A microarchitecture for resource-limited superscalar microprocessors.
T.D. Basso
Abstract
T.D. Basso
Abstract
Microelectronic components in space and satellite systems must be resistant to total dose radiation, single-even upset, and latchup in order to accomplish their missions. The demand for inexpensive, high-volume, radiation hardened (rad-hard) integrated circuits (ICs) is expected to increase dramatically as the communication market continues to expand. Motorola's Complementary Gallium Arsenide (CGaAs TM ) technology offers superior radiation tolerance compared to traditional CMOS processes, while being more economical than dedicated rad-hard CMOS processes. The goals of this dissertation are to optimize a superscalar microarchitecture suitable for CGaAs TM microprocessors, develop circuit techniques for such applications, and evaluate the potential of CGaAs TM for the development of digital VLSI circuits. Motorola's 0.5 mum CGaAs TM process is summarized and circuit techniques applicable to digital CGaAs TM are developed. Direct coupled FET, complementary, and domino logic circuits are compared based on speed, power, area, and noise margins. These circuit techniques are employed in the design of a 600 MHz PowerPC TM arithmetic logic unit. The dissertation emphasizes CGaAS TM -specific design considerations, specifically, low integration level. A baseline superscalar microarchitecture is defined and SPEC95 integer benchmark simulations are used to evaluate the applicability of advanced architectural features to microprocessors having low integration levels. The performance simulations center around the optimization of a simple superscalar core, small-scale branch prediction, instruction prefetching, and an off-chip primary data cache. The simulation results are used to develop a superscalar microarchitecture capable of outperforming a comparable sequential pipeline, while using only 500,000 transistors. The architecture, running at 200 MHz, is capable of achieving an estimated 153 MIPS, translating to a 27% performance increase over a comparable traditional pipelined microprocessor. The proposed microarchitecture is process independent and can be applied to low-cost, or transistor-limited applications. The proposed microarchitecture is implemented in the design of a 0.35 mum CMOS microprocessor, and the design of a 0.5 mum CGaAs TM micro-processor. The two technologies and designs are compared to ascertain the state of CGaAs TM for digital VLSI applications.
OpenAlex reports 1 citations for this work. Citation counts describe recorded attention and do not establish research quality.
A contribution statement is not available in the OpenAlex record.
Method details are not available in the OpenAlex metadata.
Findings are not separately available in the OpenAlex metadata.
Limitations are not available in the OpenAlex metadata.
Application details are not available in the OpenAlex metadata.
Microelectronic components in space and satellite systems must be resistant to total dose radiation, single-even upset, and latchup in order to accomplish their missions. The demand for inexpensive, high-volume, radiation hardened (rad-hard) integrated circuits (ICs) is expected to increase dramatically as the communication market continues to expand. Motorola's Complementary Gallium Arsenide (CGaAs TM ) technology offers superior radiation tolerance compared to traditional CMOS processes, while being more economical than dedicated rad-hard CMOS processes. The goals of this dissertation are to optimize a superscalar microarchitecture suitable for CGaAs TM microprocessors, develop circuit techniques for such applications, and evaluate the potential of CGaAs TM for the development of digital VLSI circuits. Motorola's 0.5 mum CGaAs TM process is summarized and circuit techniques applicable to digital CGaAs TM are developed. Direct coupled FET, complementary, and domino logic circuits are compared based on speed, power, area, and noise margins. These circuit techniques are employed in the design of a 600 MHz PowerPC TM arithmetic logic unit. The dissertation emphasizes CGaAS TM -specific design considerations, specifically, low integration level. A baseline superscalar microarchitecture is defined and SPEC95 integer benchmark simulations are used to evaluate the applicability of advanced architectural features to microprocessors having low integration levels. The performance simulations center around the optimization of a simple superscalar core, small-scale branch prediction, instruction prefetching, and an off-chip primary data cache. The simulation results are used to develop a superscalar microarchitecture capable of outperforming a comparable sequential pipeline, while using only 500,000 transistors. The architecture, running at 200 MHz, is capable of achieving an estimated 153 MIPS, translating to a 27% performance increase over a comparable traditional pipelined microprocessor. The proposed microarchitecture is process independent and can be applied to low-cost, or transistor-limited applications. The proposed microarchitecture is implemented in the design of a 0.35 mum CMOS microprocessor, and the design of a 0.5 mum CGaAs TM micro-processor. The two technologies and designs are compared to ascertain the state of CGaAs TM for digital VLSI applications.
Key concepts: Superscalar, Microarchitecture, Computer science, Computer architecture, Parallel computing, Resource (disambiguation), Speculative execution, Computer network