Five partners plan 2nm low-power AI chip for medical imaging
LSTC, Hokkaido University, Fujifilm, GE HealthCare and imec announced they will undertake a NEDO project to develop a 2nm low-power AI inference chip for medical imaging. The team plans to develop and optimize workloads on IBM Spyre before migrating them to the chip, with intended uses including photon-counting CT reconstruction, real-time ultrasound image overlay and surgical-image recognition. The development targets are more than five times the image-processing speed and ten times the energy efficiency of existing GPUs, followed by clinical use at Hokkaido University Hospital; these performance figures are unverified targets, not independent measurements.
The project combines advanced semiconductor development with clinical imaging workloads, while its speed and energy-efficiency goals remain unverified development targets.
IT之家 10 月 23 日消息，LSTC 日本尖端半导体技术中心、北海道大学、富士胶片、GE 医疗、imec 五方当地时间本月 20 日宣布承接日本 NEDO 项目委托，共同开发基于 2nm 工艺的医疗影像低功耗 AI 推理芯片。
该芯片的设计应用场景包括：
- 对光子计数 CT 提供的海量数据进行高速重构与分解，实现“实时影像诊断支持”；
- 将超声影像与 CT / MRI 影像实时叠加，实现“实时诊疗支持”；
- 从术中显微镜影像里实时识别手术器具、解剖结构、出血，实现“实时手术支持”。
图源：Pexels
合作各方计划先行在 IBM 的 Spyre 加速器上进行开发优化工作，此后再迁移到实际芯片上。其开发目标是提供现有 GPU 5 倍以上的图像处理速度和 10 倍以上的能效，并在北海道大学附属医院进行临床应用。
#LSTC#Hokkaido University#Fujifilm#GE HealthCare#imec#NEDO#IBM Spyre#Hokkaido University Hospital