Samsung Electronics reaffirmed its plan to begin mass production of its 1.4-nanometer (nm) process in 2029, confirming the revised roadmap it announced last year.
Shin Jong-shin, Vice President of the Design Platform Development Team at Samsung Foundry, said at the SAFE (Samsung Advanced Foundry Ecosystem) Forum held at Samsung Electronics' Seocho office in Seoul on July 1 that the SF1.4 process is progressing smoothly toward volume production in 2029 for lead customers.
"SF1.4 is under steady development with the goal of entering mass production in 2029 for lead customers," Shin said. "SF1.4+, an enhanced version based on SF1.4, is scheduled for mass production in 2030."
The SAFE Forum is Samsung's annual event for foundry ecosystem partners, where the company shares its latest process roadmap and design ecosystem strategy.
Samsung announced at the Samsung Foundry Forum in 2022 that it aimed to begin mass production of its 2nm process in 2025 and its 1.4nm process in 2027. The company said in 2024 that development of the 1.4nm node remained on schedule, but revised the target to 2029 last year.
Samsung's 2029 target for 1.4nm production lags Taiwan Semiconductor Manufacturing Co. (TSMC), which plans to mass-produce its A14-class process in 2028, but is broadly in line with Intel Corp. (Intel) roadmap. Intel plans to begin risk production of its 14A process in 2028 before moving to full-scale manufacturing in 2029.
SF is Samsung Foundry's process naming convention, while the number indicates the process node. The plus (+) designation refers to an enhanced version. Shin said the Plus nodes improve yield and power, performance and area (PPA) by preserving existing intellectual property (IP) infrastructure while applying design-technology co-optimization (DTCO) and additional process enhancements.
Samsung said DTCO will become increasingly important as process nodes continue to shrink because more complex interconnect structures and tighter design constraints require simultaneous optimization of manufacturing technology and chip design.
"When transitioning from SF2 to SF2P, power consumption improved by 26% and operating frequency increased by 15%, with more than half of those gains attributable to DTCO," Shin said.
Samsung also outlined its strategy to expand on-chip static random-access memory (SRAM) to strengthen competitiveness in AI semiconductors. Shin said Nvidia Corp. (Nvidia) Rubin graphics processing unit (GPU) integrates about 128 megabytes (MB) of SRAM per die, while a recently produced language processing unit (LPU) manufactured on Samsung's 4nm process incorporates more than 500MB of SRAM.
"This trend will accelerate further going forward," he said. The 4nm LPU referenced by Shin is believed to be a processor developed by Groq.
SRAM is a memory technology that retains data as long as power is supplied and, unlike DRAM, does not require periodic refresh operations, enabling faster performance. Integrating larger amounts of SRAM on a chip reduces data movement, improving AI computing performance and power efficiency.
Samsung also unveiled its next-generation 2nm roadmap. The platform will evolve through SF2, SF2P, SF2P+ and SF2X. SF2P+ is targeted for mass production between 2027 and 2028, followed by SF2X, a process optimized for artificial intelligence (AI) and high-performance computing (HPC) applications. The company said it will maintain maximum compatibility with existing IP across future process generations, allowing customers to continue leveraging previous design assets.
Samsung also showcased collaboration with domestic design solution partners (DSPs). ADTechnology Co., Ltd. (ADTechnology) said it is preparing an AI infrastructure platform based on Samsung's 2nm process. Gaonchips Co., Ltd. (Gaonchips) said it is working on AI and HPC projects using the 2nm node. CoAsia Nexell Co., Ltd. (CoAsia Nexell) introduced an advanced packaging platform targeting high-bandwidth memory (HBM) and high-speed interfaces, while Semifive Inc. (Semifive) said it is developing a 3D IC-based logic and memory integration solution. Rebellions Inc. (Rebellions) presented its AI accelerator, REBEL-100, which is currently being mass-produced on Samsung's 4nm process, saying Samsung Foundry's manufacturing technology and design ecosystem played a key role in the chip's development and commercialization.
