Efficient and Stable Topological/Ferroelectric Bi₂Te₃/SnSe Hetero-Memristor for In Situ Bionic-Visual Semi-Hardware Systems.

Abstract

As the application of artificial vision systems continues to grow, developing efficient and low-power visual sensing devices has become a key challenge. Memristors offer tunable conductivity and integrated in-situ storage and computation functions, making them ideal for low-cost visual systems. However, most memristors currently face the dual challenges of poor stability and limited optoelectronic synaptic plasticity. Here, a Bi₂Te_2.7Se_0.3/SnSe hetero-memristor is designed, which combines the advantages of two-dimensional (2D) topological insulators and 2D ferroelectric materials. The hetero-memristor performance can be tuned by the SnSe ferroelectric polarization and Bi₂Te_2.7Se_0.3 topological surface state, which improve the utilization and mobility of carriers, thereby significantly improving the performance. The high 10⁴-cycle stability, average 0.25 µW on/off power, and 2⁵ conductive states are achieved. Under different signals, the hetero-memristor can enable in situ light-electric conversion and successfully simulate various optoelectronic plasticity behaviors, such as paired-pulse facilitation, post-tetanic potentiation, spike rate-dependent plasticity, etc. Mean while, an efficient in-situ bionic-visual semi-hardware system is constructed based on the 28 × 28 perception hetero-memristor array. This system efficiently performs satellite image recognition and classification, achieving an accuracy of 97.68%. The research shows that the Bi₂Te_2.7Se_0.3/SnSe hetero-memristor is with excellent optoelectronic performances and broad application prospects, particularly in brain-like computing, smart hardware, and storage technologies.