Neurocle, a company specialized in deep learning-based vision inspection software, has announced its collaboration with Korean NPU company DeepX at CES 2026 to unveil an Edge AI-based NPU deep learning vision inspection solution, aiming to penetrate the global industrial AI market. The two companies will combine Neurocle's auto deep learning technology with DeepX's NPU hardware to present a new standard for real-time, low-power vision inspections required in industrial settings.
 
The solution being showcased is centered on optimizing deep learning vision inspection models trained using Neurocle's automatic model optimization algorithm to run on DeepX's NPU accelerator, 'DX-M1'. The DX-M1 is designed based on the M.2 form factor, making it easy to apply to industrial PCs and various Edge devices, and will be demonstrated in real-time at the DeepX booth during CES.
 
NPU is a processor specialized for deep learning operations, offering high computational efficiency and stable processing performance in vision inspection inference compared to CPU or GPU. In particular, its ability to perform inline inspections under low-power conditions is considered a key advantage in manufacturing environments where repetitive and real-time processing is essential. DeepX’s NPU is optimized for Edge environments based on low heat generation and excellent power efficiency, reducing dependency on servers or the cloud while enabling immediate judgment and response on-site.

The collaborative solution between Neurocle and DeepX can be broadly applied across various manufacturing sectors, including secondary batteries, electronics, semiconductors, and pharmaceuticals, rather than being confined to specific industries. It is expected to enhance quality inspection accuracy through stable deep learning inference in Edge environments while reducing equipment installation costs and operational burdens.
 
Neurocle also emphasized hardware flexibility. It supports a wide range of device environments, including NPU, CPU, GPU, and embedded boards, tailored to client production environments and operations, and is continuously expanding optimization strategies that are not tied to specific hardware. This policy allows each company to choose the AI infrastructure that best fits on-site conditions.
 
Furthermore, Neurocle will pre-release its deep learning training engine 'Neuro-T' engine, which allows for on-device learning and re-learning, at CES 2026. Set to officially launch in March, this engine is expected to provide a practical alternative for companies looking to build MLOps and internalize AI learning, offering only core learning functionalities based on auto deep learning algorithms without a GUI.