SHF: Small: Collaborative Research: STEMS: STatistic Emerging Memory Simulator

新兴的存储技术，如磁阻随机存取存储器（MRAM）、相变存储器（PCRAM）和电阻随机存取存储器（RRAM），正在被探索作为未来计算系统的潜在替代方案。然而，传统的存储器设计方法不足以解决这些新兴技术的概率行为，这些概率行为是由工艺变化和物理机制（例如，热波动）的内在随机性引起的。本研究的目的是开发一种称为统计新兴存储器模拟器（stem）的设计方法，用于使用这些新兴存储器技术进行电路/架构设计。智力上的优点包括：(1)开发一种通用的统计表征形式，将新兴的存储单元设计规范与设计变量、工艺变化和环境波动联系起来；(2)推导出一种可感知变化的紧凑型存储单元模型，以满足单元和阵列级别的统计设计优化需求；(3)研究一种统计存储设计方法，以探索内存结构、实现成本、并针对各种系统要求设计规范。本研究将从根本上改变未来存储器技术的设计方法，开启存储器设计的创新方向，并在架构考虑下优化和平衡新兴存储器的新设计特征，激发设计哲学从确定性时代向概率时代的转变。所提出的技术为现有的概率系统和架构研究提供了一个补充的视角，同时强调了存储器设计的良率和概率特性。该项目将促进半导体行业进一步发展和更广泛地采用新兴存储技术。设计方法和内存建模的创新将对半导体存储芯片的设计和制造方式产生影响。参与这项研究的本科生和研究生将接受下一代半导体行业劳动力的培训。

Emerging memory technologies such as Magnetoresistive random-access memory (MRAM), Phase-change memory (PCRAM), and Resistive random-access memory (RRAM) are being explored as potential alternatives for future computing systems. However, traditional memory design methodologies are not sufficient to address probabilistic behaviors, which are caused by process variations and the intrinsic randomness in the physical mechanisms (e.g., thermal fluctuations) of these emerging technologies. The objective of this research is to develop a design methodology called STatistical Emerging Memory Simulator (STEMS) for circuit/architecture designs with such emerging memory technologies. The intellectual merits include the following: (1) developing a generic statistical characterization formalism to link the emerging memory cell design specifications with design variables, process variations and environmental fluctuations, (2) deriving a variation-aware compact memory cell model to fulfill the demands of the statistical design optimizations at cell and array levels, and (3) investigating a statistical memory design methodology to explore the tradeoffs among memory structure, implementation cost, and design specifications for various system requirements. The proposed research will fundamentally change the design methodologies for future memory technologies, initiate an innovative direction in memory designs, and optimize and balance the new design characteristics of emerging memories under architectural considerations, inspiring the transition of design philosophy from the deterministic era to the probabilistic era. The proposed techniques provide a complementary perspective to the existing probabilistic system and architectural research while emphasizing the yield and probabilistic properties of memory designs. This project will facilitate further advances and wider adoption of the emerging memory technologies by the semiconductor industry. Innovations in design methods and memory modeling will have an impact on the way in which semiconductor memory chips are designed and fabricated. Undergraduate and graduate students involved in this research will be trained for the next-generation semiconductor industry workforce.