.9(a) since the noise added by the sampling module on the DCA for the optical input is much higher than for the electronic input.

92 74 Chapter 4. Testbed for High Speed All-Optical Switches (a) Eye diagram. Vertical scale is 53.4 mv/div. Horizontal scale is 20.0 ps/div. (b) Sample of pattern. Vertical scale is 65.2 mv/div. Horizontal scale is ps/div. Figure 4.9: 10 Gbit/s, length, electronic output of PRBS generator.

93 4.2. Optical Data Channel Generation 75 (a) Experimental setup. (b) Eye diagram of MZM optical output. Vertical scale is 500 µw/div. Horizontal scale is 20.0 ps/div. Figure 4.10: Optical output of MZM as it modulates a CW laser.

94 76 Chapter 4. Testbed for High Speed All-Optical Switches In the actual Optical Data Channel block, the MZM is used with the MLFL to block or allow pulses from the MLFL depending on whether the PRBS generator outputs a 1 or 0 bit. The phase shifter in Figure 4.5 adjusts where in the bit period of Figure 4.10(b) the pulse from the MLFL passes through the MZM, and is adjusted to maximize the signal to noise ratio. An eye diagram of the MLFL pulse train after it has been encoded with the 10 Gbit/s PRBS pattern is shown in Figure 4.11(a) and a sample of the pattern is shown in Figure 4.11(b). (a) Eye diagram. Vertical scale is 5.00 mw/div. Horizontal scale is 2.0 ps/div. (b) Sample of bit pattern. Vertical scale is 5.00 mw/div. Horizontal scale is ps/div. Figure 4.11: Optical output of MZM, i.e. 10 Gbit/s in Figure 4.5.

95 4.2. Optical Data Channel Generation 77 The eye diagram of Figure 4.11(a) is shown in Figure 4.12 on a larger time scale. Note that although the pulses from the MLFL are roughly 2 ps, they are displayed as having a 7.6 ps FWHM pulse width due to the limited bandwidth of the DCA sampling module. Furthermore, the impulse response of the DCA sampling module causes the ripples in the tail of the pulse. These are not due to actual power fluctuations, and do not appear in autocorrelation measurements of the MLFL pulses. These ripples will affect the eye diagram at 80 Gbit/s. Figure 4.12: Pulses from the MLFL are convolved with the frequency response of the DCA sampling module to yield 7.6 ps FWHM pulses. Vertical scale is 5.00 mw/div. Horizontal scale is 5.0 ps/div. The 10 Gbit/s RZ data is then multiplied using the Optical Clock Multiplier into 40 Gbit/s RZ data. The eye diagram at this point in the Optical Data Channel block is shown in Figure 4.13 at two different time scales.

96 78 Chapter 4. Testbed for High Speed All-Optical Switches (a) Eye diagram. Vertical scale is 3.79 mw/div. Horizontal scale is 2.0 ps/div. (b) Eye diagram of two bit periods. Vertical scale is 3.79 mw/div. Horizontal scale is 5.0 ps/div. Figure 4.13: Eye diagram of the 40 Gbit/s optical output of the Optical Clock Multiplier.

97 4.2. Optical Data Channel Generation 79 Finally, the last 2X multiplier converts the 40 Gbit/s into 80 Gbit/s data as shown in Figure The increased noise on the 1 level in Figure 4.14(b) compared to the 40Gbit/s eye diagram is likely due to the fact that the last 2X multiplier was not actively stabilized as was the Optical Clock Multiplier unit which is a commercial product. The impact of the ripples on the tail of the pulse due to the impulse response of the DCA sampling module is evident in the eye diagram of Figure 4.14(a) as ripples in the zero level. (a) Eye diagram. Vertical scale is 1.82 mw/div. Horizontal scale is 2.0 ps/div. (b) Eye diagram of four bit periods. Vertical scale is 1.82 mw/div. Horizontal scale is 5.0 ps/div. Figure 4.14: Eye diagram of the 80 Gbit/s point in the Optical Data Channel block.

