Samsung is reportedly planning to supply HBM3 memory and packaging facilities to NVIDIA for its AI GPUs as the company plans to expand its supply chain.
Currently, TSMC dominates NVIDIA's orders for manufacturing wafers and supplying advanced 2.5D semiconductor packaging. However, due to a massive influx of orders, it is being reported that TSMC is unable to handle NVIDIA's demand which is why the company is now planning to adopt a dual-sourcing approach to ensure a steady supply chain.
The primary memory chips used in the AI GPU, HBM3, are currently being supplied by SK Hynix. TSMC has expressed concerns about the huge workload involved with the 2.5D CoWoS packaging for these chips. This has caused NVIDIA to look for other suppliers, with potential suppliers being US-based Amkor Technology and Siliconware Precision Industries Taiwan (SPIL) as previously been pointed out by DigiTimes. Samsung has also entered through its AVP (Advanced Package) division, offering NVIDIA a unique proposition.
Samsung reportedly offered to acquire semiconductor wafers from TSMC and the HBM3 from its memory division. Using the company's unique I-Cube 2.5D packaging, Samsung has sorted out a way for NVIDIA to have a supplier responsible for all development stages potentially. Moreover, the company has also pledged to appoint several engineers to this task with a future proposal of directly acquiring the semiconductor wafer from the foundry division.
Samsung has already initiated the mass manufacturing of HBM3 memory, with reports suggesting that the memory boasts faster speeds (at 6.4 GB/s) and much lower energy consumption than SK Hynix. The process has gained attraction, especially from AMD, since its latest MI300 Instinct APUs are equipped
Read more on wccftech.com