Samsung held an industry Foundry event focused on solutions for “Empowering the AI Revolution,” where they announced new process nodes and technology developments. The company unveiled its technology roadmaps going out to 2027. The event included the announcement of two new process nodes, the 2nm SF2Z and SF4U nodes. The SF2Z node includes a backside power delivery network (BSPDN) to enhance power, performance, and area (PPA) compared to the first-generation 2nm node, as well as reduce IR drop. Mass production of this node is planned for 2027, while the SF4U node offers PPA improvements with mass production set for 2025. Samsung also plans to introduce the SF1.4 (1.4nm) products for mass production in 2027.
Samsung also revealed its Samsung AI foundry solutions, which include gate all around (GAA) CMOS processes providing low power AI chips. The company plans to introduce co-packaged optics (CPO) in chiplet architectures by 2027 to deliver even higher performance. Samsung reported an 80% increase in sales for its Samsung foundry AI business over the past year. The company offers a range of products such as specialty and 8-inch wafer products to meet customer needs in various industries including automotive, medical, wearable, and IoT applications. Samsung has been producing GAA CMOS for three years and will include this technology in its upcoming 2nm process.
At the event, Samsung introduced its Multi-Die Integration Alliance, which focuses on chiplet integration. This alliance allows for the combination of single-purpose chips in a high-performance package, enabling the integration of chiplets from various vendors into a cohesive system. Samsung stated that this alliance will include a focus on 2.5D and 3D IC designs. The company is also a part of the UCIe chiplet architecture introduced in 2022, based on the CXL architecture for memory expansion and pooling.
Aside from its foundry business, Samsung leads in the DRAM and NAND flash market. The company introduced industry-first CXL memory and LPCAMM2 DRAM in 2022, as well as the first GDDR7 DRAM in 2023. DRAM, especially high bandwidth memory (HBM) modules, play a crucial role in AI processing packages, such as those created by NVIDIA. Samsung showcased HBM packages in its announcements. Furthermore, Samsung and other foundries are offering non-volatile memory options to replace NOR flash at advanced manufacturing nodes and higher-level cache SRAM for storing AI inference models.
The chiplet architectures and 3D heterogeneous integration from Samsung will incorporate a wider range of memory and other components for specialized applications. At the Samsung Forum event and subsequent workshops, the company expanded its 2nm and 4nm product lines, announced chiplet integration with co-packaged optics by 2027, and reported an 80% increase in its AI foundry business over the past year. Samsung continues to innovate in the semiconductor industry, paving the way for advancements in AI technology and pushing the boundaries of what is possible in chip design and integration.
