SHF: Small: MultiSpot - Closing the power-delivery/heat-removal cycle for heterogeneous multiscale systems

SHF：小型：MultiSpot - 关闭异构多尺度系统的供电/排热循环

• 批准号: 1619127
• 负责人: Mircea Stan
• 金额: $ 45万
• 依托单位: University of Virginia Main Campus
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2016
• 资助国家: 美国
• 起止时间: 2016-06-15 至 2020-05-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1619127&HistoricalAwards=false
• 关键词: SHF; Small; MultiSpot; Closing; power

The semiconductor industry is poised to continue the historic Moore's law trend of doubling the level of integration every 18 months, even as the virtuous cycle benefits of Dennard scaling are quickly vanishing. Once devices no longer scale laterally, one way to continue to increase areal density is to use 3D integration. The main objective of the current project, called the MultiSpot project, is to provide fundamental solutions to the problem of 3D-IC power wall. These solutions could have a deep impact on the semiconductor industry, and thus the society as a whole by providing a path forward for Moore's law to take advantage of the third dimension. The project will also lead to new research infrastructure exemplified by the several high-impact architecture modeling tools for power and thermals previously developed by the PIs, such as HotSpot, HotLeakage VoltSpot, and ArchFP that have enabled extensive research at academic institutions and industry in the past. One approach is to provide very low impedance paths from the interior of the volume to the outside for power in addition to heat removal. A different dual use for microchannels is to modify their walls to create supercapacitors that can be useful to store extra charge to combat power brownouts or boost operations. An alternative method to address the power delivery volumetric/areal mismatch is voltage stacking, in which the different layers in a 3D-IC stack are electrically connected in series, instead of in parallel as in a conventional approach. MultiSpot proposes practical methods to complement and assist the implicit regulation of voltage stacking with the explicit regulation of multi-output switched-capacitor ladder networks. The project also considers cross-layer aspects, such as details of embedded capacitor technologies, the architectural implications of homogeneous vs. heterogeneous 3D-IC solutions, the modeling of power delivery and heat removal at the architecture level, and validation through modeling and simulations.

半导体行业正准备继续历史性的摩尔定律趋势，即集成度每18个月翻一番，即使Dennard缩放的良性循环效益正在迅速消失。一旦器件不再横向扩展，继续增加面密度的一种方法是使用3D集成。目前的项目称为MultiSpot项目，其主要目标是为3D-IC电源墙问题提供根本解决方案。这些解决方案可能会对半导体行业产生深远的影响，从而通过为摩尔定律提供一条利用第三维的前进道路，对整个社会产生深远的影响。该项目还将带来新的研究基础设施，例如PI以前开发的几种高影响力的电力和热架构建模工具，如HotSpot，HotLeakage VoltSpot和ArchFP，这些工具过去已经在学术机构和行业进行了广泛的研究。一种方法是提供从体积的内部到外部的非常低的阻抗路径，以用于除热之外的功率。微通道的另一个双重用途是修改它们的壁，以创建超级电容器，这些超级电容器可以用于存储额外的电荷，以对抗电力管制或提升操作。解决功率递送体积/面积失配的替代方法是电压堆叠，其中3D-IC堆叠中的不同层串联电连接，而不是如常规方法中的并联电连接。MultiSpot提出了实用的方法来补充和辅助多输出开关电容梯形网络的显式调节电压叠加的隐式调节。该项目还考虑了跨层方面，例如嵌入式电容器技术的细节，同构与异构3D-IC解决方案的架构影响，架构级别的功率传输和散热建模，以及通过建模和仿真进行验证。