MemDANCE: Memristor-based Dendritic Analog Computing Enhancement

该项目将在两个主要方向推进神经形态计算领域。首先，它将开发基于忆阻器的硬件，用于极其节能的尖峰递归神经元网络，包括树突计算，这是真实的神经元系统的主要组成部分，目前在人工神经元网络领域中大多被忽视。其次，包括树突计算将能够通过增加时间记忆来提高尖峰递归神经元网络的计算能力，这对于有效处理包括跨不同时间尺度的签名的复杂时间序列是必要的。两者结合起来，该项目有可能成为一个非常节能的基于忆阻器的尖峰递归神经元网络，可用于边缘设备和基于硬件的人工智能神经形态应用，并显着提高性能。

This project will advance the field of neuromorphic computing in two major directions. First, it will develop memristor-based hardware for extremely energy-efficient spiking recurrent neuronal networks including dendritic computing, a major component of real neuronal systems that is mostly ignored in the field of artificial neuronal networks by today. Secondly, the inclusion of dendritic computing will enable boosting the computational power of spiking recurrent neuronal networks by increasing temporal memory which is necessary for the effective processing of complex temporal sequences including signatures across different temporal scales. Taking both together, this project has the potential for an extremely energy-efficient memristor-based spiking recurrent neuronal network that may be used in edge devices and for hardware-based neuromorphic applications of AI with significantly improved performance.