98 80 Chapter 4. Testbed for High Speed All-Optical Switches In summary, the eye diagrams indicate that an 80 Gbit/s PRBS data rate was achieved in the Optical Data Channel block, and that the bit rates of 10 Gbit/s and 40 Gbit/s are also available without any reconfiguration of the testbed. The large eye opening in the 80 Gbit/s eye diagram suggests that the data rate can easily be doubled again to 160 Gbit/s Optical Control Pulse Generation The Optical Control Pulse block of Figure 4.3 has the function of creating high peak power pulses needed to realize a large nonlinear effect, particularly for the AFP switch. The MLFL is used to provide a pulse train for the Optical Control Pulse block which is then amplified to the needed power Energy Requirements for AFP Switch Previous pump probe measurements of the AFP switch [33] indicate that the optimum control pulse switching energy density is between pj/µm 2 for pulses at 1519 nm. The MLFL can be tuned to wavelengths in the C-band telecommunication range (1535 nm nm). Since the control pulse will have a longer wavelength in the testbed than in previous demonstrations, it will likely require a higher energy density due to the lower absorption of the active layer material in the AFP switch. Based on the difference in absorption of the active layer in the AFP switch between 1519 nm and 1535 nm measured in [33], it is estimated that 20% higher energy density will be needed. The measured diameters of the focused beams from the multi-beam system at the AFP switch are roughly 20 µm as shown in Chapter 2. To achieve the required switching energy density, the energy needed per control pulse is 226 pj. For a train of control pulses at a repetition rate of 10 GHz, this corresponds to an average power of 2.26 W. We were concerned that such a high power would cause damage to the AFP switch, particularly given the short pulses from the MLFL. We decided to limit the amount of power to 200 mw or +23 dbm for the control pulse in the proof-of-concept demonstrations being conducted for this thesis. More power can be used once the damage threshold of the AFP switch has been determined.

99 4.3. Optical Control Pulse Generation 81 To give the testbed users control over the peak powers of the control pulses, we designed the Optical Control Pulse block so that bursts of control pulses could be sent instead of one continuous train of pulses. By reducing the number of pulses per second that needed to be amplified by the EDFA, the individual pulse energy can still be high, while keeping the average power low enough to prevent damage to the device under test. A custom, high output power EDFA was built using facilities in our laboratory to amplify the control pulses. Since the input to the custom built EDFA is a burst of pulses, the input average power can be quite low. We need to custom build an EDFA with high gain that would be capable of accepting a low average power input, and outputting a high average power Creating a Burst of Pulses The Optical Control Pulse block s function is to output bursts of high energy pulses. The pulses themselves occur at a high repetition rate, i.e. the period between pulses is short. The period between bursts can be quite long if the testbed user wishes. To test the AFP switch, the time interval between pulses is set to 100 ps (10 GHz). The components used to do this are shown in Figure To achieve a burst of high repetition rate pulse, a MZM (MZM 2) is used to block pulses from the MLFL to the high power EDFA (EDFA 2 in Figure 4.15), thereby creating bursts. The length of the bursts is determined by how long the MZM is held in an ON state. The generation of a bursted electronic control signal for MZM 2 is first generated by the Burst Generator, which is a 2.5 Gbit/s PRBS generator borrowed from one of the undergraduate laboratories. Unlike the 10 Gbit/s PRBS generator used in the Optical Data Channel block, this PRBS generator allows the user to program in their own bit patterns. We selected to use this PRBS generator so that testbed users could have control over the burst length, and burst repetition rate. Custom burst patterns can also be programmed into the Burst Generator which could be useful for demonstration of all-optical logic gates and signal processing.

100 82 Chapter 4. Testbed for High Speed All-Optical Switches Figure 4.15: Components used in Optical Control Pulse block of Figure 4.3 are shaded in grey. The bit rate of the Burst Generator is determined by the input clock frequency. To keep the Burst Generator synchronized with the rest of the testbed, a 4 clock divider is used to convert the 10 GHz sine wave output from the RF synthesizer into a 2.5 GHz sine wave. Because the Burst Generator is designed for 2.5 Gbit/s bit rates, the rise/fall time of the output is slow - roughly 100 ps (defined as 20% to 80%). A slow rise/fall time from the Burst Generator will result in the MZM not fully turning ON or OFF for the leading or trailing pulses in the burst as shown in Figure A rise/fall time of 40 ps or less is needed to prevent this.

