Rob A. Rutenbar

This book presents a detailed summary of research on automatic layout of device-level analog circuits that was undertaken in the late 1980s and early 1990s at Carnegie Mellon University. We focus on the work behind the creation of the tools called KOAN and ANAGRAM II, which form part of the core of the CMU ACACIA analog CAD system. KOAN is a device placer for custom analog cells; ANANGRAM II a detailed area router for these analog cells. We strive to present the motivations behind the architecture of these tools, including detailed discussion of the subtle technology and circuit concerns that must be addressed in any successful analog or mixed-signal layout tool. Our approach in organizing the chapters of the book has been to present our algo­ rithms as a series of responses to these very real and very difficult analog layout problems. Finally, we present numerous examples of results generated by our algorithms. This research was supported in part by the Semiconductor Research Corpora­ tion, by the National Science Foundation, by Harris Semiconductor, and by the International Business Machines Corporation Resident Study Program. Finally, just for the record: John Cohn was the designer of the KOAN placer; David Garrod was the designer of the ANAGRAM II router (and its predeces­ sor, ANAGRAM I). This book was architected by all four authors, edited by John Cohn and Rob Rutenbar, and produced in finished form by John Cohn.

This book presents a detailed summary of research on automatic layout of device-level analog circuits that was undertaken in the late 1980s and early 1990s at Carnegie Mellon University. We focus on the work behind the creation of the tools called KOAN and ANAGRAM II, which form part of the core of the CMU ACACIA analog CAD system. KOAN is a device placer for custom analog cells; ANANGRAM II a detailed area router for these analog cells. We strive to present the motivations behind the architecture of these tools, including detailed discussion of the subtle technology and circuit concerns that must be addressed in any successful analog or mixed-signal layout tool. Our approach in organizing the chapters of the book has been to present our algo­ rithms as a series of responses to these very real and very difficult analog layout problems. Finally, we present numerous examples of results generated by our algorithms. This research was supported in part by the Semiconductor Research Corpora­ tion, by the National Science Foundation, by Harris Semiconductor, and by the International Business Machines Corporation Resident Study Program. Finally, just for the record: John Cohn was the designer of the KOAN placer; David Garrod was the designer of the ANAGRAM II router (and its predeces­ sor, ANAGRAM I). This book was architected by all four authors, edited by John Cohn and Rob Rutenbar, and produced in finished form by John Cohn.

Cell-based design methodologies have dominated layout generation of digital circuits. Unfortunately, the growing demands for transparent process portability, increased performance, and low-level device sizing for timing/power are poorly handled in a fixed cell library. Direct Transistor-Level Layout For Digital Blocks proposes a direct transistor-level layout approach for small blocks of custom digital logic as an alternative that better accommodates demands for device-level flexibility. This approach captures essential shape-level optimizations, yet scales easily to netlists with thousands of devices, and incorporates timing optimization during layout. The key idea is early identification of essential diffusion-merged MOS device groups, and their preservation in an uncommitted geometric form until the very end of detailed placement. Roughly speaking, essential groups are extracted early from the transistor-level netlist, placed globally, optimized locally, and then finally committed each to a specific shape-level form while concurrently optimizing for both density and routability. The essential flaw in prior efforts is an over-reliance on geometric assumptions from large-scale cell-based layout algorithms. Individual transistors may seem simple, but they do not pack as gates do. Algorithms that ignore these shape-level issues suffer the consequences when thousands of devices are poorly packed. The approach described in this book can pack devices much more densely than a typical cell-based layout.Direct Transistor-Level Layout For Digital Blocks is a comprehensive reference work on device-level layout optimization, which will be valuable to CAD tool and circuit designers.

Practical Synthesis of High-Performance Analog Circuits presents a technique for automating the design of analog circuits. Market competition and the astounding pace of technological innovation exert tremendous pressure on circuit design engineers to turn ideas into products quickly and get them to market. In digital Application Specific Integrated Circuit (ASIC) design, computer aided design (CAD) tools have substantially eased this pressure by automating many of the laborious steps in the design process, thereby allowing the designer to maximise his design expertise. But the world is not solely digital. Cellular telephones, magnetic disk drives, neural networks and speech recognition systems are a few of the recent technological innovations that rely on a core of analog circuitry and exploit the density and performance of mixed analog/digital ASICs. To maximize profit, these mixed-signal ASICs must also make it to market as quickly as possible. However, although the engineer working on the digital portion of the ASIC can rely on sophisticated CAD tools to automate much of the design process, there is little help for the engineer working on the analog portion of the chip. With the exception of simulators to verify the circuit design when it is complete, there are almost no general purpose CAD tools that an analog design engineer can take advantage of to automate the analog design flow and reduce his time to market. Practical Synthesis of High-Performance Analog Circuits presents a new variation-tolerant analog synthesis strategy that is a significant step towards ending the wait for a practical analog synthesis tool. A new synthesis strategy is presented that can fully automate the path from a circuit topology and performance specifications to a sized variation-tolerant circuit schematic. This strategy relies on asymptotic waveform evaluation to predict circuit performance and simulated annealing to solve a novel non-linear infinite programming optimization formulation of the circuit synthesis problem via a sequence of smaller optimization problems. Practical Synthesis of High-Performance Analog Circuits will be of interest to analog circuit designers, CAD/EDA industry professionals, academics and students.

