Enhanced resistance switching in ultrathin Ag/SrTiO3/(La,Sr)MnO3 memristors and their long-term plasticity for neuromorphic computing

Neuromorphic computing is a promising candidate for next-generation information technologies. In the present work, we report the realization of long-term plasticity and synapse emulations in Ag/SrTiO3/(La,Sr)MnO3 memristors with the SrTiO3 active layers down to 3 unit cells (u.c.) in thickness. In the 3 u.c.-thick SrTiO3 device, efficient control of Ag+-ion migration gives rise to enhanced memristive properties with the conductance continuously modulated within a large memory window of similar to 26 000% between an Ohmic low resistance state (LRS) and an electron-tunneling high resistance state (HRS). In addition, long-term plasticity of the Ag/SrTiO3/(La,Sr)MnO3 memristors is found to be dependent upon the resistance state. In the HRS, the devices exhibit excellent spike-timing-dependent plasticity characteristics with a large modulation of synaptic weight of similar to 3500% and sensitive response to electrical stimuli of as low as similar to 1.0 V and as fast as similar to 0.01 ms. Adopting the spike-timing-dependent plasticity results as database, supervised learning simulations are demonstrated in the Ag/SrTiO3/(La,Sr)MnO3-based neural networks and a high accuracy rate of 95.5% is achieved for recognizing handwritten digits. These results provide more insights into the ionic migration at nanoscale for continuous resistance modulation and facilitate the design of ultrathin memristors for high-density 3D stacking artificial neural networks.

Automated summary

This study demonstrates the realization of long-term plasticity and synapse emulations in Ag/SrTiO3/(La,Sr)MnO3 memristors with thin SrTiO3 active layers, allowing for efficient control of Ag+ ion migration and continuous conductance modulation. In the high resistance state, the devices exhibit excellent spike-timing-dependent plasticity characteristics and show sensitive responses to electrical stimuli at low voltages and fast speeds. Additionally, supervised learning simulations using spike-timing-dependent plasticity results in Ag/SrTiO3/(La,Sr)MnO3-based neural networks achieved a high accuracy rate of 95.5% for recognizing handwritten digits.