The internet has become the ubiquitous tool that has transformed the lives of all of us. New broadband applications in the field of entertainment, commerce, industry, healthcare and social interactions demand increasingly higher data rates and quality of the networks and ICT infrastructure. In addition, high definition video streaming and cloud services will continue to push the demand for bandwidth. These applications are reshaping the internet into a content-centric network. The challenge is to transform the telecom optical networks and data centers such that they can be scaled efficiently, at low cost. Furthermore, from both an environmental and economic perspective, this scaling should go hand in hand with reduced power consumption. This stems from the desire to reduce CO2 emission and to reduce network operating costs while offering the same service level as today. In the current architecture of the internet, end-users connect to the public network using the access network of an internet service provider (ISP). Today, this access network either reuses the legacy copper or coaxial network or uses passive optical network (PON) technologies, among which the PON is the most energy efficient and provides the highest data rates. Traffic from the access network is aggregated with Ethernet switches and routed to the core network through the provider edge routers, with broadband network gateways (BNGs) to regulate access and usage. These regional links are collectively called the metro network. Data centers connect to the core network using their own dedicated gateway router. The problem of increasing data rates, while reducing the economic and environmental impact, has attracted considerable attention. The research described in this work has been performed in the context of two projects part of the European Union Seventh Framework Programme (FP7), which both aim for higher data rates and tight integration while keeping power consumption low. Mirage targets data center applications while C3PO focuses on medium-reach networks, such as the metro network. Specifically, this research considers two aspects of the high-speed optical receivers used in the communication networks: increasing dynamic range of a linear receiver for multilevel modulation through automatic gain control (AGC) and integration of multiple channels on a single chip with a small area footprint. The data centers of today are high-density computing facilities that provide storage, processing and software as a service to the end-user. They are comprised of gateway routers, a local area network, servers and storage. All of this is organized in racks. The largest units contain over 100 000 servers. The major challenges regarding data centers are scalability and keeping up with increasing amounts of traffic while reducing power consumption (of the devices as well as the associated cooling) and keeping cost minimal. Presently, racks are primarily interconnected with active optical cables (AOCs) which employ signal rates up to 25 Gb/s per lane with non-return-to-zero (NRZ) modulation. A number of technological developments can be employed in AOCs of the future to provide terabit-capacity optical interconnects over longer distances. One such innovation is the use of multilevel modulation formats, which are more bandwidth-efficient than traditional NRZ modulation. Multilevel modulation requires a linear amplifier as front-end of the optical receiver. The greater part of this dissertation discusses the design and implementation of an AGC system for the data path of a linear transimpedance amplifier (TIA). The metro network is the intermediate regional network between the access and core network of the internet architecture, with link lengths up to 500 km. It is estimated that in the near future metro-traffic will increase massively. This growth is attributed mainly to increasing traffic from content delivery networks (CDNs) and data centers, which bypass the core network and directly connect to the metro network. Internet video growth is the major reason for traffic increase. This evolution demands increasingly higher data rates. Today, dense wavelength division multiplexing (DWDM) is widely recognized as being necessary to provide data capacity scalability for future optical networks, as it allows for much higher combined data rates over a single fiber. At the receiver, each wavelength of the demultiplexed incoming light is coupled to a photo diode in a photo diode array which is connected to a dedicated lane of a multichannel receiver. The high number of channels requires small physical channel spacing and tight integration of the diode array with the receiver. In addition, active cooling should be avoided, such that power consumption per receiver lane must be kept low in order not to exceed thermal operation limits. The second component of this work presents the development of an integrated four-channel receiver, targeting 4 × 25 Gb/s data rate, with low power consumption and small footprint to support tight integration with a p-i-n photo diode array with a 250 μm channel pitch. Chapter 1 discusses the impact of increasing data rates and the desire to reduce power consumption on the design of the optical receiver component, in wide metropolitan area networks as well as in short-reach point-to-point links in data centers. In addition, some aspects of integrated analog circuit design are highlighted: the design flow, transistor hand models, a software design tool. Also, an overview of the process technology is given. Chapter 2 provides essential optical receiver concepts, which are required to understand the remainder of the work. Fundamentals of feedback AGC systems are discussed in the first part of Chapter 3. A basic system model is presented in the continuous-time domain, in which the variable gain amplifier (VGA) constitutes the multistage datapath of a linear optical receiver. To enable reliable reception of multilevel modulation formats, the VGA requires controlled frequency response and in particular limited time-domain overshoot across the gain range. It is argued that this control is hard to achieve with fully analog building blocks. Therefore, an event-driven approach is proposed as an extension of the continuous-time system. Both the structural and behavioral aspects are discussed. The result is a system model of a quantized AGC loop, upon which the system-level design, presented in Chapter 4, is based. In turn, Chapter 5 discusses the detailed implementation of the various building blocks on the circuit level and presents experimental results that confirm the feasibility of the proposed approach.
Chapter 6 discusses the design and implementation of a 4 × 25 Gb/s optical receiver array for NRZ modulation with a small area footprint. The focus lies on the input stages and techniques to extend bandwidth and dynamic range are presented. Measurement results for NRZ and optical duobinary (ODB) modulation are presented, as well as the influence of crosstalk on the performance. Finally, Chapter 7 provides an overview of the foremost conclusions of the presented research and includes suggestions for future research. Two appendices are included. Appendix A gives an overview of the general network theorem (GNT), which is used throughout this work and which has been implemented numerically. The results from Appendix B, the analysis of a two-stage opamp compensated with capacitance multipliers, were used to design a building block for the AGC system.