Ismo Hänninen, Wayne Buckhanan, M. Niemier, G. Bernstein
{"title":"Network on metachip architectures","authors":"Ismo Hänninen, Wayne Buckhanan, M. Niemier, G. Bernstein","doi":"10.1145/2401716.2401719","DOIUrl":null,"url":null,"abstract":"The size of systems on a chip is limited by our ability to design and fabricate such systems, staying within the appropriate costs depending on the application. In this paper, we propose a divide-and-conquer approach, Quilt Packaging®, to be utilized for reducing the fabrication costs of large digital systems by partitioning them into a quilted \"metachip\" that offers integration density and performance merits surpassing the traditional system-on-chip. The physical partitioning and the network-on-a-quilt are closely linked, and should be designed concurrently. For this purpose, we present calculations on the silicon cost of the interconnects and partitioning, discuss the network granularity, and propose a multiprocessor design around a quilted modular network, offering novel techniques to improve the performance and enable true heterogeneous integration. Specifically, the silicon costs of the quilting method are demonstrated to be around 1% of the chip area, while the yield benefits can be in the tens of percents regime. The metachip concept enables the combination of standard high-density memory technologies and wide-bus access with improved performance, typically at least doubling the amount of memory vs. single-chip CMOS. Our modular quilted network enables the integration of non-CMOS chips into the quilt.","PeriodicalId":344147,"journal":{"name":"Network on Chip Architectures","volume":"22 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2012-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"1","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Network on Chip Architectures","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1145/2401716.2401719","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 1
Abstract
The size of systems on a chip is limited by our ability to design and fabricate such systems, staying within the appropriate costs depending on the application. In this paper, we propose a divide-and-conquer approach, Quilt Packaging®, to be utilized for reducing the fabrication costs of large digital systems by partitioning them into a quilted "metachip" that offers integration density and performance merits surpassing the traditional system-on-chip. The physical partitioning and the network-on-a-quilt are closely linked, and should be designed concurrently. For this purpose, we present calculations on the silicon cost of the interconnects and partitioning, discuss the network granularity, and propose a multiprocessor design around a quilted modular network, offering novel techniques to improve the performance and enable true heterogeneous integration. Specifically, the silicon costs of the quilting method are demonstrated to be around 1% of the chip area, while the yield benefits can be in the tens of percents regime. The metachip concept enables the combination of standard high-density memory technologies and wide-bus access with improved performance, typically at least doubling the amount of memory vs. single-chip CMOS. Our modular quilted network enables the integration of non-CMOS chips into the quilt.