Enhanced Weight-Update Characteristics in Neuromorphic Computing Using Vertical Electrochemical Random-Access Memory with PEALD-TiO₂ Channels

Cross-point arrays of analog synaptic devices are expected to realize neuromorphic computing hardware for large-scale artificial neural networks with significant enhancements in speed and energy consumption compared to conventional hardware based on the von Neumann architecture. To achieve energy efficiency and the desired characteristics of analog synaptic devices for fully parallel vector–matrix multiplication and vector–vector outer-product updates, vertically-structured metal-oxide based electrochemical random-access memory has been proposed as a promising synaptic device. This is due to its CMOS-compatibility and minimal cell size, which provide enhanced synaptic device performance and characteristics, enabling increased scalability for energy-efficient neuromorphic computing. To realize the full potential of VECRAM devices and minimize the device-to-device variations, conformal deposition of thin films along the vertical sidewall by atomic layer deposition (ALD) technique is crucial. However, depositing a channel layer with proper electronic- and ionic-characteristics by ALD has been challenging due to the surface-reaction limited reactions in conventional ALD process.<br/><br/> In this study, we employ a plasma-enhanced ALD (PEALD) technique with an additional Ar plasma step to deposit a TiO₂ channel layer for VECRAM devices. By adjusting the power and duration of the Ar plasma step during the ALD cycles, we control the degree of crystallinity and off-stoichiometry of TiO₂ films. The ionic and electronic transport characteristics of PEALD-TiO₂ are evaluated by electrochemical impedance spectroscopy and compared to film deposited by conventional ALD. We fabricate VECRAM devices to investigate the effect of the PEALD-TiO₂ properties on synaptic device characteristics. The microstructure and the conformality of the channel layer on the vertical sidewall are characterized by electron microscopy. The VECRAM device with a PEALD TiO₂ channel exhibits improved linearity and improved variations in the weight-update operations, thereby demonstrating enhanced performance for high-density synaptic array-based neuromorphic computing applications.