Samsung Confirms 1.4 Nanometer Mass Manufacturing Timeline for 2029 While Expanding Next Generation 2 Nanometer Node Roadmap and AI Partner Ecosystem
Samsung Electronics has confirmed its timescale for mass manufacturing the 1.4 nanometre silicon node in 2029. The revision was confirmed by Shin Jong shin, executive vice president of design platform development at Samsung, during the Samsung Advanced Foundry Ecosystem forum held at the Seocho headquarters according to The Elec. This provides confirmation to the amended route that Samsung announced after revising its initial schedule to follow the 1.4 nanometre node to 2027.
The next 1.4 nanometre node, known as SF1.4, is scheduled to stay on track for 2029 targeted production with early lead customers. A revised derivative node, SF1.4 plus, is expected to commence volume manufacturing in 2030. These targets place Samsung slightly behind TSMC and fairly even with Intel for these two nodes. TSMC is planning to release its A14 node in 2028, which sits in the 1.4 nanometre category. Similarly, Intel is laying a risk production node for 14A in 2028 with full volume production expected in 2029.
As lithography scaling becomes more advanced, Samsung is relying increasingly on design technology co optimization for incremental gains. Shin Jong shin stated, "Plus nodes enable customers to ramp up yield and performance and power and area metrics with the chips without having to change their intellectual property designs."
In a specific example of moving from the SF2 node to the SF2P node, power requirements were reduced by 26% and the clock frequency increased by 15%. Notably, more than 50% of those improvements came directly from design technology co optimization rather than simply scaling transistors.
In an effort to counter the bandwidth constraints inherent in artificial intelligence hardware, the Korean titan is increasing its on chip SRAM integration. The NVIDIA Rubin graphics processing unit features close to 128MB of SRAM on each die. Furthermore, a new language processing unit produced on a 4 nanometre node from Samsung manages over 500MB of integrated SRAM. This high level of memory integration effectively minimizes data latency by locating essential data proximally to execution logic without the electrical refreshing required in traditional DRAM.
Samsung further presented a roadmap for its next generation 2 nanometre platform, which will evolve sequentially from SF2 to SF2P, SF2P plus, and eventually to SF2X. The SF2P plus node is expected to move to volume production in 2027 or 2028, followed by the high performance computing and artificial intelligence capable SF2X node. The foundry aims to maintain maximum IP compatibility to allow the reuse of existing design objects to keep costs lower.
This roadmap is already under development by several industry partners. AD Technology is designing an artificial intelligence infrastructure platform based on the Samsung 2 nanometre node, while Gaonchips works on 2 nanometre processing. Counde is developing a platform for high bandwidth memory and high speed interfaces in the packaging area. Coasia Nexell is focusing on 3D integrated circuits logic design and SemiFive is working on the 3D execution system. Additionally, Rebellion presented its Rebel 100 artificial intelligence accelerator produced on the Samsung 4 nanometre line.
