三维集成电路热可靠性快速分析与在线优化技术研究

3-dimensional (3D) stacked integration is the trend of the future microprocessor designs. However, thermal management and related reliability effects have become major limiting factors of design and application of the many-core and 3D integrated circuits (3D ICs). 3D integration allows more active layers integrated on top of each other, which significantly increases the power density and leads to large thermal resistances. The temperature distributions in many-core processors of 3D ICs are highly dependent on the loads and software programs. Effective optimization technique for thermal induced reliability is a key enabler for emerging 3D high-performance microprocessors. In this project, we propose to develop new fast analysis and runtime optimization technique for thermal induced reliability of 3D ICs. The proposed technique, compared with traditional ones, has many innovations and advantages: it includes a fast and accurate reliability analysis method, a highly scalable and efficient distributed optimization method, an overshoot proof predictive control technique, and an optimization method considering a number of thermal reliability issues. The proposed technique aims to solve the key problem in 3D IC research, and is able to improve the reliability as well as the runtime efficiency of 3D ICs. It will also have significant impact on the advance of 3D IC technology.

三维层叠式（3D）集成技术是目前及未来的微处理器的发展趋势。然而，热可靠性问题已成为三维集成电路（3D IC）设计与应用的主要限制因素。三维层叠式集成允许更多的有源器件层叠在一起，这会显著提高芯片功率密度，并导致芯片散热效能降低。同时，其多核处理器的温度分布又高度依赖于软件程序负载。有效的热可靠性分析与优化技术是实现和推动三维高性能微处理器发展的核心技术。本项目中拟研发针对三维集成电路的新型热可靠性快速分析与在线优化技术。新技术相对现有技术具有多方面优势与创新：新技术含有快速而又精确的热可靠性分析方法，高度可扩展而又高效的分布式优化方法，防止过冲效应的预测控制技术，综合考虑多种热失效机制的优化方法等多种先进方法与技术。新技术将解决三维集成电路研究中的核心问题，能够大幅改善三维集成电路的可靠性与提高其运行效率，推动三维集成电路技术的发展。

新型3D IC结构高功率密度所导致的散热困难问题极大制约了其性能的发挥。在本项目中，我们对三维集成电路(3D IC)的热可靠性快速分析与在线优化技术展开了研究。在项目执行期间，项目组对硅穿孔热应力问题对 3D IC 热可靠性的影响、静态功耗估计中的温度与静态功耗之间的非线性关系问题、应用于众核芯片的动态热管理技术、具有暗硅现象芯片的动态热管理技术，以及考虑多种失效机制的 3D IC 可靠性在线优化方法等方面问题进行了研究。项目组成功研发了基于神经网络并考虑 3D IC 中硅穿孔结构的应力与温度耦合紧凑建模与高速分析技术、基于层次结构的众核芯片动态热管理技术、综合利用多种热可靠性优化技术的混合式热可靠性在线优化技术、综合考虑多种 3D IC 常见失效效应的热失效优化技术等研究成果。本项目的研究工作能够在保证可靠性的同时有效提高3D IC的运算性能。