基于硫系化合物的类神经元突触的认知存储器件研究

具有类神经元突触功能的信息存储器件，能够实现信息存储与处理的融合，被视为克服传统计算机架构中冯诺依曼瓶颈的一种有效选择。本项目将研究新型的基于硫系化合物的认知存储器件。拟利用硫系化合物的1)基于能量累积的非晶-多晶阈值相变特性,2)基于空间电荷限制电流模型的忆阻特性，3)基于金属离子导电通道的阻变特性，来模拟实现1)类神经元阈值激发特性，2)长时程增强/抑制及激发时间依赖的突触可塑性等学习记忆法则，3)记忆巩固和衰退等认知功能。并探索认知存储器件的认知功能的多样性,特性的可控性及网络集成的可行性。利用溅射、光刻、剥离等微纳工艺制备认知存储器件，搭建认知存储器表征测试平台，最终在单个无机纳米器件中实现数据存储功能的同时，实现类神经元突触认知功能，为未来大规模认知网络奠定器件基础。本项研究将对新型存储器件及认知器件的发展提供新的学术思路和技术路线。

Neuromorphic devices and systems have been built to emulate the synaptic dynamics and neural interactions in order to break the Von Neumann bottleneck, which refers to the severely constrained computation efficiencydue to the limited data transfer rate in bus between the CPU and the memory .Assuredly, the human brain that contains ~10E11 neurons and ~10E15 syanpses is the most powerful and maturest non-Von Neumann processor, as the data storage and processing are combined together rather than be devided to the separatememory and the CPU. Neurobiological cognitve process, in essence, is a macroscopic behavior based on the micro-dynamics of neurons and synapses. When the electronic signals transfer from pre-synaptic neuron to the post-synaptic neuron ,the synapse evaluates the incoming signal by contrasting with the previous state and the synaptic weight. Then a processed signal reaches thepost-synaptic neuronalong the dendrite. Meanwhile, the data computation has taken place during the signal transfer. The short-term memory changes the synaptic weight, and long-term memory is structurally encoded in the new induced spine. The synaptic interactions between different neurons creat a complicate net work association. The activities of the association ,such as long-term potentiation(LTP), long-term depression(LTD) and spiking-timing dependent plasticity(STDP) are regarded as basic rules for learning, perception, decision ,creation and other cognitive processes in cerebralcortex. The proposed research will investigate a)the amorphous-crystalline threshold phase change behavior based on energy accumulation effect, (b)the memristive characteristic based on space-charge-limited-currentmodel, (c)the resistive switching phenomena based on metal ion conductive filament, through experiments accompany with materials computation.These features are used to mimic a) neuronal threshold spiking , b)LTP、LTD、STDP and other learning rules based on synaptic plasticity, c)memory consolidation and memory decay.We intend to fabricate the brain-inspired cognitive memory devicewith neuronal/synaptic functionsusing magnetron sputtering, photolithography, lift-off and oter nano-fabrication processes.Electrical characterization system will be designed and built to measure the phase change, memristiveand resistive switching characteristics. Finally,brain-inspired cognitive funcitons will be implemented accompany with the data storage functionin one single inorganic nano-device. Results of this research will promote the mergence of material science, information technology, cognitive science and neurobiology, providinga fresh solution to get beyond the Von Neumann architecture and develop the neuromorphic computing.