项目名称: 针对暗硅片的多核处理器体系结构研究和优化平台搭建
项目编号: No.61202026
项目类型: 青年科学基金项目
立项/批准年度: 2013
项目学科: 计算机科学学科
项目作者: 梁晓峣
作者单位: 上海交通大学
项目金额: 25万元
中文摘要: 半导体工艺的深入发展可以将更多的晶体管集成在芯片上。但是现代多核处理器芯片受到功耗墙,良品率和程序并行度等因素的制约,难以继续靠增加核的数目来提高性能。这些问题都预示着未来多核处理器设计会遇到瓶颈:我们可以在芯片上集成更多的晶体管,但是却无法有效的利用它们。这些没有办法被充分利用的芯片面积被称之为"暗硅片"。本课题重点研究未来处理器发展中的暗硅片问题,讨论如何充分利用这些"冗余"的芯片面积,针对不同类型的应用设计新型处理器体系结构以达到性能和功耗的优化。 研究内容为1)暗硅片的发展趋势建模2)适用于暗硅片的芯片部件选择和参数提取,包括集成电压调制器,可编程门阵列,特殊接口,各类特殊电路等3)不同类别处理器的性能功耗,适用领域及主要应用的研究和分析4)建立一个面向暗硅片体系结构设计的智能优化平台5)针对不同类型的处理器应用,在暗硅片中集成特定的芯片部件优化体系结构。
中文关键词: 暗硅片;碳纳米管;通用图形处理器;;
英文摘要: The continued scaling of semiconductor industry promises more and more transistors to be integrated onto a single chip. But modern multicore processors cannot keep improving the performance simply by adding the number of cores due to the limitation of power wall, yield and program parallism. All these issues predict that we might be able to integrate many more transistors than those we are capable of using. These under-utilized transistors and silicon area are called "Dark Silicon." This proposal is to study the problem of dark silicon in the future processor design. We will discuss how to leverage the "redundant" dark silicon to design specical silicon components and optimize architecture for different types of microprocessors. The research mainly focus on 1) modeling the future trend of dark silicon 2) selection and parameter extraction of silicon components suitable for dark silicon. Examples are: voltage regulators, FPGA, special IOs and circuits, etc. 3) study the power and performance trade-offs of different microprocessors, their usage pattern and target applications 4)build a smart platform for the simulation and optimization of computer architecture in the dark silicon era 5) apply different silicon components to different microprocessors and optimize the processor architectures.
英文关键词: Dark Silicon;CNFET;GPGPU;;