101 4.3. Optical Control Pulse Generation 83 Figure 4.16: A slow rise/fall time in the burst control signal to MZM control results in pulses at the beginning and end of the burst being attenuated. Grey triangles represent optical pulses from the MLFL. Dashed lines indicate the burst voltage. The solution to the problem of a slow rise/fall time is to use a high speed D flip-flop (DFF), which was a commercial product from Inphi Corp. The DFF has the net effect of sending out a copy of the burst signal, but with shorter rise/fall times due to the faster circuitry in the DFF. A comparison of the fall times is shown in Figure Figure 4.17(a) shows the fall time of the Burst Generator is 99 ps. The burst is sent to the DFF, and the DFF output is shown in Figure 4.17(b) with a fall time of 30 ps. This is short enough that there should not be any attenuation of pulses at the edges of the burst due to slow rise/fall time. Note that the rise time in Figure 4.17(b) is quite long due to the dip at the top of the rising edge. The dip is likely due to voltage reflections as a result of poor impedance matching between the trace and coaxial connector in the DFF board. To further ensure that the edge of the burst does not coincide with a pulse from the MLFL, a phase shifter is used to adjust the time delay between burst edge and the pulse from the MLFL. A modulator driver is once again used to amplify the DFF output voltage to a high enough level that it can fully turn the MZM ON and OFF. The result of the burst on the pulse train from the MLFL is shown in Figure In this case, a burst 8 pulses long (800 ps) is repeated every 3.2 ns. The ratio of the burst length to the burst repetition ( ) burst length period is referred to as the duty cycle. For Figure 4.18 the duty burst repetition period cycle is 25%. In Figure 4.18(b) some pulses are not visible because of the jitter of the DCA.

102 84 Chapter 4. Testbed for High Speed All-Optical Switches (a) Burst Generator electronic output. Vertical scale is 100 mv/div. Horizontal scale is ps/div. (b) DFF electronic output when the Burst Generator is the input. Vertical scale is 100 mv/div. Horizontal scale is ps/div. Figure 4.17: Electronic output of Burst Generator.

103 4.3. Optical Control Pulse Generation 85 (a) Each burst has 8 pulses from the MLFL. Vertical scale is 2.00 mw/div, Horizontal scale is ps/div. (b) Bursts occur every 3.2 ns. Vertical scale is 2.00 mw/div, Horizontal scale is ns/div. Figure 4.18: The burst of 10 GHz pulses from MZM 2.

104 86 Chapter 4. Testbed for High Speed All-Optical Switches High Power EDFA After the MZM 2 creates a burst of pulses, the burst is sent to the high power EDFA (EDFA 2 in Figure 4.15) for amplification. The output power curve for the EDFA is shown in Figure 4.19 for a CW input at 1550 nm. Figure 4.19: The output power of the high power EDFA for a CW input at 1550 nm at various input powers. The EDFA was designed to amplify low input powers as shown in Figure The reason for this is so that a low duty cycle burst can be used with the EDFA, thus generating very high energy pulses if needed. Note that Figure 4.19 is a measure of the total output power of the EDFA, and not the output signal power alone. At lower input powers, the signal-to-noise ratio in the output signal will be lower as well. To compensate for this, a bandpass filter with an appropriate bandwidth centered around the signal wavelength can be used at the EDFA output to filter out some of the noise. Nonetheless, the signal quality of the pulses is maintained after being amplified. This is shown in Figure 4.20 for a burst length of 3.2 ns (32 pulses at 10 GHz) and a 10% duty cycle, with an average input power to the EDFA of dbm. The pulses shown in Figure 4.20 have an energy of approximately 135 pj per pulse. To operate the AFP switch at a good contrast ratio, duty cycles of less than 6% are needed to get the required energy of 226 pj per pulse. In summary, the Optical Control Pulse block was designed and tested to provide a high

105 4.4. Wavelength Conversion Demonstration 87 Figure 4.20: The last 8 pulses of a 32 pulse burst at the EDFA output. Vertical scale is 5.00 mw/div. Horizontal scale is ps/div. pulse energy, and high repetition rate optical clock that can act as the control signal for demonstrations of all-optical switching devices. The burst feature allows users of the testbed to adjust the pulse energy and average control signal power to their needs, and may be useful in other optical networking experiments such as circulating loop measurements to demonstrate transmission over long fiber lengths Wavelength Conversion Demonstration With the completion of the testbed, high bit rate experiments could now be performed on the AFP switch with the multi-beam system. We performed a high speed wavelength conversion where data at a rate of 10 Gbit/s was converted from one wavelength to another. The setup for this is shown in Figure In this demonstration, the Optical Control Pulse block of Figure 4.21 represents a data channel in an optical network that is to be converted to a different wavelength. The MLFL which generates the pulses for the Optical Control Pulse block was tuned to a wavelength of 1550 nm. A tunable wavelength CW laser is set to a different wavelength, in this case at 1572 nm. This represents the wavelength that the data should be converted to.

