Bionic gas sensing systems, mimicking biological olfaction, have advanced significantly, finding applications in various fields including air quality monitoring, healthcare, and hazardous gas detection. Semiconductor gas sensors, known for their sensitivity and low cost, are commonly used. However, integrating sensor arrays, circuits, and algorithms poses challenges regarding size, power consumption, and fault tolerance. Multichip approaches offer flexibility but require careful coordination between modules. To enhance selectivity and achieve high-performance systems, multivariate data processing methods, especially artificial neural networks (ANNs), are employed. However, traditional ANNs often require extensive training data, which can be time-consuming to obtain. This research draws inspiration from the rapid response of the *Drosophila* olfactory system, leveraging its efficient odor labeling and processing mechanisms. The goal is to create a compact, energy-efficient system for accurate gas recognition.

The existing literature extensively covers the development and application of bionic gas sensing systems, highlighting advancements in various industrial applications. Several studies explore different gas sensors and their integration into systems. The use of semiconductor gas sensors, particularly metal oxide semiconductors (MOS), is widely reported due to their advantages. However, limitations regarding system integration, power consumption, and fault tolerance are discussed. The use of ANNs for data processing is also common, but the need for substantial training data is highlighted as a significant challenge. This paper addresses this challenge by adopting a bio-inspired approach.

This study presents a biomimetic electronic olfactory sensing system composed of an 18-channel MEMS sensor array, CMOS readout circuits, and a *Drosophila*-inspired algorithm. The MEMS array includes 16 gas sensors, a temperature sensor, and a humidity sensor, all fabricated using a microhotplate with various MOS gas-sensitive nanomaterials. The CMOS circuit simulates the biological information processing of the olfactory system. The *Drosophila*-inspired algorithm uses a sparse, binary random matrix for dimensionality increase and a winner-takes-all mechanism for dimensionality reduction. The system's design mimics the biological olfactory process, from odor molecule detection to signal transmission and processing in the brain. The fabrication process involves multiple steps including dry etching, anisotropic wet etching, platinum heater integration, and inkjet printing of various MOS nanomaterials onto the microplates. The CMOS readout circuit design includes a heating circuit and a resistance-to-voltage conversion circuit. The MEMS sensor array and the CMOS circuit are integrated using flip-chip bonding. The system's performance was evaluated by testing its response to various gases under different conditions, including varying humidity and temperature. The algorithm's performance was assessed by measuring its accuracy in identifying gas types and predicting gas concentrations.

The fabricated 18-channel MEMS sensor array, measuring just 2.5 x 2.5 mm², exhibited excellent performance. The sensors demonstrated rapid response times (11s for 10 ppm acetone), good long-term stability (minimal heater resistance drift after 100 days), and effective heat shielding. Eight types of MOS nanomaterials (SnO2, ZnO, Au-SnO2, WO3, Pt-SnO2, Fe2O3, TiO2, and Pd-SnO2) were used, each on two sensor units. The integrated CMOS-MEMS device, measuring 5 x 8 mm² with a thickness under 1 mm, provides a compact solution. The system achieved 98.5% accuracy in qualitatively identifying seven types of gases and 93.2% accuracy in quantitatively predicting gas concentrations across 3–5 concentration gradients. The system demonstrates robustness to environmental factors and partial sensor damage, owing to the algorithm's ability to handle noisy data and the inclusion of temperature and humidity sensors.

The results demonstrate the successful development of a compact, low-power, and highly accurate biomimetic gas sensing system. The *Drosophila*-inspired algorithm proved highly effective in processing the sensor data, achieving high accuracy even with limited training data and noisy sensor readings. The system's resilience to environmental fluctuations and partial sensor damage addresses significant limitations of traditional gas sensing systems. The findings highlight the potential of bio-inspired approaches to create advanced, efficient, and robust gas sensing technologies. This work contributes to the field by demonstrating a practical implementation of a biomimetic olfactory system with superior performance compared to existing systems.

This research successfully developed a biomimetic electronic nose based on the *Drosophila* olfactory system. This system demonstrates high accuracy in gas identification and concentration prediction, showing resilience to environmental factors and sensor imperfections. Future work could focus on miniaturization, integration with wireless communication for remote monitoring, and expansion of the gas library for broader applications.

While the system demonstrates high accuracy, the study focused on a limited set of gases. Further research is needed to evaluate its performance with a wider range of gases and in more complex real-world scenarios. The long-term stability of the sensor array needs further investigation over extended periods. The algorithm's generalization capability to unseen gas mixtures should also be investigated.