SUWON, South Korea — In Korean chipmaking, the pay stub has become an organizing principle.
Samsung Electronics Co. has created a “Core Team” inside its Semiconductor Research Institute, the main research body of its Device Solutions division, to develop next-generation DRAM and NAND flash products. The unit has recently begun operating, according to industry sources cited by Yonhap and other Korean outlets on Wednesday.
What sets the team apart is not its mandate but its compensation. Members keep their affiliation with the research institute. If they meet development targets set in advance, they will be paid performance bonuses on the same scale as employees of the Memory Business, the division that sells the chips and books the profit.
That difference is worth roughly 30 percentage points. Under the 2026 wage agreement reached between Samsung’s management and its union in May, the DS division pays a special management performance bonus funded by 10.5% of its operating profit. Staff at the research institute were classified as a “common organization” and receive about 70% of what memory division employees get. Core Team members move to 100%.
The arrangement addresses a complaint that has circulated inside Samsung for years: that the engineers who design the next memory chip are paid as overhead, while the colleagues who sell the current one collect the upside. Korean chipmakers pay a large share of white-collar compensation through a profit-linked bonus once a year, which makes the unit on a badge as consequential as the job title on it.
The instrument matters because it is the one Korean chipmakers can adjust quickly. Base salaries are set through bargaining and move slowly; the annual bonus is where a good year and a bad year are expressed. That makes it the obvious place to pay for a specific outcome, and the obvious place for staff to look when they compare offers.
Samsung is not the first company to discover that research staff notice. SK Hynix, which took the lead in high-bandwidth memory and supplies Nvidia, has used its run of record profits to hold onto process and design engineers. American chipmakers and software companies have hired from Korean memory teams for the same reason, and the flow has been steady enough that Korean newspapers treat each departure as a data point.
The technical pressure behind the reorganization is straightforward. Memory is in the middle of an upcycle driven by AI servers, with DRAM contract prices rising sharply and NAND recovering from the collapse of 2022 and 2023. The products that carry those prices, high-bandwidth memory stacked with logic dies and NAND with more than 300 layers, are hard to design and harder to manufacture at yield.
Samsung has ground to make up. It has spent two years working to qualify its high-bandwidth memory for Nvidia’s accelerators while SK Hynix shipped, and its foundry business has struggled to hold large customers. In commodity DRAM and NAND its position is stronger, but Chinese suppliers including CXMT and YMTC are adding capacity and moving up the product stack, which compresses the part of the market where Samsung has historically earned its steadiest margins.
The Core Team is designed to shorten the distance between research and product. Korean reports describe the unit as combining technology development with process design and working alongside the Memory Business rather than at a remove from it, which is a structural change as much as a pay change. In Samsung’s system, the research institute has often been measured on papers and patents; the memory division is measured on yields and shipment dates.
Paying researchers on the memory scale imports the second set of measures into the first organization. It also creates a two-tier workforce inside a single institute, and the tier is decided by assignment rather than seniority. Samsung has not said how many people are on the team or how it will define the development targets that trigger the higher payout, and those two details will determine whether the unit functions as an incentive or as a retention tool for a select group.
The precedent matters beyond Samsung. Memory and logic are both consolidating around a small number of companies with the capital to keep building fabs, and the constraint on those companies is increasingly people rather than equipment. A process engineer who understands why a stack of a few hundred layers warps is not interchangeable, and no amount of capital expenditure produces one on schedule.
Samsung’s experiment is an attempt to price that scarcity from the inside, by treating compensation as part of the development schedule instead of a consequence of it. If it works, the structure will spread to other units and other companies, and the annual wage negotiation will become the place where research priorities are set in public.
If it does not, the company will have raised expectations for a group of engineers, given them a benchmark to compare against, and left the rest of the institute on the old scale. That is a harder position to defend than the one it started from. The first test arrives at the end of the year, when Samsung pays its bonuses and its researchers find out what a target is worth.