106 88 Chapter 4. Testbed for High Speed All-Optical Switches Figure 4.21: Experimental setup for 10 Gbit/s wavelength conversion demonstration. The Optical Data Channel block of the testbed is not needed in this experiment. Pulses from the Optical Control Pulse block will be present whenever a 1 bit is transmitted. 1 bits will turn the AFP switch ON, and allow light from the CW laser to be transmitted to the receiver, in this case the DCA where it can be viewed. In doing so, the AFP switch and multi-beam system can convert a data channel of 1 and 0 bits encoded in OOK from one wavelength to another. The wavelength converted output from the AFP switch is shown in Figure In this demonstration, a 1.6 ns burst (16 pulses at a repetition rate of 10 GHz) was sent from the Optical Control Pulse block every 51.2 ns. Figure 4.22 shows a sample of the burst. The output from the AFP switch clearly shows wavelength conversion was achieved. This is evident from the reshaping of the pulses; essentially what is viewed on the DCA is the recovery time of the AFP switch and the resulting change in transmission of the CW laser. The contrast ratio in this case is 5.2 db, which is low compared to the results we previously obtained in the pump probe experiments of Chapter 3. We believe this is because the alignment of the optics in the multi-beam system during this experiment was not optimized

107 4.4. Wavelength Conversion Demonstration 89 Figure 4.22: Experimental setup for 10 Gbit/s wavelength conversion demonstration. Data at a carrier wavelength of 1550 nm is converted to data at 1572 nm. The Optical Data Channel block of the testbed is not needed in this experiment. Vertical scale is 665 µw/div. Horizontal scale is 50.0 ps/div.

108 90 Chapter 4. Testbed for High Speed All-Optical Switches for maximum beam overlap between the control pulses and the CW laser. Contrast ratios very similar to those that were obtained in the pump probe experiments are possible with a more meticulous alignment. A basic demonstration of 10 Gbit/s wavelength conversion was achieved. This demonstrates the potential of the testbed to perform high bit rate experiments on all-optical switching devices, and also shows that the testbed can be reconfigured to perform a number of different types of experiments. This experiment also shows that the multi-beam system can be used at high bit rates with the AFP switch, and that future multiple channel high speed demonstrations of the AFP switch can be performed with the testbed and multi-beam system as the platform on which to perform these experiments. In this chapter, we presented the design of a high bit rate testbed for testing of ultrafast all-optical switches, and other optical telecommunication devices. The testbed is very flexible and can be adjusted for the particular needs of a given experiment. Variable data rates from 10 Gbit/s up to 80 Gbit/s were demonstrated, and the testbed can easily be upgraded to 160 Gbit/s. Testing of all-optical switches in the testbed is facilitated by providing a high energy, short pulse duration optical clock. High speed experiments have already begun using the testbed, such as the 10 Gbit/s wavelength conversion demonstration of the AFP switch with the multi-beam system. The testbed allows researchers developing all-optical switches to quickly characterize and demonstrate the performance capabilities of their devices.

109 Chapter 5 Conclusions In conclusion, this thesis reports two major advances to the AFP switch demonstrating its potential in future all-optical networks. First, a multi-beam system was designed, built, and demonstrated with the AFP switch. This multi-beam system greatly enhances the functionality of the AFP switch by focusing multiple beams to the same spot on the switch, allowing simultaneous demultiplexing, wavelength conversion, and the realization of many input optical logic gates. The multi-beam system is comprised of a fiber array, collimating lens array, and a focusing lens. The current multi-beam system can deliver four beams to the AFP switch, focused to a spot size of roughly 20 µm, and is fiber terminated for integration into a fiber optic network or testbed. Good contrast ratio, as high as 16.2 db, and fast recovery time as short as 6.8 ps was demonstrated with the multi-beam system and AFP switch. Low crosstalk between channels, less than -50 db, was also demonstrated. A wavelength conversion experiment was performed demonstrating the large bandwidth of the AFP switch with the multi-beam system. Wavelength conversion of pulses from nm was demonstrated. The second advance is the design and building of a high bit rate optical testbed to characterize all-optical switches and other optical telecommunication devices. The testbed has an adjustable OOK RZ bit rate from Gbit/s and the wavelength is tunable from 1535 to 1565 nm. The testbed also provides a bursted, high energy optical clock with short 91

