Promising to lead the AI semiconductor revolution, Samsung Electronics emphasized its exclusive “one-stop” service for foundry, memory, and advanced packaging at the Foundry Forum 2024 on June 12. In San Jose, President Choi Si-young highlighted how this comprehensive approach sets Samsung apart in the global market.
The Samsung Foundry Forum is an annual event where Samsung Electronics unveils its semiconductor foundry strategies. With TSMC, the world’s leading foundry company, securing orders for NVIDIA’s AI semiconductors, attention has turned to the strategies of Samsung Electronics, the second-largest foundry.
Being an integrated device manufacturer (IDM), handling everything from semiconductor design to manufacturing and sales, has been seen as a disadvantage for Samsung’s foundry business due to concerns about potential information leaks, making it difficult to secure orders from competitors like Apple and MediaTek. However, with the recent shortage of AI semiconductors, the ability to reduce manufacturing time through one-stop services has become an advantage. Song Tae-joong, vice president of Samsung’s foundry business, stated in a media briefing that the one-stop AI solution can shorten production time by more than 20% compared to when clients use separate foundry, memory, and packaging companies.
Rene Haas, CEO of ARM, a global semiconductor design company, also expressed his expectation that Samsung could solve the issue of semiconductor manufacturing taking over two years from IP provision to actual production.
At the event, Samsung focused on applying new technologies that reduce power consumption and improve performance in their manufacturing processes, rather than competing in the “nano war” of extreme miniaturization. Their strategy aims to attract foundry customers by addressing the significant power consumption issues hindering AI semiconductor growth. Samsung announced plans to introduce the new 2nm process SF2Z, featuring backside power delivery network (BSPDN) technology by 2027. This technology places the power wiring layer on the backside of the wafer, reducing power consumption and enhancing data signal efficiency. Additionally, from 2027, they plan to implement silicon photonics technology, which uses photons to transmit signals instead of electrons, increasing processing speed while reducing power consumption and heat, a key technology for AI semiconductor development.
Choi stated, “At a time when numerous technologies are evolving around AI, the key to its implementation lies in high-performance, low-power semiconductors.” The semiconductor industry noted that Samsung is focusing on actual process improvements rather than the extreme miniaturization race.
Samsung also announced the introduction of Custom HBM, a high-bandwidth memory tailored to customer needs, starting with the next-generation HBM4. As AI semiconductor designs from companies like NVIDIA, AMD, and Google differ, Samsung aims to design the most efficient HBM for each case. SK Hynix has also partnered with TSMC to produce customized HBM, indicating intense future competition in this market between Samsung and SK Hynix.
Whether the newly announced foundry strategy will serve as a turning point for Samsung’s recently sluggish semiconductor business remains to be seen. According to market research firm TrendForce, Samsung’s global foundry market share was 11% in the first quarter of this year, trailing TSMC’s 61.7% by 50.7 percentage points, a slight increase from the previous quarter’s 49.9 percentage points. Samsung aims to quadruple its AI semiconductor-related customer base and increase sales by 9.1 times by 2028 compared to last year, leveraging fast production and low-power, high-efficiency technologies. Orders for CPUs, GPUs, and NPUs from domestic and international AI semiconductor companies are gradually increasing. A Samsung spokesperson mentioned that they will meet the growing demand by expanding their plants in Korea and Taylor, Texas.
