We are developing "smart sensors" that identify and make inferences from acquired information right where it is sensed. Using a method called physical reservoir computing, we embed AI functionality directly into MEMS sensors to realize perception devices that operate with low power consumption. Serving as the "eyes and ears" of physical AI in applications such as robots and autonomous driving, we aim to create edge sensors that, like living organisms, pick out only the information that truly matters.

We are developing ferroelectric nonvolatile memory that retains information even after the power is turned off, combining high integration density, nondestructive readout, and low power consumption. We further pursue its operation as a "analog memory (artificial synapse)" whose state changes in response to the input, aiming to realize next-generation memory that unifies storage and computation.

Using a thin-film fabrication technique called combinatorial sputtering, we simultaneously produce dozens of films with different compositions and growth temperatures in a single deposition. Rather than relying on trial and error, we efficiently search for new materials best suited to sensors and memory. By advancing the entire process from materials development to device prototyping in a seamless, integrated workflow, we provide the foundation that underpins Themes 1 and 2.