Samsung Introduces Future Semiconductor Node Plans
Introduction to Semiconductor Plans
At the SAFE Forum (Samsung Advanced Foundry Ecosystem), Samsung shared its ambitious vision for future semiconductor technology. One of the main highlights was the introduction of their first 2nm smartphone chip, the Exynos 2600, produced using the SF2 node. This chip marks a significant step forward in mobile device efficiency and performance.
Innovation Through the SF2P Node
Next, Samsung plans to release the SF2P node, which is expected to reduce power consumption by up to 26% and increase clock speed by 15%. However, these improvements do not rely solely on the new node technology. Shin Jong-shin, Vice President of the Design Platform Development Team at Samsung Foundry, explained that more than half of the performance gains come from the Design-Technology Co-Optimization (DTCO) approach.
DTCO is a new method in chip manufacturing that combines semiconductor manufacturing processes and chip design. Typically, these two steps are done separately, but by integrating them, a more optimal design can be achieved. The result is increased speed, reduced power consumption, and even potential cost savings.
Next Steps: SF2P+ and SF2X
Following SF2P, Samsung also introduced SF2P+, which is expected to begin mass production between 2027 and 2028. Additionally, there is development of the SF2X node, designed specifically for artificial intelligence (AI) hardware. With a focus on AI, Samsung aims to stay at the forefront of technological innovation, considering the growing need for fast and efficient data processing.
SRAM Performance Improvements for Future Chips
Besides producing smaller and more efficient chips, Samsung is also working to increase SRAM (Static Random-Access Memory) capacity on chips. SRAM is used to build CPU or GPU registers and caches. Although very fast, SRAM requires more space compared to DRAM. For example, an SRAM cell storing one bit of information requires six transistors, while a DRAM cell only needs one transistor.
For instance, Nvidia's Rubin GPU produced on TSMC's N3 node contains 128MB of SRAM on the chip. Samsung and Groq are developing an LLM accelerator that will have more than 500MB of SRAM, and this is being done on an older 4nm node. Additionally, Rebellions, supported by Samsung, has introduced the AI accelerator REBEL-100, which is currently being produced on Samsung's 4nm node.
The Future of the 1.4nm Node
After the 2nm era, Samsung will continue development toward the 1.4nm node, starting with SF1.4, scheduled for mass production in 2029. This plan is followed by SF1.4+, expected in 2030. Although SF1.4 was initially planned for launch in 2027, the schedule was postponed a few years ago, reflecting Samsung's commitment to quality and innovation at every step.
With all these innovative plans, Samsung is not only striving to be a leader in the semiconductor industry but also committed to delivering more efficient and beneficial technology to users worldwide.
