MOSAIC: Multimodal On-device Sensory AI Computing
Keyword Keywords
OBJECTIVE Objective
This project aims to secure the core technology of MOSAIC (Multimodal On-device Sensory AI Computing), a platform that integrates sensory input, AI inference, and actuation output in a single system, by developing four sensory input devices, sensor-NPU interface circuits, and a multimodal NPU core. Through this, it seeks to verify the feasibility of a Physical AI accelerator platform capable of low-power, on-device multimodal inference.
INTRODUCTION Message from the Principal Investigator
Greetings.
Director of MOSAIC (Multimodal On-device Sensory AI Computing), the Physical AI Semiconductor Research Center.
Physical AI is advancing on the premise of on-device intelligent systems that interact with the physical environment in real time. To realize this potential, our center aims to integrate sensory input devices, sensor-NPU interface circuits, and a multimodal NPU core into a unified design, securing core technology in which sensory input, AI inference, and actuation output are organically connected on a single platform, and to extend this technology into real-world Physical AI applications.
By closely linking GIST's strengths in AI semiconductor design and sensory input devices with the systemization and demonstration infrastructure of research institutes, and through postdoc-centered convergence research, we will create core semiconductor technologies that serve industry beyond academic achievement, and cultivate the key talent who will lead the field of Physical AI semiconductors.
Thank you.
IMPACT Expected Impact
By fusing and processing heterogeneous sensory signals in real time on a single semiconductor platform, this research significantly improves the response speed and energy efficiency of on-device multimodal inference.
The resulting NPU and sensor-AI interface design methodologies can serve as design assets for domestic AI semiconductor fabless companies, which can accelerate technology transfer and commercialization into fields such as robotics, autonomous driving, medical devices, and edge AI.
By fostering convergence experts spanning devices, circuits, NPUs, and applications, it contributes to producing core research talent for the Physical AI semiconductor field.
TALENT Ideal Talent Profile
The center aims to cultivate convergence researchers who can encompass and organically connect sensory input devices, interface circuits, NPU architectures, on-device AI algorithms, and Physical AI applications.
At the same time, it seeks individuals with the collaborative ability to lead joint research with national research institutes and other organizations, a commitment to reproducible research culture and the sharing of data and results, and the leadership and international communication skills to grow into independent researchers.