Real-time thermokinetic combinatorial fingerprinting of toxic gases with a MEMS e-nose

Real-time monitoring of toxic gases requires instruments capable of detecting trace analytes while discriminating among chemically similar species. Although electronic noses (e-noses) are promising, robust quantitative gas analysis requires enhanced selectivity and specificity. Here, we present a micro-electromechanical system (MEMS) e-nose utilizing an array of CuO-modified SnO 2 nanoparticle sensors. By coupling temperature modulation with transient feature extraction, we activate gas/oxide interfacial redox processes to generate information-rich electrical signatures. These thermokinetic descriptors are analyzed using linear discriminant analysis (LDA) to optimize feature separation. Furthermore, a sliding-window convolutional neural network-long short-term memory (CNN-LSTM) framework is implemented for real-time gas fingerprinting and continuous concentration regression. Through tightly coupled artificial intelligence (AI)-sensor integration, the CNN-LSTM model decodes subtle patterns within transient signals. This capability allows the e-nose system to promptly discriminate between gases with similar response signatures, establishing a scalable pathway for high-specificity toxic gas monitoring in compact instruments.