Synaptic plasticity has been traditionally credited for learning in the brain. The prevalent view on learning through synapses forms the backbone behind all the significant developments in the area of artificial neural networks (ANN). However, more recent studies in Neuroscience reveal that dendritic junctions play a crucial role in the dynamics of learning, leading to increased efficiency and faster learning. Consequently, there is a need to implement dendritic computation/learning at a hardware level in ANNs. Resistive Random Access Memory (RRAM) devices have been frequently used as non-volatile synaptic elements for in-memory-computing (IMC) or neuromorphic computing applications. However, their usage in implementing dendritic dynamics has been rarely investigated. This work reports a SiOx-based RRAM device with gradual and volatile resistance switching behavior. We demonstrate that the switching dynamics of this device can be used to emulate the behavior of a dendritic junction. We also report robust endurance of this device up to 1 million cycles and investigate the transport mechanism responsible for the switching dynamics. The demonstrated dendritic devices makes room for the monolithic integration of all-SiOx neural networks in the future.

中文翻译：

---

基于 PECVD SiO x 的 RRAM 树枝状器件的演示

传统上，突触可塑性被认为是大脑学习的元凶。通过突触学习的普遍观点构成了人工神经网络（ANN）领域所有重大发展背后的支柱。然而，神经科学领域最近的研究表明，树突连接在学习动态中发挥着至关重要的作用，可以提高效率和加快学习速度。因此，需要在人工神经网络的硬件级别上实现树突计算/学习。电阻随机存取存储器 (RRAM) 器件经常用作内存计算 (IMC) 或神经形态计算应用的非易失性突触元件。然而，它们在实现树突动力学方面的用途却很少被研究。这项工作报告了一种基于 SiOx 的 RRAM 器件，具有渐进且易失的电阻切换行为。我们证明该器件的开关动力学可用于模拟树突结的行为。我们还报告了该器件高达 100 万次循环的强大耐用性，并研究了负责切换动态的传输机制。所展示的树突状器件为未来全SiOx神经网络的单片集成腾出了空间。