Page 1123 - The Mechatronics Handbook
P. 1123

Embedded microprocessors have been categorized into DSP processors and embedded CPUs due to
                                 historic reasons. DSP processors have been designed and marketed as special-purpose devices that are
                                 mostly programmed by hand to perform digital signal processing computations. A recent trend in the
                                 DSP market is to use compilers to alleviate the need for tedious hand-coding in DSP development. Another
                                 recent trend in the DSP market is toward integrating a DSP processor core with application-specific logic
                                 to form a single-chip solution. This approach is enabled by the fast increasing chip density technology.
                                 The major benefit is reduced system cost and energy consumption. Two general types of DSP cores are
                                 available to application developers today. Foundry-captive DSP cores and related application-specific logic
                                 design services are provided by major semiconductor vendors such as Texas Instruments, Lucent Technol-
                                 ogies, and SGS-Thompson to application developers who commit to their fabrication lines. A very large
                                 volume commitment is usually required to use the design service. Licensable DSP cores are provided by
                                 small to medium design houses to application developers who want to be able to choose fabrication lines.
                                   There are several ways that the needs of embedded computing differ from those of the more traditional
                                 general-purpose systems. Constraints on the code size, weight, and power consumption place stringent
                                 requirements on embedded processors and the software they execute. Also, constraints rooted in real-
                                 time requirements are often a significant consideration in many embedded systems. Furthermore, cost
                                 is a severe constraint on embedded processors.
                                   Embedded CPUs are used in products where the computation involved resembles that of general-
                                 purpose applications and operating systems. Embedded CPUs have been traditionally derived from out-
                                 of-date computer microprocessors. They often reuse the compiler and related software support developed
                                 for their computer cousins. Recycling the microprocessor design and compiler software minimizes engi-
                                 neering cost. A trend in the embedded CPU domain is similar to that in the DSP domain: to provide
                                 embedded CPU cores and application specific logic design services to form single-chip solutions. For
                                 example, MIPs customized its embedded CPU core for use in Nintendo64, in return for engineering fees
                                 and royalty streams. ARM, NEC, and Hitachi offer similar products and services. Due to an increasing
                                 need to perform DSP computation in consumer and telecommunication products, an increasing number
                                 of embedded CPUs have extensions to enable more effective DSP computation.
                                   Contrary to the different constraints and product markets, both computer and embedded micropro-
                                 cessors share traditional elements of computer architecture. These main elements will be described.
                                 Additionally, over the past decade, substantial research has gone into the design of microprocessors
                                 embodying parallelism at the instruction level, as well as aggressive compiler optimization and analysis
                                 techniques for harnessing this opportunity. Much of this effort has since been validated through the
                                 proliferation of mainstream general-purpose computers based on these technologies. Nevertheless, grow-
                                 ing demand for high performance in embedded computing systems is creating new opportunities to
                                 leverage these techniques in application-specific domains. The research of Instruction-Level Parallelism
                                 (ILP) has developed a distinct architecture methodology referred to as Explicitly Parallel Instruction
                                 Computing (EPIC) technology. Overall, these techniques represent fundamental substantial changes in
                                 computer architecture.


                                 42.3 Major Components of a Microprocessor

                                 The main hardware of a microprocessor system can be divided into sections according to their function-
                                 alities. A popular approach is to divide a system into four subsystems: the central processor, the memory
                                 subsystem, the input/output (I/O) subsystem, and the system interconnection.  Figure 42.1 shows the
                                 connection between these subsystems. The main components and characteristics of these subsystems will
                                 be described.

                                 Central Processor

                                 A modern microprocessor’s central processor system can typically be further divided into control, data
                                 path, pipelining, and branch prediction hardware.

                                 ©2002 CRC Press LLC
   1118   1119   1120   1121   1122   1123   1124   1125   1126   1127   1128