SkyANN: Skyrmionic Artificial Neural Networks

SkyANN：Skyrmionic 人工神经网络

• 批准号: 10108371
• 金额: $ 133.95万
• 依托单位: THE UNIVERSITY OF MANCHESTER
• 依托单位国家: 英国
• 项目类别: EU-Funded
• 财政年份: 2024
• 资助国家: 英国
• 起止时间: 2024 至 无数据
• 项目状态: 未结题
• 来源: https://gtr.ukri.org/projects?ref=10108371
• 关键词: SkyANN; Skyrmionic; Artificial; Neural; Networks

The Skyrmionic Artificial Neural Network (SkyANN) presents a groundbreaking paradigm for neuromorphic computing, closely emulating brain neurophysiology by combining skyrmionic quasiparticles, which mimic neurotransmitters and facilitate complex computations at the synapse level, with electrical CMOS connections that simulate the propagation of action potentials among neurons for rapid and dense inter-layer connectivity. Our innovative magneto-electric devices aim to achieve energy consumption four orders of magnitude lower than CMOS technology and double the bandwidth for the same device footprint, enhancing edge inference and learning capabilities. This approach challenges contemporary neural networks implemented with CMOS digital, mixed-signal, and emerging in memory computing technologies, which are limited by lower energy efficiency and reliability. Building on preliminary results from SkyANN partners, we plan an ambitious endeavor to develop a first-of-its-kind magneto-electric neural network, showcasing the promising potential of this novel technology. Along the way, we will refine materials, processes, design methodologies, and architectures to prepare the European micro- and nano-electronics ecosystem for the future, while supporting the EU's Green Deal vision. Our well balanced consortium brings together complementary expertise and extensive knowledge, spanning from device physics to circuits and architectures across multiple layers of design abstraction. As a result, the SkyANN consortium is poised to facilitate the rapid transfer of fundamental discoveries to relevant industrial stakeholders, accelerating impact and reinforcing European strengths in the economically, geopolitically, and socially vital semiconductor sector.

Skyrmionic人工神经网络(SkyANN)为神经形态计算提供了一种突破性的范例，通过将Skyrmionic准粒子(模拟神经递质并促进突触水平的复杂计算)与电连接(模拟神经元之间的动作电位传播)相结合，紧密模拟脑神经生理学，实现快速而密集的层间连接。我们的创新磁电设备旨在实现比CMOS技术低四个数量级的能耗，并将相同设备占地面积的带宽增加一倍，增强边缘推理和学习能力。这种方法挑战了当代用CMOS数字、混合信号实现的神经网络，以及新兴的存储计算技术，这些技术受到较低的能量效率和可靠性的限制。在SkyANN合作伙伴初步成果的基础上，我们计划进行一项雄心勃勃的努力，开发首个此类磁电神经网络，展示这项新技术的前景潜力。在此过程中，我们将完善材料、工艺、设计方法和架构，为未来的欧洲微纳电子生态系统做好准备，同时支持欧盟的绿色协议愿景。我们平衡良好的联盟将互补的专业知识和广泛的知识结合在一起，从设备物理到电路和架构，跨越多个设计抽象层。因此，SkyANN财团准备促进将基础发现迅速转移到相关行业利益相关者，加快影响并加强欧洲在经济、地缘政治和社会关键半导体领域的实力。