As VLSI technology moves to the nanometer scale for transistor feature sizes, the impact of manufacturing imperfections result in large variations in the circuit performance. Traditional CAD tools are not well-equipped to handle this scenario, since they do not model this statistical nature of the circuit parameters and performances, or if they do, the existing techniques tend to be over-simplified or intractably slow. Novel Algorithms for Fast Statistical Analysis of Scaled Circuits draws upon ideas for attacking parallel problems in other technical fields, such as computational finance, machine learning and actuarial risk, and synthesizes them with innovative attacks for the problem domain of integrated circuits. The result is a set of novel solutions to problems of efficient statistical analysis of circuits in the nanometer regime.

In the early days of VLSI, the design of the power distribution for an integrated cir­ cuit was rather simple. Power distribution --the design of the geometric topology for the network of wires that connect the various power supplies, the widths of the indi­ vidual segments for each of these wires, the number and location of the power I/O pins around the periphery of the chip --was simple because the chips were simpler. Few available wiring layers forced floorplans that allowed simple, planar (non-over­ lapping) power networks. Lower speeds and circuit density made the choice of the wire widths easier: we made them just fat enough to avoid resistive voltage drops due to switching currents in the supply network. And we just didn't need enormous num­ bers of power and ground pins on the package for the chips to work. It's not so simple any more. Increased integration has forced us to focus on reliability concerns such as metal elec­ tromigration, which affects wire sizing decisions in the power network. Extra metal layers have allowed more flexibility in the topological layout of the power networks.

This book, and the research it describes, resulted from a simple observation we made sometime in 1986. Put simply, we noticed that many VLSI design tools looked "alike". That is, at least at the overall software architecture level, the algorithms and data structures required to solve problem X looked much like those required to solve problem X'. Unfortunately, this resemblance is often of little help in actually writing the software for problem X' given the software for problem X. In the VLSI CAD world, technology changes rapidly enough that design software must continually strive to keep up. And of course, VLSI design software, and engineering design software in general, is often exquisitely sensitive to some aspects of the domain (technology) in which it operates. Modest changes in functionality have an unfortunate tendency to require substantial (and time-consuming) internal software modifications. Now, observing that large engineering software systems are technology­ dependent is not particularly clever. However, we believe that our approach to xiv Preface dealing with this problem took an interesting new direction. We chose to investigate the extent to which automatic programming ideas cold be used to synthesize such software systems from high-level specifications. This book is one of the results of that effort.

As VLSI technology moves to the nanometer scale for transistor feature sizes, the impact of manufacturing imperfections result in large variations in the circuit performance. Traditional CAD tools are not well-equipped to handle this scenario, since they do not model this statistical nature of the circuit parameters and performances, or if they do, the existing techniques tend to be over-simplified or intractably slow. Novel Algorithms for Fast Statistical Analysis of Scaled Circuits draws upon ideas for attacking parallel problems in other technical fields, such as computational finance, machine learning and actuarial risk, and synthesizes them with innovative attacks for the problem domain of integrated circuits. The result is a set of novel solutions to problems of efficient statistical analysis of circuits in the nanometer regime.

Cell-based design methodologies have dominated layout generation of digital circuits. Unfortunately, the growing demands for transparent process portability, increased performance, and low-level device sizing for timing/power are poorly handled in a fixed cell library. Direct Transistor-Level Layout For Digital Blocks proposes a direct transistor-level layout approach for small blocks of custom digital logic as an alternative that better accommodates demands for device-level flexibility. This approach captures essential shape-level optimizations, yet scales easily to netlists with thousands of devices, and incorporates timing optimization during layout. The key idea is early identification of essential diffusion-merged MOS device groups, and their preservation in an uncommitted geometric form until the very end of detailed placement. Roughly speaking, essential groups are extracted early from the transistor-level netlist, placed globally, optimized locally, and then finally committed each to a specific shape-level form while concurrently optimizing for both density and routability. The essential flaw in prior efforts is an over-reliance on geometric assumptions from large-scale cell-based layout algorithms. Individual transistors may seem simple, but they do not pack as gates do. Algorithms that ignore these shape-level issues suffer the consequences when thousands of devices are poorly packed. The approach described in this book can pack devices much more densely than a typical cell-based layout.Direct Transistor-Level Layout For Digital Blocks is a comprehensive reference work on device-level layout optimization, which will be valuable to CAD tool and circuit designers.

