Posco DX Develops Industrial Unstructured Data Analysis Platform Utilizing Domestic NPU

Integrated Management Platform for Video Data and AI Models
Collaboration with Domestic NPU Developers such as DeepX and Mobilint
GPU Handles Model Training, NPU Responsible for Inference

Posco DX announced on the 18th that it has developed an industrial site unstructured data analysis platform utilizing domestically produced artificial intelligence (AI) semiconductors.

Posco DX integrated a domestically developed neural processing unit (NPU) into its vision AI platform. The newly developed platform operates video data collected from industrial sites as training data in a single environment and has standardized features that are repeatedly required for service implementation, such as AI model management, performance metrics, and application history.

POSCO DX announced on the 18th that it has developed an industrial site unstructured data analysis platform using domestic artificial intelligence (AI) semiconductors. POSCO DX AI researchers are testing an AI program that estimates loading capacity in logistics environments. POSCO DX

An NPU is a semiconductor specialized for AI computations including deep learning and machine learning. Compared to a graphics processing unit (GPU), it is optimized for AI inference, which can reduce infrastructure costs and power consumption.

When an NPU is directly installed in on-site facility control systems, it enables the implementation of edge AI without routing data through an AI data center or server. This allows for inference to occur on-site without external data transfer, which is advantageous for ensuring both security and real-time operation in manufacturing environments where data protection is a core requirement.

The vision AI platform uses a hybrid approach: GPUs are still used for training and developing AI models, but domestically produced NPUs are applied for real-time computation and decision-making at industrial sites. Even in the research and verification stages, the NPU can be used to pre-validate inference performance and effectiveness before actual on-site application. This lets users verify AI model accuracy, processing speed, and energy efficiency in advance of deployment.

Previously, Posco DX has promoted the localization of industrial AI based on partnerships with domestic NPU developers such as DeepX and Mobilint. Through these efforts, the company confirmed that infrastructure construction and operation costs were reduced by approximately 50% and power consumption by about 90%, compared to achieving the same inference performance with GPUs.

A Posco DX representative stated, "The NPU-based vision AI platform will serve as a foundation for implementing AI optimized for industrial sites more efficiently," adding, "We will continue to systematize the use of domestic NPUs and enhance industrial inference capabilities, contributing to the revitalization of the AI semiconductor ecosystem."

This content was produced with the assistance of AI translation services.