The first-generation medical AI specialist Deepnoid has signed a memorandum of understanding (MOU) with domestic AI semiconductor company FuriosaAI and will begin commercial operation of AI semiconductors. Deepnoid was founded in 2008 and was listed on KOSDAQ in 2021 under the special listing regime for technology growth companies. The company is currently developing solutions for medical image reading and diagnostic reporting AI such as M4CXR, DEEP:NEURO, DEEP:LUNG for pulmonary disease response, Skymaru:Security, DEEP:Security, and DEEP:Factory for security and factory automation.


FuriosaAI is an AI semiconductor startup founded in 2017 that offers neural network processors (NPU) for data centers and edge devices. Its first-generation NPU Warboy is a semiconductor optimized for vision recognition computation and delivers performance of up to 64 TOPS (64 trillion operations per second) at INT8 (8-bit integer). The second-generation NPU RNGD, now in mass production, is an NPU for data centers and infrastructure, a high-end product with 512 TOPS performance on an INT8 basis and 48GB of HBM3 memory.

In terms of applications, it supports high-performance large language models (LLM), multimodal LLMs, vision models, and other deep learning models. At the OpenAI Korea opening ceremony in September, FuriosaAI demonstrated OpenAI’s gpt-oss 120B model as an on-device AI using two RNGD units, proving excellent LLM processing performance and efficiency.


This collaboration focuses on using FuriosaAI’s second-generation semiconductor RNGD (Renegade) to process data analysis for M4CXR, a generative AI-based solution that creates draft reading reports for chest X-rays. M4CXR is a solution that automatically generates draft reading reports from chest X-ray data.

Deepnoid began deep learning research in 2015 and has developed technologies that detect cerebral aneurysms based on magnetic resonance angiography (MRA) results, as well as analyze and detect diseases such as colorectal cancer, renal cancer, and gastric cancer from various pathology images. Through R&D, the company developed the AI-based chest X-ray diagnostic support solution DEEP:CHEST and is upgrading its functions into M4CXR.

Last year, through the National IT Industry Promotion Agency’s (NIPA) “AI Semiconductor Application Demonstration Support Project,” Deepnoid carried out a proof-of-concept project that used RNGD instead of a graphics processing unit (GPU) for M4CXR inference tasks. At that time the focus was on text applications, but RNGD is expected to support vision models such as Stable Diffusion XL, offering potential use in future image analysis. As an extension of that collaboration, the two companies plan to jointly promote domestic and overseas commercialization of M4CXR and RNGD and to conduct ▲integration and review of existing technologies ▲pilot projects ▲joint business development ▲participation in national projects.


Deepnoid CEO Choi Woo-sik said, “This collaboration is significant in that it secures both cost efficiency and operational stability, which are essential for the large-scale deployment of medical AI solutions,” adding, “Through this, we will lower the barriers to adoption of AI solutions, including the generative AI-based solution M4CXR, and build a sustainable business model.”

FuriosaAI CEO Baek Jun-ho said, “Through this collaboration, we expect this to become an important case demonstrating excellent performance and efficiency of NPU-based medical systems in real medical settings,” and added, “FuriosaAI will actively contribute to the digital transformation of the medical field.”

IT Donga reporter Nam Si-hyun (sh@itdonga.com)

Popular News