神经形态器件及其类脑计算应用

Neuromorphic Devices for Brain-like Computing

在传统冯·诺依曼计算机中,中央处理器和存储器分离,数据吞吐量受到很大限制。因此,传统计算在处理非结构化数据时,其能量效率很难进一步提高。同时,传统CMOS器件的尺寸已接近物理极限,很难延续摩尔定律,这进一步限制了计算机的性能提升。为了在“大数据”时代提高计算系统的性能,必须改变计算机的计算范式。人类的大脑是一台高效、高容错和超低功耗的生物超级计算机。神经形态工程从生物大脑和感觉神经系统中汲取灵感,有望显著降低模式识别和决策判断等智能感知和计算任务的能耗和设备成本。

In the traditional von Neumann computer,the data throughput is greatly limited due to the separation of the CPU and memory.Therefore,it is difficult to further improve the energy efficiency of traditional computer when dealing with unstructured data.At the same time,the size of traditional CMOS devices is approaching the physical limit,and it is difficult to continue Moore's law,which also limits the performance improvement of computers.To improve the performance of computing systems in the so-called"big data"era,the computing paradigm of computers must be changed.Our brain is a biological supercomputer with high efficiency,high fault tolerance and ultra-low power consumption.Neuromorphic engineering draws inspiration from biological brain and sensory nervous system,and is expected to significantly reduce the energy consumption and device cost of intelligent perception and computing tasks such as pattern recognition and decision judgment.