Characterization of a 10 × 10 Amorphous-InGaZnO Memristor Crossbar Array and Its Binary-Neural-Network Performance for 1T1M Integration

Abstract

This study reports the fabrication and characterization of a 10 & times; 10 passive amorphous-InGaZnO (a-IGZO) memristor crossbar array together with a-IGZO thin-film transistors (TFTs), aiming at selector-assisted one transistor and one memristor (1T1M) integration for neuromorphic hardware. Mo/Al2O3/IGZO/Pd memristors exhibit stable bipolar switching governed by Schottky barrier modulation via electric-field-driven oxygen-vacancy redistribution, achieving an ON/OFF ratio above 10(2) at a 1-V read condition and endurance over 500 cycles. Retention is evaluated over a similar to 3600-s time window at 1 V after ensuring a fully programmed low-resistance state (LRS) using a 10-V SET pulse, and the relaxation behavior is quantified using a stretched-exponential fit with a characteristic time constant on the order of 10(3) s. As a proof-of-concept, a 4 & times; 4 binary character classification is demonstrated using the passive array and a winner-takes-all readout with a differential multibottom electrode scheme to quantify sensing margins. In addition, SPICE simulations based on the measured TFT and memristor characteristics are performed for an assumed 1T1M active-matrix configuration, showing enlarged readout margins compared with the passive-array measurements. These results highlight the feasibility of a-IGZO as a unified material platform for both selector and memristive elements and its potential for scalable, energy-efficient neuromorphic arrays.