Development of Self-rectifying TaO_y/Nanoporous TaO_x Memristor Synapse for Suppressing Non-neural Signal in the Large-scale Neuromorphic Array System

Memristor that consists of a metal-oxide layer sandwiched between two conductors is being greatly envisioned as a platform to imitate the principal of biological synapses due to its nonlinear and dynamic electrical properties depending on the history of applied electrical programming^[1]. In this study, we fabricated the nanoporous (NP) TaO_x memristor device by an anodic treatment in the room temperature and utilized the device as the two-terminal artificial synapses^[2]. The device exhibits a stable self-rectifying I-V switching behavior with ~10⁴ nonlinearity, which can effectively prevent the undesired neural signals in the densely-integrated synaptic array. Based on analog shift of the Ohmic-contact site by diverse electrical stimuli, the essential synaptic functions were successfully mimicked. A 16 × 16 crossbar array with only our device was fabricated and statistically evaluated. In addition, we investigated the effect of the nonlinearity of the synaptic device on the accuracy of the pattern recognition using artificial neural network simulation. Taken all together, we believe the designed device can provide a route toward the large-scale neuromorphic computing technology.

References
[1] Wang, G. et al. Nano Lett. 2015, 15, 6009-6014.
[2] Choi, S. et al. under review (2018).