Superior artificial synaptic properties applicable to neuromorphic computing system in HfOx-based resistive memory with high recognition rates

Abstract

With the development of artificial intelligence and the importance of big data processing, research is actively underway to break away from data bottlenecks and modern Von Neumann architecture computing structures that consume considerable energy. Among these, hardware technology for neuromorphic computing is in the spotlight as a next-generation intelligent hardware system because it can efficiently process large amounts of data with low power consumption by simulating the brain’s calculation algorithm. In addition to memory devices with existing commercial structures, various next-generation memory devices, including memristors, have been studied to implement neuromorphic computing. In this study, we evaluated the synaptic characteristics of a resistive random access memory (ReRAM) with a Ru/HfO_x/TiN structure. Under a series of presynaptic spikes, the device successfully exhibited remarkable long-term plasticity and excellent nonlinearity properties. This synaptic device has a high operating speed (20 ns, 50 ns), long data retention time (> 2 h @85 ℃) and high recognition rate (94.7%). Therefore, we propose that memory and learning capabilities can be used as promising HfO_x-based memristors in next-generation artificial neuromorphic computing systems.