Doping Engineering for Optimized Self‐Rectifying TaOx Memristor for Crossbar Array Neuromorphic applications

Abstract Oxide‐based memristors are promising candidates for artificial neural network computations using neuromorphic hardware. However, when arranged in large‐scale arrays, their performance is often hindered by challenges such as poor reliability and sneak currents. To address these, Al, N‐doped TaO x (ANTO) memristors are precisely engineered in this study using atomic layer deposition (ALD). The oxygen vacancy (V o ) content in TaO x can be regulated by controlling the dopant concentrations via the number of ALD cycles, thereby effectively adjusting the Schottky barrier height and surface resistance. This optimizes the oxygen ion‐based interface‐type resistive switching while improving the self‐rectifying properties. Furthermore, V o passivation by N‐doping effectively suppresses sneak currents in crossbar arrays. Optimized devices exhibit endurance exceeding 3.5 × 10 4 cycles and 3‐bit retention over 10 3 s. Their long‐term potentiation and depression characteristics are highly accurate in synaptic neural network‐based image recognition tasks. Hardware‐level demonstrations using the fabricated 32 × 32 crossbar array confirm stable analog switching, reliable multibit programming, and effective suppression of sneak‐path currents under practical operating conditions. Notably, selective multilevel programming is validated through conductance mapping, demonstrated by visualizing a word pattern within the crossbar array. These results establish ANTO memristors as a promising platform for high‐density, energy‐efficient neuromorphic computing systems.