Transformation of digital to analog switching in TaOx-based memristor device for neuromorphic applications

Abstract

An oxidizable metal diffusion barrier inserted between the active metal electrode and the switching layer decreases the electroforming voltage and enhances the switching stability and synaptic performances in TaOx-based conducting bridge memristor devices. The TiW barrier layer avoids an excessive metal ion diffusion into the switching layer, while the TiWOx interfacial layer is formed between the barrier and the switching layer. It modulates the oxygen vacancy distribution at the top interface and contributes to the formation and rupture of the metal ion-oxygen vacancy hybrid conducting bridge. We observe that the device that relies upon non-hybrid (metal ions only) conducting bridge suffers from poor analogous performance. Meanwhile, the device made with the barrier layer is capable of providing 2-bit memory and robust 50 stable epochs. TaOx also acts as resistance for suppressing and a thermal enhancement layer, which helps to minimize overshooting current. The enhanced analog device with high linear weight update shows multilevel cell characteristics and stable 50 epochs. To validate the neuromorphic characteristic of the devices, a simulated neural network of 100 synapses is used to recognize 10 × 10 pixel images.