Solution‐Processed HZO–IGZO Thin Film Transistors for Multiclass Classification Tasks

In recent years, thin film transistors (TFTs) based on HfZrO (HZO) have made significant progress in the fields of memory and neuromorphic computing. However, the fabrication of conventional HZO TFTs usually relies on sophisticated techniques such as atomic layer deposition and rapid thermal annealing. Although these techniques improve device performance, they may increase cost and process complexity. To address this issue, this work proposes a low‐cost and simple solution‐based strategy for fabricating HZO TFTs. The fabricated devices exhibit typical counterclockwise hysteresis, successfully emulating various synaptic plasticity behaviors, as well as achieving stable multilevel conductance states under consecutive pulse stimulation. A three‐layer fully connected neural network constructed using the experimentally measured discrete conductance states of the devices achieves a recognition accuracy of 95% on the MNIST handwritten digit recognition task after 50 training epochs. Furthermore, the weight‐replacement model is extended to a ResNet‐18 deep convolutional neural network, attaining an accuracy of 87% on a 102‐class flower image classification task. The proposed solution‐based process, which requires no complex vacuum equipment, is simple and low‐cost, providing a feasible technological pathway for the development of low‐cost neuromorphic computing and image enhancement applications based on HZO TFTs.