Study on Resistive Switching Random Access Memory (RRAM) for synapse device and Neuromorphic system

Abstract

Recently, the current memory Silicon-based Flash memory technology are rapidly developed. However, Flash memory is faced with limitation such as low operation speed, poor endurance, and high write voltage. Moreover, continuation in increasing the density of flash will reach their physical limitation in the near future. As a result, the new memory device must be developed including high performance characteristics such as high operation speed, good reliability, and so on. To overcome the problems of current memory device, a various memory concepts such as magnetoresistive random access memory (MRAM), phase change random access memory (PRAM), ferroelectric random access memory (PRAM), resistive switching random access memory (RRAM). Among them, RRAM is considered as the promising candidates for next-generation non-volatile memory (NVM) due to their low power consumption, high speed operation, high density integration, low operation voltage, good reliability. Moreover, RRAM can be used as synapse device for the hardware implementation of human brain based neuromorphic system due to their gradual conductance modulation capability. A general neuromorphic computation system is based on neurons with processing signal and synapses with memory. Especially, there are several promising memory device having synaptic behavior. Among them, Pr0.7Ca0.3MnO3 (PCMO) based RRAM is most promising candidates for building neuromorphic system due to non-volatile, high on/off ratio, good uniformity, and gradual conductance modulation with same voltage stimulus. In this thesis, PCMO based RRAM was fabricated for realizing electrical synapse. TiN / PCMO shows interface switching due to oxidation-reduction between TiN and PCMO. TiN / PCMO has excellent resistive switching properties such as high on/off ratio, good uniformity, and controllable conductance modulation. We showed TiN / PCMO can be utilized for emulating synaptic plasticity such as Spike-Time-Dependent Plasticity (STDP), which is one of key mechanism for neuromorphic system. Furthermore, associativity, which is one general example of human behavior, is implemented with complementary metal-oxide-semiconductor (CMOS) device and TiN / PCMO. By adopting CMOS devices as neurons and TiN/PCMO devices as synapses, we implemented neuromorphic hardware that mimics associative memory characteristics in real time for the first time. Owing to their excellent scalability, resistive-switching devices, shows promise for future high-density neuromorphic applications.