CoinDesk reports:
IT Home, October 23: On the 20th of this month, LSTC Japan Advanced Semiconductor Technology Center, Hokkaido University, Fujifilm, GE Healthcare, and imec jointly announced they have been entrusted by Japan's NEDO to develop a low-power AI inference chip based on a 2nm process for medical imaging.
The design applications for this chip include:

- Rapidly reconstruct and decompose the massive data provided by photon-counting CT to enable real-time imaging diagnostics;
- Overlay ultrasound images with CT/MRI images in real time to enable "real-time clinical decision support";
- Real-time identification of surgical instruments, anatomical structures, and bleeding from intraoperative microscope images to enable "real-time surgical support."
The collaborating parties plan to initially develop and optimize on IBM’s Spyre accelerator before migrating to actual chips. Their development goal is to achieve more than five times the image processing speed and over ten times the energy efficiency of current GPUs, with clinical applications to be conducted at the affiliated hospital of Hokkaido University.
