IN Brief:
- South Korea has established a roadmap covering eight application-specific AI chips and two semiconductor IP projects by 2030.
- The KRW 800.2bn programme involves equipment manufacturers, fabless designers, foundries, and software developers.
- Samsung Foundry plans to prioritise 21 prototype manufacturing runs during 2027 across processes ranging from 2 nm to 28 nm.
South Korea’s Ministry of Trade, Industry and Resources has established a development roadmap for eight application-specific AI chips and two semiconductor intellectual property projects, with the ten developments scheduled for completion and verification by 2030.
The roadmap forms part of the KRW 800.2bn K-On-Device AI Semiconductor Technology Development Project, which began in June 2026. Its latest programme arrangements were detailed on 12 November at the second-half meeting of the AI Semiconductor M.AX Alliance in Seoul.
The initiative combines semiconductor designers, foundries, software companies, and industrial manufacturers. Its objective is to develop computing hardware that can be integrated into actual products, supported by the necessary modules, operating software, and application testing.
Eight chips are planned across six application groups. The programme targets one AI chip for autonomous vehicles, three for smart appliances, and one each for collaborative robots, humanoid robots, agricultural robots, and unmanned drones.
Two additional projects concern reusable semiconductor design IP. One is intended for advanced manufacturing processes, while the other addresses operation in extreme environments. These are separate design assets rather than additional completed AI processors.
The distinction is relevant because the programme combines complete semiconductor products with technology intended to support subsequent designs. Its target of ten developments does not mean ten independently manufactured AI chips.
Application-specific AI hardware is increasingly being considered for equipment that must interpret information and make decisions locally. Processing sensor data within a machine can reduce dependence on external computing infrastructure and the communication bandwidth required to transmit information to a remote service.
Local execution can also reduce delays associated with network transfer, although the benefit depends on the workload and the available processing hardware. Power consumption, memory capacity, software performance, and thermal conditions remain important constraints.
Automotive AI processing can involve interpreting information from cameras and other sensors under demanding timing and reliability requirements. The resulting hardware must provide appropriate computing performance while supporting the wider vehicle system’s operational and safety architecture.
Smart appliances often impose different requirements. Their designs may need to accommodate AI functionality within comparatively restricted power budgets and compact electronics, making the balance between processing capability, memory, and manufacturing cost important.
Robotics presents further variations. Collaborative and humanoid robots combine sensing with motion and control functions, while agricultural machines may operate under changing environmental conditions and have limited access to communications infrastructure.
Unmanned drones add constraints involving electrical power, system weight, and the processing of sensor information during flight. These differences explain why the roadmap specifies separate semiconductor developments rather than assuming a single processor will meet every application’s requirements.
Industrial equipment manufacturers will participate directly in defining the specifications for the chips. Rather than purchasing completed devices after development, participating companies are expected to supply operating requirements and provide access to real equipment for prototype evaluation.
Testing a processor within its intended application can reveal constraints that are difficult to identify using isolated semiconductor benchmarks. Sensor interfaces, software compatibility, power management, thermal behaviour, and interactions with other control electronics can affect the performance of the complete machine.
Providing those requirements before semiconductor design is finalised may help reduce the risk of producing technically capable chips that do not satisfy the integration needs of the equipment manufacturer.
Manufacturing access represents another part of the programme. Samsung Foundry plans to coordinate priority allocation for 21 prototype fabrication runs during 2027, covering semiconductor processes from 2 nm to 28 nm.
The provisional allocation consists of three runs at 2 nm, four at 4 nm, three at 5 nm, three at 8 nm, three at 14 nm, and five at 28 nm. Samsung intends to consider these requirements alongside its existing production schedule when allocating capacity.
The 21 runs represent opportunities for prototype fabrication, not 21 separate chip products or guaranteed commercial manufacturing contracts. The number of completed devices will depend on the designs submitted, their fabrication arrangements, and the results of subsequent testing.
Some prototype activity will use multi-project wafer manufacturing, in which different semiconductor designs share a fabrication run. This allows several projects to distribute certain manufacturing costs, reducing the expenditure associated with obtaining initial silicon.
Dedicated runs provide a different manufacturing option for designs requiring their own wafer allocation. Their cost and suitability depend on the selected process, design complexity, prototype quantity, and development timetable.
Access to fabrication capacity can become a bottleneck for smaller semiconductor companies. Once a design has completed verification and physical implementation, engineers must secure the appropriate foundry process and manufacturing schedule before working silicon becomes available for testing.
Advanced process nodes also introduce demanding design rules and substantial engineering costs. Semiconductor designers must verify timing, electrical behaviour, manufacturability, and the interactions between functional blocks before committing the design to fabrication.
Reusable IP can reduce some of that development effort by providing established functions that do not need to be recreated for each chip. However, the blocks must still be integrated and verified against the electrical and physical requirements of the final design.
The programme includes support from domestic and international semiconductor IP companies. Proposed measures include more flexible payment arrangements and shared IP packages intended to reduce development costs for fabless designers.
A new public-private strategy committee will oversee the development programme. It is co-chaired by a government vice-minister and Kyung Hee University President Kim Jin-sang, with responsibility for reviewing annual plans, monitoring delivery, and addressing coordination problems between participating organisations.
The ministry has already brought forward the planned fabrication of one smart appliance AI chip by more than a year. The adjustment concerns a specific development milestone; it does not establish an accelerated commercial launch timetable for the entire portfolio.
Completing the roadmap will require working semiconductor prototypes, suitable software, and verification within the relevant industrial applications. Production qualification and commercial adoption will follow separate decisions by the participating manufacturers.
Samsung’s planned foundry allocation provides an early manufacturing framework for the programme. Its effectiveness will depend on whether semiconductor designers can turn those prototype opportunities into functional devices that meet the requirements of the equipment in which they are intended to operate.



