Multilevel Analog Switching in HfO₂/TiO₂ Flexible Memristor for Precision Neuromorphic Computing

Ongoing progress in scientific research continues to extend technological frontiers, with flexible and wearable electronic systems becoming increasingly central to nextgeneration device development. A major bottleneck in this field lies in managing the high thermal budget during the complementary-metal-oxide-semiconductor (CMOS)compatible fabrication line and device performance. Here, we report a CMOS process-compatible stacked flexible memristorbased HfOx, realized on flexible polyimide substrates and specifically designed to tolerate elevated thermal processing for neuromorphic hardware integration. The fabricated device achieves a high ION/IOFF ratio above 1.6 × 10 3 under a 100 µA compliance current and displays excellent uniformity across device arrays. Mechanical durability tests confirmed stable functionality beyond 10 4 cycles at a curvature radius of 5 mm. In addition, endurance exceeding 10 6 WRITE operations demonstrates strong potential for online learning in neural architecture, while the data retention surpassing 10 4 s at 120ºC ensures reliable long-term data storage and inference. The effectiveness of the devices is further verified through multilayer perceptron (MLP) neural network (NN) simulations employing experimentally derived parameters, yielding a nonlinearity (NL) factor of 1.5 and enhanced classification accuracy from baseline for input patterns.