Abstract
Various memristor-based synaptic devices have been proposed for implementing a neuromorphic system. However, memristor devices typically suffer from various inherent problems such as nonlinearity and asymmetry of conductance modulation and the sneak path issue of the crossbar array structure. To solve these drawbacks, we propose a one transistor-two memristor (1T2M) synaptic device, its array structure, and its operation method for neuromorphic system applications. For the channel of the transistor and switching layer of the memristor, amorphous InGaZnO was used. The proposed 1T2M synaptic device exhibited more linear and symmetric characteristics of conductance modulation compared with the single memristor device. In addition, the proposed array structure was robust to the sneak path problem. To investigate the switching mechanism, a depth profile analysis of X-ray photoelectron spectroscopy was conducted for each resistance state. Finally, we confirmed an excellent pattern recognition accuracy by using an artificial neural network simulation.
Original language | English |
---|---|
Pages (from-to) | 2837-2844 |
Number of pages | 8 |
Journal | ACS Applied Electronic Materials |
Volume | 2 |
Issue number | 9 |
DOIs | |
State | Published - 22 Sep 2020 |
Keywords
- MNIST
- amorphous InGaZnO
- linear/symmetric synaptic behavior
- neuromorphic
- one transistor-two memristor synaptic device