Skip to main content
main-content

Über dieses Buch

This work began in 1995 as an outgrowth of the InfoPad project which showed us that in order to reduce the energy consumption of a portable multimedia terminal that something had to be done about the consumption of the microprocessor subsystem. The design of the InfoPad attempted to reduce the requirements of this general pur­ pose processor by moving the computation into the network or by the use of highly optimized integrated circuits, but in spite of these efforts it still was a major consumer of energy. The reasons for this became apparent as we determined that the energy required to perform a function in dedicated hardware could be several orders of magnitude lower than that consumed in the InfoPad microprocessor. We therefore set out on a full fledged attack on all aspects of the microprocessor energy consumption [1 J. After considerable analysis it became clear that though better circuit design and a stream­ lined architecture would assist in our goal of energy reduction, that the biggest gains were to be found by operating at reduced voltages. For the busses and VO this could be accomplished without significant degradation of the processor performance, but this was not a straightforward solution when applied to the core of the processor sub­ system (CPU and memory).

Inhaltsverzeichnis

Frontmatter

Chapter 1. Introduction

Abstract
The explosive proliferation of portable electronic devices has compelled energy-efficient VLSI and system design to provide longer battery run-times, and more powerful products that require ever-increasing computational complexity. In addition, the demand for low-cost and small form-factor devices has kept the available energy supply roughly constant by driving down battery size, despite advances in battery technology which have increased battery energy density. Thus, energy-efficient design must continuously provide more performance per watt.
Thomas D. Burd, Robert W. Brodersen

Chapter 2. Energy Efficient Design

Abstract
To effectively optimize the energy efficiency of a processor system, it is critical to first understand the computational demands placed upon it and the usage model of the processor. This information is then coupled with simple CMOS circuit models suitable for deep sub-micron process technologies to define metrics that can be used to derive energy-efficient design principles.
Thomas D. Burd, Robert W. Brodersen

Chapter 3. Microprocessor System Architecture

Abstract
While it is important to be aware of energy issues at all levels of the design hierarchy, energy-efficiency optimizations at the level of architectural design generally yield the largest gains. The closer the design approaches to the final physical implementation, the more difficult the gains become because the scope of possible optimizations narrows.
Thomas D. Burd, Robert W. Brodersen

Chapter 4. Circuit Design Methodology

Abstract
The key to energy-efficient circuit implementation, much like architecture and system design, is to focus on energy consumption throughout the entire process, rather than addressing it only as the design nears completion. There are a number of simple rules that will yield an energy-efficient circuit implementations that will be described in this Chapter. Also, if a set of simple rules are followed most CMOS circuits can be made to be robust against the voltage variations of DVS which will also be presented. To demonstrate these principles the design of several complex blocks such as the arithmetic and memory circuits will be described in detail. Because of the importance of bus transceivers, a discussion of an ultra-low energy design will also be given.
Thomas D. Burd, Robert W. Brodersen

Chapter 5. Energy Driven Design Flow

Abstract
The most critical aspect of energy-efficient design is to be energy conscious throughout the entire design flow. A typical design flow treats energy consumption as an afterthought, and is not thoroughly analyzed until the design has reached the transistor schematic stage. This is too late in the design process for radical modifications that can lead to a more energy-efficient implementation. Therefore, much as performance is analyzed at the initial high-level specification of the design, so must energy consumption be analyzed, as well. The primary goal of this design flow is to evaluate energy consumption early on so that the largest energy reductions can be attained.
Thomas D. Burd, Robert W. Brodersen

Chapter 6. Microprocessor and Memory IC’s

Abstract
The chip’s processor core implements the ARM V4 instruction set architecture (ISA) [6.1]. The implementation was derived from an RTL behavioral model (provided by ARM Ltd.) which fixed both the ISA as well as the processor core interface. However, both the custom physical implementation of the core, as well as the rest of the microprocessor design, were fully optimized for energy efficiency.
Thomas D. Burd, Robert W. Brodersen

Chapter 7. DC-DC Voltage Conversion

Abstract
As seen in previous chapters, use of the appropriate voltage for the performance required is one of the most powerful techniques for improving energy efficiency. This can be either done statically by choosing the single most appropriate voltage for the throughput required or dynamically by using information obtained by the operating system and applications. This chapter introduces switching regulators which allow translations of DC voltages with efficiencies of over 90%. Design equations and closed-form expressions for losses are presented for the three basic low-voltage CMOS switching regulator topologies — buck, boost, and buck-boost — controlled via pulse-width or pulse-frequency modulation. This then will be followed for the enhancements to the basic design that are required for efficient dynamic voltage regulation.
Thomas D. Burd, Robert W. Brodersen

Chapter 8. DC-DC Converter IC for DVS

Abstract
In a DVS microprocessor subsystem, the processor core and surrounding peripherals are run from a dynamically scaled voltage supply, enabling up to a 10x improvement in average energy per operation. This section describes the implementation of a prototype dynamic DC-DC converter for application in a DVS system along with measured results.
Thomas D. Burd, Robert W. Brodersen

Chapter 9. DVS System Design and Results

Abstract
A complete embedded microprocessor system was designed and implemented to demonstrate the processor system design methodology described in the previous chapters. By combining Dynamic Voltage Scaling with energy-efficient architecture and circuit design, the system is able to demonstrate more than an order of magnitude improvement in energy efficiency over more conventionally implemented designs.
Thomas D. Burd, Robert W. Brodersen

Chapter 10. Software and Operating System Support

Abstract
The basic goal of energy reduction from the software standpoint is to maximize the battery lifetime of portable general-purpose microprocessor devices by reducing the energy necessary to complete a given task without significantly changing system behavior. Software energy reduction techniques can be divided into two categories: static, which optimizes software before it is executed, and dynamic, which alters the operation of the device at run-time. Static energy reduction techniques for a microprocessor in a general-purpose system can be divided into two categories: high-level application design and compile-time optimization. Dynamic techniques rely on the software running in a portable electronic device to monitor and adjust the device operation at run-time. For example, a laptop that can turn off its LCD backlight requires a dynamic algorithm to determine when the backlight is not needed. Typically, dynamic techniques also require some modification to the base hardware, i.e. the ability to turn off the LCD display, which might not inherently reduce energy consumption. DVS is a dynamic technique that controls the speed of the CPU.
Thomas D. Burd, Robert W. Brodersen

Chapter 11. Conclusions

Abstract
Processor systems are widely prevalent in portable devices which demand increasingly higher levels of energy-efficiency. However, processor energy-efficiency has lagged behind custom ASIC’s and DSP chips, such that while the processor carries only a fraction of the computational load, it is a significant, if not dominant, component of the overall system energy consumption. In order to address this problem it is necessary to use a design methodology that incorporates energy consumption as a primary consideration in all stages of the design process, from the software down to the circuit design. A critical component of this methodology is maximizing the degrees of design freedom by taking into account the characteristics of the application. If this is done, it is possible to significantly improve processor energy-efficiency, thereby enabling smaller, more powerful, and longer running portable devices.
Thomas D. Burd, Robert W. Brodersen

Backmatter

Weitere Informationen