The core of the semiconductor business is overseas expansion. Semiconductors utilize internationally standardized specifications and interfaces, making them usable in any country worldwide regardless of the manufacturing country. Additionally, the supply chain for the design, manufacturing, assembly, and sale of semiconductors spans multiple countries, allowing businesses to start in an already internationalized ecosystem. The fact that semiconductors are essential in almost all products, such as smartphones, automobiles, home appliances, and industrial equipment, and the recent spread of artificial intelligence (AI) leading to multifaceted demand for AI semiconductors across all industries, are reasons why the semiconductor industry must expand overseas.


Considering the characteristics of the semiconductor industry, the Ministry of Science and ICT and the National IT Industry Promotion Agency (NIPA) are assisting semiconductor companies in their overseas expansion through the '2025 AI-Semiconductor Overseas Demonstration Support Project.' This project aims to secure global adoption cases for promising domestic AI semiconductors and AI services and to lay the groundwork for overseas expansion. Initially, the selection was completed in April this year, but considering the importance of the project, a second selection was conducted in July, with a total of eight consortia receiving support.

Comprehensive Support for Server and Edge Semiconductors, AI Solutions, etc.

The AI-Semiconductor Overseas Demonstration Support Project targets two types of companies: semiconductor companies and AI service companies. The AI semiconductor support field is divided into two categories: the server-type sector, which implements an overseas demonstration environment using domestic AI semiconductors and enables local service of AI models on overseas servers, and the edge-type sector, which enhances edge AI models that can perform inference on digital devices based on domestic AI semiconductors and conducts local demonstrations. AI service companies are those that have completed the development of services based on AI models and possess solutions.


In the first phase of the project, ▲ Simple Platform-Ditec ▲ Vernect-Rebellion, and B2N were selected for the server sector, while ▲ Ottom-UX Factory, Free Soft ▲ DeepX-Nota were selected for the edge sector. Although the project was initially planned for only 10 companies, a second selection was conducted following the confirmation of an additional budget on May 1. Consequently, ▲ Douzone Bizon-FuriosaAI ▲ EcoPeace-Rebellion were additionally selected for the server sector, and ▲ LBS Tech-Dinoticia ▲ RaonPeople-Mobilint were selected for the edge sector. Meanwhile, to support entry into as many countries as possible, the project supports entry into Taiwan, Mongolia, the Philippines, and Malaysia, while the additional budget supports entry into the UK, the Middle East (UAE), Japan, and Thailand.

Douzone Bizon and FuriosaAI, selected in the second phase, are conducting demonstrations of resource and accounting management services targeting Japanese SMEs, while EcoPeace and Rebellion are conducting demonstrations of AI video recognition-based water pollution detection and autonomous purification solutions in collaboration with UAE mid-sized companies. LBS Tech and Dinoticia are conducting demonstrations of wheelchair data collection and safety navigation construction in collaboration with the West Midlands Transport Authority in the UK, and RaonPeople and Mobilint are developing on-device AI systems for disaster, safety, and crime prevention in collaboration with the Bongkan Provincial Government in Thailand.

Case of Douzone Bizon and FuriosaAI's Entry into Japan

Among the eight consortia, the case of Douzone Bizon and FuriosaAI is particularly noteworthy. The Ministry of Economy, Trade and Industry (METI) of Japan announced the DX (Digital Transformation) Promotion Guidelines 2.0, aiming to increase the digital adoption rate of SMEs to over 60% by 2025, and provides IT Introduction Subsidy to support part of the cost when SMEs and small businesses introduce IT tools suitable for their business.


With the spread of IT adoption in Japan, the number of AI startups entering the Japanese market from Korea is increasing, but the entry barriers to the Japanese market remain high. Amidst this, as Douzone Bizon, Korea's leading ERP company and a pioneer in AI transformation, collaborates with unicorn AI semiconductor company FuriosaAI to target Japan, not only startups but also SMEs and mid-sized companies need to pay attention to this move. Furthermore, it is important that the project targets the entire range of companies needing ERP, rather than just pursuing specific businesses.

