Correlation between Sensing Accuracy and Read Margin of a Memristor-Based NO Gas Sensor Array Estimated by Neural Network Analysis

Memristor-based gas sensors (gasistors) have been considered as the most promising candidate for detecting NO gas suitable for neural network (NN) analysis. In this work, in order to solve an overfitting issue arising from the training data when using a single gasistor, which degrades the accuracy of NN, we here propose a metal–insulator-silicon (MIS)-structured Zr 3 N 4 -based gasistor array that results in an improvement in both the accuracy of the NN analysis and the efficiency of the operating power. As a result, the proposed gasistor array showed a decrease of epoch and a 2.5% improvement of prediction accuracy at room temperature compared to single cells with metal/insulator/metal (MIM) and MIS structures. These results imply that an array structure based on MIS can efficiently solve the overfitting issue by receiving multiple responses at once, compared to a single gas sensor that obtains one response per sensing.