Artificial Synapse Consisted of TiSbTe/SiCx:H Memristor with Ultra-high Uniformity for Neuromorphic Computing

Abstract

To enable a-SiCx:H-based memristors to be integrated into brain-inspired chips, and to efficiently deal with the massive and diverse data, high switching uniformity of the a-SiC0.11:H memristor is urgently needed. In this study, we introduced a TiSbTe layer into an a-SiC0.11:H memristor, and successfully observed the ultra-high uniformity of the TiSbTe/a-SiC0.11:H memristor device. Compared with the a-SiC0.11:H memristor, the cycle-to-cycle coefficient of variation in the high resistance state and the low resistance state of TiSbTe/a-SiC0.11:H memristors was reduced by 92.5% and 66.4%, respectively. Moreover, the device-to-device coefficient of variation in the high resistance state and the low resistance state of TiSbTe/a-SiC0.11:H memristors decreased by 93.6% and 86.3%, respectively. A high-resolution transmission electron microscope revealed that a permanent TiSbTe nanocrystalline conductive nanofilament was formed in the TiSbTe layer during the DC sweeping process. The localized electric field of the TiSbTe nanocrystalline was beneficial for confining the position of the conductive filaments in the a-SiC0.11:H film, which contributed to improving the uniformity of the device. The temperature-dependent I-V characteristic further confirmed that the bridge and rupture of the Si dangling bond nanopathway was responsible for the resistive switching of the TiSbTe/a-SiC0.11:H device. The ultra-high uniformity of the TiSbTe/a-SiC0.11:H device ensured the successful implementation of biosynaptic functions such as spike-duration-dependent plasticity, long-term potentiation, long-term depression, and spike-timing-dependent plasticity. Furthermore, visual learning capability could be simulated through changing the conductance of the TiSbTe/a-SiC0.11:H device. Our discovery of the ultra-high uniformity of TiSbTe/a-SiC0.11:H memristor devices provides an avenue for their integration into the next generation of AI chips.