The solution that Douzone Bizon will introduce to the Japanese market is ONE AI. ONE AI is a company-customized AI service embedded in ▲ ERP covering production, accounting, cost, HR, sales, and employee management ▲ groupware consisting of schedule and resource management, messenger, work management, portal, electronic approval, email ▲ electronic document management solutions (EDM) such as document centralization and authority management, information leakage prevention, document history verification, and document sharing.


Douzone Bizon has equipped ONE AI with generative AI functions using two models: the Lama-Douzone 70B AI model, optimized for company work management, and the LG Exaone 3.5 32B model, optimized for fast and lightweight operation. Based on this, company-customized AI functions such as ERP data analysis to enhance corporate productivity, AI schedule registration and approval writing, and automatic report generation are provided. Douzone Bizon is enhancing ERP to suit the local situation in Japan and applying a Japanese language-based UI/UX.


The computation for running ONE AI is powered by FuriosaAI's second-generation AI neural processing unit (NPU), RNGD. RNGD is a semiconductor optimized for building high-performance, high-efficiency inference infrastructure for large enterprises and clouds, emerging as an alternative to NVIDIA GPUs. In July, LG AI Research Institute also decided to adopt RNGD, announcing that RNGD recorded a power consumption-to-performance ratio 2.25 times that of existing GPUs.


On the software side, it supports ▲ large language model (LLM) operation and parallel execution ▲ a compiler and runtime environment that converts and executes AI models in a form optimized for RNGD ▲ cloud-native toolkit that helps support RNGD in cloud environments ▲ infrastructure that assists in managing and controlling hardware devices. This allows for the integrated use of the entire process from AI model development to deployment, management, and utilization. FuriosaAI supports semiconductor migration with a professional support organization consisting of solution architects and engineering teams to ensure smooth service transition. FuriosaAI's software stack is well-received by customers for its speed and ease in the model support process, and it is already evaluated as being in the commercialization stage in the global market in terms of maturity.

Douzone Bizon and FuriosaAI propose a phased introduction plan aiming for comprehensive verification of the function, performance, usability, and stability of AI NPU-based SaaS. In the first phase, they verify the normal operation of NPU-based ERP AI functions and check the basic functions of the PoC target service. In the second phase, they verify the UI/UX for Japanese users and check local suitability. In the third phase, they conduct performance optimization by comparing performance indicators such as power consumption to response speed, NPU inference speed, and accuracy. In the fourth phase, they modify functions and improve qualitative satisfaction based on interviews and satisfaction surveys with participants.

Additionally, to ensure quality, the developer conducts tests for each function of ONE AI, verifies functionality with Japanese users, and aims to secure a response accuracy of over 90% and reduce response time by over 20% compared to comparable GPU equipment. Specific AI technology verification items include ▲ AI service accuracy such as voucher classification, document summarization, and meeting minutes analysis ▲ measurement of NPU computation speed based on RNGD ▲ comparison of computation per power consumption with existing GPUs ▲ system stability analysis such as service interruption and error occurrence rate ▲ user satisfaction evaluation on UX, ease of use, and function quality.

Significance Beyond AI Semiconductor Demonstration in Japan Entry Collaboration Case


The AI Semiconductor Overseas Demonstration Project is an important project for both AI solution companies and AI semiconductor companies. Semiconductor companies participating in this project include FuriosaAI, Rebellion, Mobilint, DeepX, Dinoticia, UX Factory, and Ditec. While some companies have recorded profits, none have achieved dramatic profits through large-scale projects. Through this demonstration project, they must prove the success potential of AI semiconductors in the overseas market to lead to new business opportunities.

The AI Semiconductor Overseas Demonstration Project is more about promoting the industrial value of Korean semiconductors than generating revenue from the project itself. It is hoped that the contracts and achievements obtained through the project will serve as a catalyst for the overseas expansion of Korea's AI semiconductor industry in the future.

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

Popular News