High-Performance Computing

MOSAIC:Memory-cube Operability in a Stacked AI Chip

Features
3D stacking technology reduces data transmission distances from micrometers to nanometers.
No need for high silicon IP costs of inter-chip communication.
Software-defined computing cores can flexibly adjust memory bandwidth and capacity according to different application specifications.
Modular, multi-tiered, easily scalable, supporting various AI application scenarios.

Description

Generative AI is a highly popular topic at the moment, driving significant economic growth and transforming our lives. It has gradually expanded from cloud computing to edge computing, encompassing consumer electronics, smart homes, the Internet of Things, and more. However, the development of specialized AI chips still faces many challenges, such as low production volumes and diverse customized specifications. This is especially true for deep neural networks, which require massive computational and memory resources. MOSAIC is the world's first flexible extension 3D chip designed for computing and storage integration. It maximizes space, performance, and power consumption benefits by combining AI wafer-level systems to suit generative AI's common operation requirements. Scalability, high performance, and low production costs are the distinguishing features of MOSAIC's semiconductor performance.
 

Dept:Electronic and Optoelectronic System Research Laboratories
POC:王耀宗
Tel:03-5913316
E-mail:victor.wang@itri.org.tw