The list reads like a specification sheet for a small industrial economy. One AI chip for autonomous driving, three for smart home appliances, and one each for collaborative robots, humanoids, agricultural robots and unmanned drones, plus two design assets, one for advanced processes and one for extreme environments.
Ten items in total, all to be developed and verified in Korean products by 2030. The Ministry of Trade, Industry and Energy confirmed the roadmap on Nov. 12 at the second-half general meeting of the AI Semiconductor M.AX Alliance in Seoul, before an audience of about 150 people drawn from demand companies, fabless design firms, foundries and intellectual-property vendors.
The roadmap sets the schedule for an 800.2 billion won project that began in June. The structure is unusual. Instead of paying a chip designer to build something and then looking for a buyer, the ministry has made the buyers the lead organisations. Appliance makers, carmakers, robot manufacturers and defence contractors are expected to specify what they need for mass production and to open their test lines for demonstration.
That arrangement addresses the oldest problem in Korean semiconductors. The country dominates memory, where Samsung Electronics and SK Hynix sell standard parts in enormous volumes to customers who have little choice. It has never built a comparable position in logic chips, where the value comes from designing a part for a particular product and holding that socket for a hardware generation.
Samsung’s foundry division is the second half of the bargain. The company will set aside 21 prototype production runs for the programme: three at 2-nanometre class, four at 4 nanometres, three at 5, three at 8, three at 14 and five at 28, mixing multi-project wafers with dedicated runs. It will collect 2027 prototype demand in advance and reflect it in process allocation, so that a domestic fabless firm does not lose six months waiting behind a larger customer.
The ministry also said it had pulled forward the schedule. Production timing for one of the home-appliance chips moves up by more than a year from the original plan.
Oversight sits with a joint public-private committee co-chaired by the ministry’s vice minister and Jin-sang Kim, president of Kyung Hee University. The committee is to review and confirm annual development plans each year, check progress, and settle technical disputes and conflicts of interest among the companies involved.
The money is modest by the standards of the industry. At current exchange rates the project is worth a few hundred million dollars spread across five years and ten designs, a fraction of what a single advanced logic fab costs. Samsung and SK Hynix each spend more than that on memory capacity in a good quarter.
That is roughly the point. The programme is not trying to build a national champion in data-centre accelerators, where Nvidia sets the price and the design cost runs into the billions. It is aiming at inference on the device: the chip inside a robot arm that has to recognise a part in 20 milliseconds, or inside a washing machine that has to weigh a load without a network connection.
Those markets are fragmented, cost-sensitive and slow to standardise, which is why they have attracted dozens of small designers rather than a single dominant supplier. Chinese and Taiwanese chipmakers are already selling into them. A Korean entrant needs a cost advantage, a software stack that customers can integrate, and a foundry relationship that guarantees supply when volumes are small.
The roadmap tries to supply all three at once, using the country’s appliance and automotive industries as the anchor demand. If Samsung Electronics puts a domestic AI chip in a refrigerator line that ships millions of units a year, the fabless firm that designed it gets volume without having to win a merchant market first.
The risk is that anchor demand never materialises. Procurement departments at large manufacturers buy on price and reliability, and a government programme cannot order them to prefer a domestic part. The ministry’s own framing acknowledges this: demand companies are described as innovative buyers participating in development, rather than as guaranteed customers.
A second risk is foundry priority. Two-nanometre capacity is the most contested resource in the industry, and Samsung’s own high-bandwidth memory and AI accelerator customers will be competing for the same lines. A promise to prioritise domestic prototypes is a promise that gets tested the first time a large customer calls.
Min-woo Lee, who heads the ministry’s Industry Growth Office, put the test plainly. Real partnership in semiconductors, he said, should be proven not by an agreement to cooperate but by jointly developed technology being adopted in actual products.
That is the right measure and the harder one. Korea has run semiconductor support programmes before, and the scoreboard has usually been counted in patents and prototypes rather than sockets designed into shipping products. Ten chips by 2030 is a target that can be verified. Whether a Korean robot or a Korean car in 2031 contains one of those chips is the question the roadmap cannot answer in advance.
