Oxide interface-based polymorphic electronic devices for neuromorphic computing

Abstract

Abstract Aside from recent advances in artificial intelligence (AI) models, specialized AI hardware is crucial for addressing large volumes of unstructured and dynamic data. Conventional complementary metal-oxide-semiconductor (CMOS)-based AI hardware faces several critical challenges including scaling limitations, the separation of computation and memory units, and overall system energy efficiency. While emerging materials have been proposed to overcome these limitations, issues such as scalability, reproducibility, and compatibility remain critical obstacles. Here, we demonstrate polymorphic electronic devices with programmable transistor, memristor, and memcapacitor functionalities by manipulating the quasi-two-dimensional electron gas in LaAlO 3 /SrTiO 3 heterostructures using lateral gates. A circuit utilizing transistor and memcapacitor functionalities exhibits digit recognition, enabling implementation in physical reservoir computing. An integrated circuit incorporating transistor and memristor functionalities performs logic operations with in-situ output storage and supports advanced reconfigurable synaptic logic operations for multi-input decision-making tasks such as patient monitoring. Our findings pave the way for oxide-based monolithic integrated circuits in a scalable, silicon-compatible, energy-efficient single platform for polymorphic and neuromorphic computing.