The tools and techniques you need to break the analog design bottleneck! Ten years ago, analog seemed to be a dead-end technology. Today, System-on-Chip (SoC) designs are increasingly mixed-signal designs. With the advent of application-specific integrated circuits (ASIC) technologies that can integrate both analog and digital functions on a single chip, analog has become more crucial than ever to the design process. Today, designers are moving beyond hand-crafted, one-transistor-at-a-time methods. They are using new circuit and physical synthesis tools to design practical analog circuits; new modeling and analysis tools to allow rapid exploration of system level alternatives; and new simulation tools to provide accurate answers for analog circuit behaviors and interactions that were considered impossible to handle only a few years ago. To give circuit designers and CAD professionals a better understanding of the history and the current state of the art in the field, this volume collects in one place the essential set of analog CAD papers that form the foundation of today’s new analog design automation tools. Areas covered are: *Analog synthesis *Symbolic analysis *Analog layout *Analog modeling and analysis *Specialized analog simulation *Circuit centering and yield optimization *Circuit testing Computer-Aided Design of Analog Integrated Circuits and Systems is the cutting-edge reference that will be an invaluable resource for every semiconductor circuit designer and CAD professional who hopes to break the analog design bottleneck.

Practical Synthesis of High-Performance Analog Circuits presents a technique for automating the design of analog circuits. Market competition and the astounding pace of technological innovation exert tremendous pressure on circuit design engineers to turn ideas into products quickly and get them to market. In digital Application Specific Integrated Circuit (ASIC) design, computer aided design (CAD) tools have substantially eased this pressure by automating many of the laborious steps in the design process, thereby allowing the designer to maximise his design expertise. But the world is not solely digital. Cellular telephones, magnetic disk drives, neural networks and speech recognition systems are a few of the recent technological innovations that rely on a core of analog circuitry and exploit the density and performance of mixed analog/digital ASICs. To maximize profit, these mixed-signal ASICs must also make it to market as quickly as possible. However, although the engineer working on the digital portion of the ASIC can rely on sophisticated CAD tools to automate much of the design process, there is little help for the engineer working on the analog portion of the chip. With the exception of simulators to verify the circuit design when it is complete, there are almost no general purpose CAD tools that an analog design engineer can take advantage of to automate the analog design flow and reduce his time to market. Practical Synthesis of High-Performance Analog Circuits presents a new variation-tolerant analog synthesis strategy that is a significant step towards ending the wait for a practical analog synthesis tool. A new synthesis strategy is presented that can fully automate the path from a circuit topology and performance specifications to a sized variation-tolerant circuit schematic. This strategy relies on asymptotic waveform evaluation to predict circuit performance and simulated annealing to solve a novel non-linear infinite programming optimization formulation of the circuit synthesis problem via a sequence of smaller optimization problems. Practical Synthesis of High-Performance Analog Circuits will be of interest to analog circuit designers, CAD/EDA industry professionals, academics and students.

In the early days of VLSI, the design of the power distribution for an integrated cir­ cuit was rather simple. Power distribution --the design of the geometric topology for the network of wires that connect the various power supplies, the widths of the indi­ vidual segments for each of these wires, the number and location of the power I/O pins around the periphery of the chip --was simple because the chips were simpler. Few available wiring layers forced floorplans that allowed simple, planar (non-over­ lapping) power networks. Lower speeds and circuit density made the choice of the wire widths easier: we made them just fat enough to avoid resistive voltage drops due to switching currents in the supply network. And we just didn't need enormous num­ bers of power and ground pins on the package for the chips to work. It's not so simple any more. Increased integration has forced us to focus on reliability concerns such as metal elec­ tromigration, which affects wire sizing decisions in the power network. Extra metal layers have allowed more flexibility in the topological layout of the power networks.

This book, and the research it describes, resulted from a simple observation we made sometime in 1986. Put simply, we noticed that many VLSI design tools looked "alike". That is, at least at the overall software architecture level, the algorithms and data structures required to solve problem X looked much like those required to solve problem X'. Unfortunately, this resemblance is often of little help in actually writing the software for problem X' given the software for problem X. In the VLSI CAD world, technology changes rapidly enough that design software must continually strive to keep up. And of course, VLSI design software, and engineering design software in general, is often exquisitely sensitive to some aspects of the domain (technology) in which it operates. Modest changes in functionality have an unfortunate tendency to require substantial (and time-consuming) internal software modifications. Now, observing that large engineering software systems are technology­ dependent is not particularly clever. However, we believe that our approach to xiv Preface dealing with this problem took an interesting new direction. We chose to investigate the extent to which automatic programming ideas cold be used to synthesize such software systems from high-level specifications. This book is one of the results of that effort.