Beyond neuromorphic: Exploiting the extended frequency response of memristive devices and systems to process information in new ways.

超越神经形态：利用忆阻设备和系统的扩展频率响应以新的方式处理信息。

• 批准号: EP/X017001/1
• 负责人: Anthony Kenyon
• 金额: $ 25.75万
• 依托单位: University College London
• 依托单位国家: 英国
• 项目类别: Research Grant
• 财政年份: 2023
• 资助国家: 英国
• 起止时间: 2023 至 无数据
• 项目状态: 未结题
• 来源: https://gtr.ukri.org/projects?ref=EP%2FX017001%2F1
• 关键词: Beyond; neuromorphic; Exploiting; extended; frequency

We propose to establish a wholly new direction in computing hardware. We will develop devices and circuits that initially take inspiration from the way the brain processes information (so-called neuromorphic computing), but will go beyond this to develop a radical new approach that uses the dynamics of the frequency response of neuromorphic devices known as memristors to process information in novel and powerful ways. Such devices are currently used as variable electrical resistors whose resistance depends on their past history. In this work we will exploit their rich dynamics by using their complex frequency response to process signals in the frequency domain and to modify the behaviour of other coupled devices in simple circuits.This work promises to open up a new direction for memristor research and address a pressing issue for modern computing systems: their increasingly unsustainable energy demands. Given that a single state-of-the-art machine learning system can generate as much CO2 during training as five cars emit over their lifetime, and that global data centres currently consume around 250TWh per year, new low-power computing approaches are needed urgently. Neuromorphic systems take inspiration from biology to close the six order of magnitude power consumption gap between the brain and digital computing systems. In this work we will go further and add capabilities not found in biology: processing using modification of the complex frequency response of systems. This project will establish a key toolbox of novel devices to underpin next generation neuromorphic and "neuromorphic plus" computing systems.

我们建议在计算硬件方面建立一个全新的方向。我们将开发最初从大脑处理信息（所谓的神经形态计算）的方式中吸收灵感的设备和电路，但将超越此功能，以开发一种激进的新方法，该方法使用了神经形态设备的频率响应的动力学，被称为Memristors的频率响应来处理新颖和强大的方法。此类设备目前被用作可变电阻器，其电阻取决于其过去的历史。在这项工作中，我们将通过对频域中的过程信号进行复杂的频率响应来利用它们的丰富动态，并在简单电路中修改其他耦合设备的行为。这项工作有望为Memristor研究打开新的方向，并解决现代计算系统的紧迫问题：它们日益不受限制的能源需求。鉴于单个最先进的机器学习系统可以在训练期间产生与五辆汽车在其一生中发射的二氧化碳一样多，而且目前每年每年消耗250TWH的全球数据中心，因此紧急需要采用新的低功率计算方法。神经形态系统从生物学中汲取灵感，以缩小大脑和数字计算系统之间的六个数量级功耗差距。在这项工作中，我们将走得更远，并添加生物学中找不到的功能：使用系统的复杂频率响应的修改处理。该项目将建立一个新型设备的关键工具箱，以支持下一代神经形态和“神经形态加上”计算系统。