Exploration of Neuromorphic Memristive Devices

近年来，人们对新兴的存储技术产生了极大的兴趣，其中阻性随机存取存储器(RRAM)是最有前途的候选技术之一，它具有低功耗、高速切换和3D阵列中的高密度存储。双端电阻开关元件可以在适当的偏压下改变电阻。即使在没有电压偏置的情况下也能保持最后的电阻状态的能力暗示了包括非易失性存储器在内的一系列潜在应用。此外，值得注意的是，这些设备类似于不同的神经元功能--最重要的是，通过逐渐改变它们的阻力(“突触重量”)，具有突触般的可塑性。通过利用“内存中计算”，有可能解决“冯·诺伊曼瓶颈”这一长期存在的问题：需要在处理核心和内存之间不断地转移数据。这样的系统在提供在实现机器学习(ML)和机器智能方面远远优于现有CMOS硬件的解决方案方面具有巨大的潜力。其主要优点之一是显著降低了电路复杂性，并极大地提高了电源效率。在这个项目中，学生将通过使用和优化RRAM设备构建神经形态功能单元来广泛探索突触调制和神经元激活(人工神经网络中的两个基本功能)。突触功能可能包括增强、抑制和学习，如尖峰时间依赖可塑性(STDP)和“神经元样”功能将包括整合和尖峰。学生将有机会通过实验确定基于氧化物的RRAM器件的特征，为器件的优化和制造做出贡献，并构建将在功能和能效之间找到最佳平衡的物理和电路模型。

Recent years have seen great interest in emerging memory technologies, with resistive random access memory (RRAM) being one of the most promising candidates, with low power consumption, high-speed switching, and high-density storage in 3D arrays demonstrated. Two-terminal resistance switching elements can change resistance under appropriate voltage bias. The ability to retain the last resistance state even without a voltage bias suggests a range of potential applications including non-volatile memories. Furthermore, remarkably, these devices resemble different neuronal functions - most importantly synaptic-like plasticity by gradually changing their resistance ("synaptic weights"). By utilising the "computing-in-memory", it is possible to solve the long-lasting problem of the "von Neumann bottleneck": the need to continually shuffle data between processing cores and memory. Such systems have massive potential in delivering solutions that are far superior to existing CMOS hardware in the implementation of machine learning (ML) and machine intelligence. One of the main benefits is a significant reduction in circuit complexity and vast improvements in power efficiency. In this project, students will extensively explore the synaptic modulation and neuronal activation (two fundamental functions in artificial NNs) by construct neuromorphic functional units using and optimising RRAM devices. Synaptic functionality might include potentiation, depression, and learning such as spiking-time dependent-plasticity (STDP) and "neuronal-like" functionality will include integration and spiking. Students will have an opportunity to characterise oxide- based RRAM devices experimentally, contribute to optimisation and fabrication of the devices, and to construct physical and circuit models that will find the optimal balance between functionality and power efficiency.