In-Sensor Computing

 Research Topics-In-Sensor Computing

In conventional sensor systems, raw signals are digitized, transmitted, and processed in separate processors, which causes considerable latency, power consumption, and data-transfer overhead as the number of sensors increases. In the emerging era of Physical AI and edge intelligence, sensors are required not only to detect physical and chemical signals but also to interpret them on site. Our laboratory conducts research on in-sensor computing, in which sensing, memory, and computing functions are merged at the device and circuit levels so that information is processed where it is generated. In particular, we develop CMOS-compatible FET-type gas, humidity, pressure, and temperature sensors integrated with embedded micro-heaters, amplifiers, and non-volatile memory arrays, enabling multimodal sensing and signal preprocessing within a single chip. We also exploit the intrinsic physical behaviors of sensors, such as transient responses, low-frequency noise, and sensory adaptation, as computational resources for selective gas identification, and implement in-memory and neuromorphic computing schemes that perform feature extraction and inference directly at the sensor node. Through these studies, we aim to realize low-power, real-time intelligent sensor systems for applications such as artificial olfaction, environmental monitoring, food quality assessment, healthcare, and robotics.