(Invited) CMOS-Based Multimodal Sensing

In the coming Society 5.0, information will be collected, aggregated, and utilized in a wide range of areas through the IoT, and as the concept of "Trillion Sensors" states, the importance of "sensors" as an interface between physical space and cyberspace is increasing. In the development of sensor devices to meet this demand attempts to realize the sensing of multiple items by integrating single-function sensors with high detection selectivity have been the main focus in the past. On the other hand, our research group has proposed the measurement concept of multimodal sensing, in which sensors with broad detection characteristics are used to realize the sensing of multiple items, and has developed semiconductor CMOS-based sensors to realize this concept. In this talk, we introduce ion image sensors and odor sensors as examples of CMOS-based sensors suitable for multimodal sensing, and data analysis using machine learning, which is important for this measurement concept will be mentioned and introduce an example of machine learning-based sensing. The application of these sensors to smart agriculture and other fields will also be discussed, as well as the current status and issues for the practical application of sensor systems. An ion-sensitive transistor (ISFET) can be fabricated by changing the MOSFET's gate to an ion-sensitive membrane. An ion image sensor can be constructed by arraying ion-sensitive elements that operate based on the same principle using CMOS image sensor technology. This sensor visualizes the solution's hydrogen ion concentration (pH) and enables real-time observation of ion dynamics in the solution. By changing the ion-sensitive membrane of the sensor, it is possible to image specific ions such as potassium and calcium, and by adding an enzyme-immobilized membrane, it is also possible to capture chemical substances such as lactic acid. By combining these methods, the development of a multimodal sensor that can simultaneously image multiple ions and chemical substances is in progress as an approach from the hardware side. The future approach from the software side is expected to include measurement data processing by machine learning, thereby expanding the range of substances to be measured. In addition, making this sensor extremely thin makes it possible to perform ion imaging by inserting it into the object to be measured. Functional demonstration of visualization of ion distribution in a plant stem has been performed successfully, and we are currently studying its application in the field of smart agriculture. Since the ion image sensor mentioned above captures minute changes in electric potential generated by ion-sensitive membranes, it can be operated as a gas sensor by forming a membrane on the sensor whose electric properties change depending on gas adsorption. Certain polymer membranes respond to a wide range of gases, they can be used as odor-sensitive membranes. When multiple gas- and odor-sensitive membranes with different response characteristics are formed on a CMOS sensor, gas- and odor-specific response patterns can be obtained. Machine learning is effective in analyzing these patterns, and this presentation will introduce an example of gas-type discrimination.