Samsung Electronics has revealed its foundry innovations for boosting its AI business during Samsung Foundry Forum (SFF) U.S., an annual event held at the company’s Device Solutions America headquarters in San Jose, California.
Samsung announced its process technology roadmap, including two new nodes — SF2Z and SF4U — as well as its integrated Samsung AI Solutions platform using the strengths of its Foundry, Memory and Advanced Package (AVP) businesses.
Samsung plans to offer a one-stop shop for clients to get their AI chips made faster, integrating its global No. 1 memory chip, foundry, and chip packaging services, to utilize the AI boom.
Samsung is one of the few companies that sells memory chips, offers foundry services and designs chips under the same roof. As a result, Samsung said it has cut the time it takes to produce AI chips by 20 percent.
“Alongside our GAA process optimized for AI chips, we plan to introduce integrated, co-packaged optics (CPO) technology for high-speed, low-power data processing, providing our customers with the one-stop AI solutions they need to thrive in this transformative era,” said Dr. Siyoung Choi, President and Head of Foundry Business at Samsung Electronics.
Samsung plans to introduce the all-in-one, CPO-integrated AI solution in 2027, provide customers with one-stop AI solutions.
Samsung expects global chip industry revenue to grow to $778 billion by 2028, boosted by AI chips, Siyoung Choi said.
Revenue from artificial intelligence (AI) semiconductors is expected to total $71 billion in 2024, the latest forecast by Gartner indicated.
But there are challenges in AI chipset market. While much of the focus is on the use of high-performance graphics processing units (GPUs) for new AI workloads, the major hyperscalers (AWS, Google, Meta and Microsoft) are all investing in developing their own chips optimized for AI, Gartner said.
Samsung SF2Z and SF4U
The event featured presentations from thought leaders such as Arm CEO Rene Haas and Groq CEO Jonathan Ross. Around 30 partner companies exhibited at booths.
Samsung’s latest 2nm process, SF2Z, incorporates optimized backside power delivery network (BSPDN) technology, which places power rails on the backside of the wafer to eliminate bottlenecks between the power and signal lines.
Applying BSPDN technology to SF2Z enhances power, performance and area (PPA) compared to SF2. Samsung said it also significantly reduces voltage drop (IR drop), enhancing the performance of HPC designs. Mass production of SF2Z is slated for 2027.
SF4U, a 4nm variant, offers PPA improvements by incorporating optical shrink. Samsung said mass production of SF4U is scheduled for 2025.
Samsung said its preparations for SF1.4 (1.4nm) are progressing smoothly, with performance and yield targets on track for mass production in 2027.
Samsung emphasized the maturity of its GAA technology, a key technological enabler in empowering AI.
Samsung’s GAA process is demonstrating maturity in both yield and performance. Samsung plans to mass produce its second-generation 3nm process (SF3) in the second half of this year and deliver GAA on its upcoming 2nm process.
Over the past year, customer collaborations have led to an increase in Samsung Foundry’s AI sales by 80 percent.
Samsung offers specialty and 8-inch wafer derivatives with PPA improvements and cost competitiveness.
At a briefing with reporters ahead of the event, Executive Vice President of Foundry Sales and Marketing Marco Chisari said the company believes OpenAI CEO Sam Altman’s projections on demand for AI chips are realistic.
Altman has told executives at contract chipmaker TSMC, he wanted to build roughly three dozen new chip factories, Reuters has previously reported.
Baburajan Kizhakedath
