en.Wedoany.com Reported - South Korean AI semiconductor company Mobilint plans to begin mass production of its second-generation neural processing unit (NPU), REGULUS, by the end of this year. It has recently delivered customer samples (CS) to clients, officially entering the full product certification phase.
Mobilint CEO Dongjoo Shin stated at an AI semiconductor forum breakfast seminar held in Seoul on July 9 that REGULUS customer samples have begun commercial sales, and reliability verification with QRT has also been completed. He noted that although mass production is three to four years behind overseas competitors, the timing coincides with the rise of the physical AI market, where the chip will be deployed.
Currently, approximately 90 companies are evaluating Mobilint's products (including REGULUS), up from around 70 at the beginning of the year. Potential customers include LG Electronics, POSCO DX, Doosan Robotics, Kolon Benit, LOTTE Innovate, Shinsegae I&C, Daedong, and Intellivix.
REGULUS is a 17 mm × 17 mm system-on-chip (SoC) that integrates a central processing unit (CPU), video codec, NPU core, image signal processor (ISP), and scratchpad memory (SPM) intellectual property (IP). The chip supports Ethernet and USB interfaces, uses LPDDR4X memory, and is manufactured using TSMC's 12-nanometer process.
The processor targets applications such as smart cameras, AI self-service kiosks, industrial robots, and agricultural machinery. Shin stated that the company has received an increasing number of inquiries from customers seeking alternatives to Chinese chips supplied by vendors like Rockchip.
Mobilint is also targeting on-premises and edge device AI systems, enabling edge AI inference without relying on cloud connectivity. Its main global competitors in the edge AI chip market include NVIDIA's Jetson platform, Qualcomm, and Hailo.
Shin stated that Mobilint's chips offer significant efficiency advantages over competitor products. He noted that edge AI requires high performance, competitive pricing, and energy efficiency, and that its chips can halve customers' capital expenditure (CAPEX), reduce operating costs by three to four times, and improve energy efficiency by five times compared to competing solutions. He added that edge computing offers lower latency, stronger communication security, and higher energy efficiency compared to cloud-based computing.
According to Shin, software compatibility is another key differentiator. He stated that entering the market without broad compatibility is impossible. Mobilint's software development kit (SDK) supports 400 AI models in commercial production environments, spanning large language models (LLMs), vision language models (VLMs), and vision-language-action (VLA) models. Shin noted that no other global edge AI NPU supplier currently supports VLA models, describing this capability as Mobilint's unique competitive advantage.
The company has also begun developing a successor to REGULUS, adopting a chiplet-based architecture and manufactured using a 4-nanometer process. Shin stated that developing and commercializing a 4nm chip requires at least 100 billion Korean won. Mobilint recently raised 70 billion Korean won in a Series C funding round. The next-generation chip will be manufactured using Samsung Electronics' 4nm process, with SEMIFIVE serving as the design house. Chiplet technology packages multiple smaller chips into one processor rather than relying on a single large monolithic chip. As system-on-chip functionality continues to expand, larger chip sizes reduce manufacturing yields. The chiplet architecture distributes key functions across multiple chips, improving yields while offering greater scalability.
