We have made significant progress in developing low-power gas sensors using micro-heaters and micro-LEDs. The micro-heaters are suspended to minimize heat loss through conduction, reducing energy consumption to under 10 mW, ideal for long-term battery operation [1, 2, 3, 4, 5]. Additionally, our monolithic micro-LED based gas sensor enables efficient light energy transfer for ultra-low power gas detection at around 0.1 mW [6]. To support these sensors, we designed a control module with multi-channel transmission, and individual micro-heater and micro-LED control, ensuring precise sensor singal measurements [7]. The module integrates multiple variable resistors and high-speed multiplexed switching to accurately measure the wide resistance range of semiconductor metal oxide (SMO) in response to target gases, enhancing detection accuracy and reliability. Furthermore, the MEMS sensors developed by our lab exhibit excellent sensor-to-sensor uniformity, are scalable for mass production, and operate at low power. These characteristics ensure stable and reliable long-term operation in various environments, making them highly suitable for integration with IoT technologies. This opens up significant potential for widespread use in a variety of applications, regardless of location or conditions.

We have integrated deep learning techniques with gas and physical sensor technologies to overcome sensor limitations and expand into various applications. By applying deep learning algorithm, we developed an electronic nose (e-nose) system that recognizes patterns in signals from different gas sensor arrays, allowing it to distinguish between different gases and accurately predict their concentrations, providing more valuable information [8, 9, 10]. Additionally, we applied deep learning to improve the spatial resolution of pressure sensors, enabling precise detection of force location and intensity, which opens up applications in VR/AR and wearable sensors [11].

To address the high computational demands of deep learning, we have developed a bio-inspired artificial olfactory neuron module that mimics the spiking behavior of human olfactory neurons [12]. This neuromorphic approach allows for in-sensor computing, significantly reducing the power required for processing CNN algorithms. By integrating this technology with our gas sensors, we have created a more efficient e-nose system with dramatically lower power consumption. Beyond gas sensing, we are extending this neuromorphic computing technology to tactile sensors, exploring its potential in augmented reality (AR) and virtual reality (VR) applications. By incorporating AI-driven tactile feedback, we are working towards enhancing the interactivity and realism of AR/VR systems, making them more responsive and immersive.