ECCS - EPSRC Development of uniform, low power, high density resistive memory by vertical interface and defect design

未来的节能计算将逐渐从以计算为中心转向以内存为中心。存储和移动数据需要巨大的能源成本，这比计算数据要高得多。实际上，在数据密集型应用程序中，内存支配着计算能量（bb0 4倍）。新型超低功耗非易失性存储器（NVM）是计算的各个方面的核心，从用于数据中心的独立存储级存储器（SCM），到用于物联网、汽车行业等的嵌入式非易失性存储器（e-NVM），再到新的计算形式。随着相关能源消耗的增加，该领域正在迅速增长，例如，到2030年，数据中心将使用全球10%的电力，市场每年增长10%，到2023年将达到1500亿美元。由于摩尔定律的终结和新的处理器技术尚未建立，这种效率无法从处理中获得。最大的性能和节能收益将在内存方面。用高性能内存NVM替换标准内存可以减少60%以上的功耗。在众多正在深入研究的候选NVM形式中，氧化电阻RAM （RRAM）在成本、密度、简单性和3D集成潜力方面具有最大的潜力。然而，目前存在一些挑战，特别是对成型电压、均匀性差、结垢和耐用性的需求。这个项目的目的就是克服这些挑战。这将通过将我们的突破性成果从理想系统应用到行业平台来实现。到目前为止，在理想的系统中，我们已经证明了a)精确和非随机的导电通道可以被设计成薄膜，以消除对高压形成过程的需要；B)高且可控的氧空位浓度导致高且可重复的开关比；C)消除过渡金属的使用产生低泄漏，并大大减少了膜与膜之间的变化。本项目将探索的产业平台是掺杂HfO2，通过溅射和原子层沉积生长。一个在提案领域拥有超过15年非常密切合作的国际领先团队将承担这项工作。首先，将在普渡大学使用脉冲激光沉积（PLD）来生长VAN薄膜。PLD是一种最简单的方法，可以快速制作出最完美的金属氧化物薄膜。这将使我们能够更全面地了解RS过程，并将提供有关如何通过工业可扩展过程生长薄膜的信息。普渡大学和剑桥大学都将参与可编程器件HfO2纳米结构薄膜的设计。从PLD生长中获得的知识将被转化为剑桥大学开展的溅射和ALD方法。布法罗大学的工作将集中在使用普渡大学和剑桥大学沉积的RS薄膜制造和测试原型忆阻器和rram。最先进的表征工具也将成为材料理解和设备优化的核心，这些工具将在普渡大学和剑桥大学使用。小组之间将有很强的互动，定期进行抽样和知识转移。我们的最终目标是一个无成形器件，其通/关比~104，耐用性bbb1012，每个开关<10pJ，均匀性几%，缩放到20nm。在培训方面，我们将培养材料科学和电子工程方面的毕业生。我们将在美国和英国的世界领先的研究环境中培养3名以上的早期职业研究人员，涉及几家公司（大小不等），包括剑桥公司ARM，他们在英国和美国的内存领域都非常活跃。

The future of energy-efficient computing will increasingly move from being compute-centric to being memory-centric. There is a huge-energy cost of storing and moving data, which is much higher than computing it. Indeed, memory dominates compute energy (>4X) in data-intensive applications. New ultralow power non-volatile memory (NVM) is central to all aspects of computing, from stand-alone, storage class memory (SCM) for use in data centres, to embedded non-volatile memory (e-NVM) for IoT, the automotive industry, etc, to new forms of computing. The area is growing hugely, along with the associated energy consumption, e.g. data centers will use >10% of global electricity by 2030 with a market growing ~10% per year, to >$150 billion by 2023. Such efficiency cannot come from processing as Moore's law reaches an end and new processor technologies are not yet established. The biggest performance and energy saving gains will be in memory. The replacement of standard memory with high performance memory NVM could reduce power usage by more than 60%.Of the many candidate NVM forms under intense investigation, oxide resistive RAM (RRAM) has the greatest potential in terms of cost, density, simplicity and potential for 3D integration. However, several challenges currently exist, notably the need for a forming voltage, poor uniformity, scaling, and endurance. The aim of this project is to overcome these challenges. It will be done by adopting our ground-breaking results from an ideal system to an industry platform. So far, in the ideal system, we have demonstrated that a) precise and non-random conducting channels can be engineered into films to eliminate the need for a high voltage forming process; b) high and controlled oxygen vacancy concentrations lead to highly and reproducible on-off ratios; c) eliminating use of transition metals produces low leakage and strongly reduces variabilities from film to film. The industry platform we will explore in this project is doped HfO2, grown by sputtering and atomic layer deposition. An internationally leading team with more than 15 years of very strong collaboration in the area of the proposal will undertake the work. First, growth of VAN films using pulsed laser deposition (PLD) will be undertaken at Purdue University. PLD is the simplest way to make the most perfect metal oxide films in a rapid way. These will enable us to understand the RS processes more fully and will provide information on how to grow the films by the industrially scaleable processes. Both Purdue and Cambridge will be involved in the HfO2 nanostructure film design for PLD. The knowledge from the PLD growth will then be translated to the sputtering and ALD approaches undertaken at the University of Cambridge. The effort at the University at Buffalo will focus on the fabrication and testing of prototype memristors and RRAMs using the RS films deposited by Purdue University and University of Cambridge. State-of-the art (some in-operando) characterisation tools will also be central to materials understanding and device optimisation and these will be used at Purdue and Cambridge. A very strong interaction between the groups, with regular sample and knowledge transfer will take place. Our ultimate goal is a forming free device, with on/off ratio~104, endurance >1012, <10pJ per switch, uniformity of few %, scaled to 20 nm.In terms of training, we will educate graduates in materials sciences and electronic engineering. We will train more than 3 early career researchers in world-leading research environments in the US and UK, with several companies involved (small to large), including the Cambridge company ARM who are very active, both in the UK and US, in the memory area.