This thesis spans two levels of the design process by examining optimization at both the register-transfer level and at the logic level. More specifically, this thesis addresses the following two problems: 1) performing logic synthesis for custom layout rather than the traditional approach that focuses on synthesis for standard cells, and 2) performing optimization for custom layout from register-transfer level netlists. Thus optimization is performed on the microarchitecture design and at a lower level for individual microarchitecture components.

First, techniques are introduced for generating gate-level netlists that take advantage of custom layout capabilities. Such techniques include limiting serial/parallel transistor chains, transistor sizes, and capacitive loads in forming complex gates. These considerations have not been incorporated in previous logic synthesis systems.

Second, techniques are introduced for improving the microarchitecture structure and using estimates from lower-level optimization tools to guide microarchitecture design optimizations that attempt to meet user specified area and time constraints. These techniques include the capability for mixing layout styles such as custom layout for random-logic components and bit-slicing for regularly structured components. In this manner the entire design, control logic and datapath, can be optimized at the same time. Further, this paper presents a new methodology for microarchitecture-level optimization that greatly reduces the amount of technology-specific knowledge necessary to perform the optimizations.

In this report we discuss strengths and weaknesses of logic synthesis systems and describe a system for microarchitectural and logic optimization. Our system uses a set of algorithms for synthesizing SSI/MSI macros from parameterized microarchitecture components. In addition, it uses rules for optimizing both at the microarchitecture and logic level. The system increases designer productivity and requires less design knowledge and experience from circuit engineers.

This report describes different strategies for area, power, and time optimization for designs on microarchitectural and logic levels as implemented in MILO system.

In recent years the drive to produce more complex integrated circuits while spending less design time has driven the demand for design automation tools. The search for design automation methods has resulted in the design of numerous behavioral synthesis and logic synthesis tools. This report describes a system that fills the gap between traditional behavioral synthesis and logic synthesis tools. Techniques are introduced for improving the microarchitecture structure and using feedback from lower-level optimization tools to guide design optimizations while attempting to meet user specified area and time constraints. These techniques include the capability for mixing layout styles such as custom layout for random-logic components and bit-slicing for regularly structured components. In this manner the entire design, control logic and datapath, can be optimized at the same time. Further, this paper presents a new methodology for microarchitecture-level optimization that greatly reduces the amount of technology-specific knowledge necessary to perform the optimizations.

This paper describes a method for incorporating layout parameters to better meet performance contraints. We define an algorithm for synthesis of high-performance designs and present a synthesis tool for use with custom layout generators. Experimental results indicate such an approach produces faster layouts and permits greater area/time tradeoffs than traditional logic synthesis systems.

This paper describes a novel algorithm for automatic transistor sizing which is one technique for improving timing performance in CMOS circuits. The sizing algorithm is used to minimize area and power subject to timing constraints. We define the transistor sizing problem as a graph problem and use a non-linear optimization technique. The algorithm consists of three separate tasks: critical path analysis, transistor sizing and transistor desizing. The main contribution of the presented algorithm is that the delays of all paths in a given design can be tuned simultaneously to satisfy timing constraints. Furthermore, the minimal transistor area and minimal power dissipation under giving timing constraints can be achieved. Experimental results show that this approach has greater control over area/time tradeoffs than traditional sizing algorithms.

This report describes the design process and synthesis tools used in the UC Irvine CADLAB design environment to design a representative benchmark. The steps taken and rationale used in each stage of the design process are discussed. The benchmark is initially described using a VHDL behavioral description; results produced by each intermediate tool are presented, showing the system flow and integration of tools. The final silicon layout is performed in 3 micron CMOS technology.