Architectural Techniques to Sustain Moore's Law to Enable Reliable, Secure, and Efficient Memory Systems

摩尔定律在实现可扩展的高容量存储系统方面发挥了重要作用。不幸的是，维持摩尔定律是具有挑战性的，因为扩展存储单元的特征大小也会降低它们的可靠性，使它们变得不安全。这项研究的长期任务是研究一种聪明的方案，为跨不同计算系统的广泛应用提供可靠和安全的存储系统。我们的研究将显著提高能源效率，降低能源成本，并有助于构建高性能计算系统。这项工作将使加拿大人能够开发提供重大经济和环境收益的计算系统。在技术方面，启用安全可靠的大容量内存系统对于构建数据中心、智能手机、台式机和超级计算机至关重要。此外，这些系统有助于使加拿大成为云计算、机器学习和人工智能、商业和科学计算领域的领导者。为此，该计划旨在提供三个主要的研究主题。1. 用于容错的计算机硬件和体系结构：通过使用体系结构和系统级策略来提高内存可靠性，从而帮助维持摩尔定律。目前，内存系统将数据分布在多个内存模块中，并保留错误代码以纠正错误数据。我们的项目将超越这些方法，并研究跨层方案、数据冗余策略、新内存技术、单元可变性和页面映射方案，以进一步提高可靠性。2. 安全和隐私：在这个大数据、机器学习和人工智能的时代，内存系统越来越多地用于存储和提供敏感数据。随着存储系统可靠性的降低，安全存储数据变得更加具有挑战性。更糟糕的是，由于当前的体系结构和系统级解决方案往往具有显著的性能和功耗开销。我们的项目将着眼于以低成本实现高安全性的架构和系统级解决方案。3. 数据管理：随着内存容量的增加，其带宽也需要按比例增长。该计划将研究数据压缩策略，以无缝地增强带宽。当前系统中的数据压缩往往需要大量的软件支持，还会产生额外的硬件开销。我们的项目将在架构和系统级别探索低成本的数据压缩策略，用于主存、缓存和存储。培养高素质人才是加拿大的一项重要要求。该计划的关键目标之一是训练HQP在计算机体系结构、统计和概率论、数学分析、性能分析、操作系统、编译器和编程语言等领域发展重要技能。

Moore's law has been instrumental in enabling scalable high-capacity memory systems. Unfortunately, sustaining Moore's law is challenging, as scaling the feature sizes of memory cells also reduces their reliability and makes them insecure. The long-term mission of this research is to investigate clever schemes that enable reliable and secure memory systems for a wide range of applications across different computing systems. Our research will significantly enhance energy efficiency, reduce energy costs, and help build high-performance computing systems. This work will enable Canadians to develop computing systems that provide significant economic and environmental gains. On a technological front, enabling secure and reliable high-capacity memory systems is essential for building data centers, smartphones, desktops, and supercomputers. Furthermore, these systems are instrumental in enabling Canada to be a leader in cloud computing, machine learning and artificial intelligence, commerce, and scientific computing. To this end, this program aims to provide three main research themes. 1. Computer hardware and architecture for fault tolerance: To help sustain Moore's law by improving memory reliability by using architecture and system-wide strategies. Currently, memory systems distribute data across several memory dies and keep error-codes to correct faulty data. Our program will go beyond these approaches and investigate cross-layer schemes, data redundancy strategies, new memory technologies, cell variability, and page mapping schemes to improve reliability further. 2. Security and Privacy: In this era of big data, machine learning and artificial intelligence, memory systems are increasingly being used to store and provide sensitive data. As the reliability of memory systems reduces, it is becoming even more challenging to store data securely. To make matters worse, as current architecture and system-level solutions tend to have significant performance and power overheads. Our program will look at architecture and systems-level solutions that will enable high security at low-costs. 3. Data Management: As memory capacity increases, its bandwidth also needs to grow proportionally. This program will investigate data compression strategies to enhance bandwidth seamlessly. Data compression in current systems tends to require massive software support and also incurs additional hardware overheads. Our program will explore low-cost data compression strategies at architecture and system-level for main memory, caches, and storage. Development of highly qualified personnel (HQP) is a critical requirement for Canada. One of the critical goals of this program is to train HQP to develop vital skills in the areas of computer architecture, statistics and probability theory, mathematical analysis, performance analysis, operating systems, compilers, and programming languages.