110 92 Chapter 5. Conclusions pulse durations as low as 1.6 ps at FWHM. The testbed is already under use to demonstrate the ultrafast switching capabilities of novel all-optical switches. One such demonstration that we performed was shown in this thesis - a 10 Gbit/s wavelength conversion from 1550 nm to 1572 nm Significance and Contribution For the work presented in Chapter 2, I was responsible for the design of the multi-beam system and determining the specifications of the optics. With the assistance of the students at CREOL, I fabricated the collimating lens array. I also did most of the work in the alignment and characterization of both the lens array and the multi-beam system. For the work presented in Chapter 3, I performed the pump probe measurements using an existing experimental setup. I also performed the crosstalk measurements myself. For the work presented in Chapter 4, I was responsible for the design of the testbed schematic, the setup and integration of the components, and performed the signal quality measurements myself. I was assisted in performing the 10 Gbit/s wavelength conversion. As a result of the work completed in this thesis, one refereed conference paper was published [49]. The significance of this thesis is that multiple channel switching using the AFP switch and multi-beam system has been demonstrated. This shows that the AFP switch with the multi-beam system is a versatile and compact device, capable of of simultaneously performing many operations. High bit rate operation of the AFP switch has also been demonstrated. This shows the potential of the AFP switch for use in future optical networks at high serial bit rates. Lastly, the testbed setup has been prepared that can be used to test a variety of all-optical switching devices and conduct other high speed optical telecommunication experiments. The testbed is a facility that will encourage collaboration with other researchers who wish to make use of its capabilities.

111 5.2. Future Work Future Work The work that was completed in this thesis has laid the foundation for future demonstrations of the multiple channel capabilities of the AFP switch. The current multi-beam system can deliver four beams to the AFP switch which was sufficient for the proof-of-concept demonstrations in this thesis. The number of beams can be increased significantly by using a two dimensional fiber array in a honeycomb arrangement with a pitch of 2 mm to match the collimating lens array. Such fiber arrays are commercially available from a few different vendors [50, 51, 52]. A larger focusing lens, or perhaps an array of focusing lenses will be needed to focus additional channels onto the AFP switch. Integration of the optics in the multi-beam system is another future improvement. Integration of the optics into a robust package will allow the system to be moved without realignment. It should also enhance the system s stability to thermal fluctuations and vibrations. The multi-beam system can be further improved by exploring fabrication methods that allow as many of the components to be grown using the same process. The completion of the high bit rate testbed creates a facility that can be used to demonstrate the high speed operation of the AFP switch with multi-beam system. 40 Gbit/s wavelength conversion has already been achieved by other members in the group using the testbed, and an 80 Gbit/s demonstration is planned for the near future. Simultaneous high speed demultiplexing at 80 Gbit/s of multiple channels is also planned with the AFP switch and multi-beam system.

112 94 Chapter 5. Conclusions

113 Appendix A: Structure of the Asymmetric Fabry-Pérot Switch The physical structure of the AFP switch is shown in Figure A.1. The Bragg reflection stack acts as the high reflectance layer, while the SiON coating acts as the low reflectance layer. Figure A.1: Physical structure of the AFP switch. Drawing is not to scale, and not all the layers in the Bragg reflection stack are shown. The Bragg reflection stack and the saturable absorber are grown using an MBE process in the same chamber, and can be grown one after another. The Bragg reflection stack is first grown using a conventional MBE process. The saturable absorber is then grown in the same chamber, again using an MBE process. However, a high energy helium plasma is injected into the MBE chamber during growth of the saturable absorber. The helium plasma is generated by an Electron Cyclotron Resonance source during growth [34]. Previous experiments have shown that the carrier recovery time of InGaAsP grown in a 95