Research On Phase Change Memory Based On GeGaSb And Its Neuromorphic Application

The storage and processing of mass data have always been a major issue in the age of information.In this context,various emerging data storage technologies and novel computing architectures have been proposed and developed.Among them,the phase change memory(PCM)not only has excellent storage performance,but also exhibits the potential to imitate the behavior of biological synapses and neurons due to its threshold switching and progressive process of crystallization/amorphization.Therefore,PCM is expected to realize brain-inspired computing as the neuromorphic device.However,the research on neuromorphic applications of PCM is still at the early stage,and there are several problems such as single function and poor performance of the device.In this paper,a GeGaSb-PCM is proposed,while its performance,phase change mechanism and neuromorphic applications are systematically studied.The main works in this paper are as follows:Firstly,the fabrication and performance measurements of the GeGaSb film and GeGaSbPCM are introduced.The results of in-situ annealing resistance tests exhibit that GeGaSb has high crystallization temperature and high 10-years data retention temperature,showing great thermal stability.And the results of electrical tests show that GeGaSb exhibits a unique abruptto-progressive SET behavior by applying electrical pulses with small width,which has the advantages of fast phase transition speed,large resistance contrast,low resistance drift coefficient,great number of intermediate resistance states and excellent linearity.Secondly,the mechanism analysis of the abrupt-to-progressive SET behavior of GeGaSb is studied.The lattice structure,crystallization degree and elemental distribution of GeGaSb in different resistance states are analyzed by X-ray diffraction,transmission electron microscopy and X-ray energy spectroscopy,and it is concluded that the generation of the crystallization channel in the phase change material leads to the abrupt change in resistance,while the combined effect of increasing crystallization degree of the channel and Sb aggregation due to phase separation leads to the subsequent progressive change.At last,the synaptic and neuronal functions of GeGaSb-PCM-based neuromorphic devices are realized by simulation.A simple simulation of convolutional neural network based on the GeGaSb-PCM is achieved,and the synaptic performance of devices is evaluated by carrying out a standard handwritten digit recognition task.And the simulation results show that the neural network achieves a high recognition accuracy up to 96.0% and a small cross entropy loss as low as 0.23,which mean that the synaptic performance of GeGaSb-PCM is good.In addition,a neuron circuit based on GeGaSb-PCM is designed and the integrate-andfire behavior is also realized by simulations.The simultaneous implementation of synaptic and neuronal functions in a single device provides the possibility of building full phase-change neural networks in the future.