POSCO DX develops unstructured data analysis platform using domestic AI semiconductors

POSCO DX has developed an unstructured data analysis platform utilizing domestic Korean AI semiconductors, specifically Neural Processing Units (NPUs). The platform is designed to support the localization of AI in industrial settings by integrating video data and AI model management.

The system employs a hybrid architecture that uses Graphics Processing Units (GPUs) for AI model training and development, while utilizing NPUs for real-time inference and decision-making on-site. This approach enables ‘Edge AI,’ allowing for immediate processing within local equipment control systems without the need to transmit sensitive data to remote data centers, thereby enhancing security.

By collaborating with domestic NPU developers such as DeepX and Mobilint, POSCO DX has demonstrated significant efficiency gains. Compared to using GPUs for inference, the NPU-based platform has shown a reduction in infrastructure and operation costs by approximately 50% and a decrease in power consumption by about 90%. The company plans to expand the application of this Vision AI platform to various industrial tasks, including fire monitoring, worker safety, and logistics management.