"From the 2-Nanometer Process to SSDs and GPUs," Samsung Electronics Bets on Semiconductor Competitiveness, With the Core Imperative Being "Organic Growth"

Samsung Electronics’ 2-nanometer pilot production nears, ushering in full-scale competition at the leading edge of foundry manufacturing
Samsung Electronics parts ways with Arm, aggressively embraces the open-source RISC-V architecture
"Aiming to end its reliance on AMD by 2027," the company accelerates its drive toward GPU design independence

Samsung Electronics’ pilot production of its 2nm (nanometer, one-billionth of a meter) foundry process is fast approaching. With both the design IP needed to push beyond the limitations of the 2-nanometer process and customer demand now in place, the company has effectively fired the starting gun in the race for next-generation process leadership. Beyond that, Samsung Electronics has also begun pursuing design independence in solid-state drives (SSDs) and graphics processing units (GPUs), concentrating its efforts on bolstering its fundamental semiconductor competitiveness.

Samsung Electronics Foundry Steps Up the 2-Nanometer Race

On April 5, IT outlet Wccftech reported that Samsung Electronics’ yield rate for its 2-nanometer gate-all-around (GAA) process had reached roughly 60%, with preparations for pilot production now effectively complete. According to the report, Samsung Electronics is currently converting its Taylor, Texas plant in the United States, originally built for its existing 4-nanometer process, into a production base for 2-nanometer manufacturing. Some facilities have already received temporary certificates of occupancy (TCO), enabling the deployment of engineers and the installation of equipment, while Dutch semiconductor equipment maker ASML has reportedly dispatched personnel to the site to support the installation of extreme ultraviolet (EUV) systems.

The company has also secured IP aimed at addressing the heat and area-efficiency constraints widely seen as the principal limitations of the 2-nanometer process. As semiconductor nodes become more advanced, heat density inside chips rises, while leakage current increases exponentially, eroding power efficiency. Until now, fabless chipmakers had installed temperature sensors in the FEoL (Front-End-of-Line) region at the bottom of the chip to monitor internal heat, but that method carried a critical drawback: it encroached on the space reserved for actual computing elements such as transistors, which serve as the chip’s functional core. In response, Samsung Electronics developed a technology that relocates the temperature sensor from the conventional FEoL position to the upper interconnect layer, or BEoL, thereby reducing wasted space inside the chip. Industry observers believe Samsung Electronics is likely to apply the technology aggressively to high-performance chips such as its in-house Exynos mobile application processor series going forward.

The Exynos lineup is expected to be one of the flagship products manufactured on Samsung Electronics’ 2-nanometer foundry process. Samsung Electronics is likely to equip the standard and Plus models of the Galaxy S26, as well as this year’s new Galaxy Z Flip release, with the 2-nanometer-based Exynos 2600. In addition, Tesla’s artificial intelligence (AI) chips, AI5 and AI6, are slated for mass production on the 2-nanometer process, while a substantial portion of Qualcomm’s new AP orders is also likely to be manufactured on the same node. U.S. autonomous-driving AI semiconductor firm Ambarella has likewise placed an order with Samsung Electronics for 2-nanometer chips for advanced driver-assistance systems (ADAS).

Breaking from Arm and Adopting an Open-Source Architecture

Beyond the 2-nanometer leading-edge process, Samsung Electronics is also devoting considerable effort to securing “in-house competitiveness” across its semiconductor business. Its move to distance itself from the U.K.’s Arm is a prime example. Arm is the dominant force in chip design IP, holding particularly unrivaled standing in CPU architecture, or instruction sets. Until now, Samsung Electronics had depended directly on Arm’s architecture and core design IP not only for the CPU cores in its mobile APs, but also for SSD controllers and embedded processors used in a range of IoT and modem chips.

Recently, however, the ties between the two companies have been loosening rapidly. On April 4, Wccftech reported that Samsung Electronics would apply the open-source RISC-V architecture to the controller chip that serves as the brain of its next-generation enterprise SSD lineup, the BM9K1. In effect, the company has declared a forceful drive toward “design independence” by reducing its technological reliance on Arm. To date, Samsung Electronics has never fully introduced RISC-V into an actual mass-produced product beyond the prototype stage.

The principal reason behind Samsung Electronics’ adoption of RISC-V is widely seen as the substantial licensing cost involved. Companies using Arm designs must pay royalties proportional to chip shipment volumes. In the SSD market, where large-volume deliveries are standard, that structure has weighed on profitability. RISC-V, by contrast, is an open-source standard that anyone can use free of charge. That is why the market views Samsung Electronics’ latest shift as a strategic move aimed at cutting royalty expenses and securing capital for next-generation semiconductor R&D.

Limited design flexibility has also emerged as a key weakness of Arm-based architecture. Arm strictly restricts customer-side design changes, treating such modifications as licensing violations. RISC-V architecture, by contrast, allows users to add whatever functions they need. Samsung Electronics, in fact, enhanced the performance of its new product by incorporating proprietary extensions into RISC-V. The company independently developed a dedicated controller capable of finely tuning functions such as NAND cell management, error correction coding (ECC), and the irregular read-write patterns characteristic of AI workloads. As a result, the product’s sequential read speed improved by 1.6 times from the previous-generation BM9C1, while energy efficiency increased by 23%.

The Blueprint for GPU Self-Reliance Also Takes Shape

The push for independence is also continuing in the GPU segment. Samsung Electronics formed a partnership with U.S. fabless chipmaker AMD in 2019 and has used GPUs based on AMD’s RDNA architecture in the Exynos series since 2022. That strategy was designed to offset its relative weakness in graphics performance. The Exynos 2500 and Exynos 2600, both released this year, also use GPUs incorporating AMD technology. Samsung Electronics, however, plans to use an internally developed GPU beginning with the tentatively named Exynos 2800, which is scheduled for mass production in 2027. That chip is set to serve as the brain of the Galaxy S28 series, due for release in 2028.

To secure self-reliance in the GPU field, Samsung Electronics has spent the past three years recruiting large numbers of GPU engineers from global semiconductor companies. The company is reportedly offering annual salaries ranging from $199,000 to $266,000 to senior-level engineers in order to secure talent. A particularly notable figure is John Rayfield, whom Samsung recruited last year. He is a heavyweight in the GPU field who has led mobile and AI chip design efforts at AMD, Broadcom, and Intel. With Rayfield at the forefront, Samsung Electronics is accelerating development of its own architecture through its research centers in Austin and San Jose (SARC/ACL).

GPU independence is a long-term strategic move aimed at securing competitiveness in advanced industries, extending well beyond the optimization of smartphone performance alone. Once it secures its own GPU architecture, Samsung Electronics will be able to respond more flexibly across a broad range of platforms, including autonomous vehicles, humanoid robots, smart glasses, and other extended reality (XR) devices. It would also establish the technological foundation needed to compete with Broadcom and Marvell in the market for custom AI semiconductors (ASICs) for big tech clients such as Google, Microsoft (MS), and OpenAI. At present, however, only a handful of companies worldwide, including Nvidia, AMD, Qualcomm, Apple, and Arm, have successfully commercialized independent GPU architectures. In that regard, one industry expert noted, “Samsung Electronics’ GPU internalization is an essential step for future growth,” while adding, “The key question is whether Samsung Electronics can overtake, within a short period of time, the patent barriers and software ecosystem that have been built up over decades.”