Um MPSOC GALS baseado em rede intrachip com geração local de relógio

Abstract

Due to the evolution of deep submicron technologies for semiconductor fabrication, it is possible nowadays to manufacture increasingly complex systems inside a single sili-con die. However, this evolution in some cases mandates the abandonment of traditional design techniques. The development of purely synchronous complex systems begins to be influenced by relatively long intrachip distances as well as by parasitic effects in wires with growingly small cross-sections. Besides, it is important to enable the design of devices with enhanced processing capabilities to fulfill the demand for multiple applications in re-search and industry environments, while at the same time improving energy efficiency. This is motivated by the significant increase on the demand for multifunctional portable equipments like tablets and smart phones that must everyday become faster and yet present reasonable battery life. In view of these facts, new paradigms for the design of globally asynchronous locally synchronous (GALS) systems come to the forefront in the construction of multiprocessor systems on chip (MPSoCs). This work has as main strateg-ic objective to explore GALS MPSoC architectures that target the control of power dissipa-tion. The decision to work with MPSoCs comes from the natural need to increase the number of processing elements in current designs, as a way to take full advantage of the silicon technological evolution. During the development of this work five distinct contribu-tions are worth mentioning. First, the architectures of the Hermes-GLP NoC router and of the HeMPS MPSoC were subject to a set of corrections and modifications, to provide these modules with better support to the implementation of GALS systems. This allowed the proposition of a new MPSoCs, called HeMPS-GLP. Second, a set of changes in the embedded processor microkernel of the HeMPS MPSoC enabled the smooth interconnec-tion and configuration of new hardware structures to the system processors. Third, a new high-level language verification environment for the HeMPS-GLP MPSoC was made avail-able, which supports up to 256 distinct operating frequencies for the NoC, together with the independent definition of each processing element´s clock. Fourth, there is the propo-sition of a new local clock generator targeting minimum area, low power dissipation, oper-ating frequency stability and insensitivity to process, voltage and temperature variations. Finally, this work provides a simulation and code generation environment for silicon im-plementations of the HeMPS-GLP MPSoC. This environment emulates the local clock ge-nerators, based on the designed local clock generator.