SK Hynix Could Partner With Intel Foundry to Produce Next-Generation HBM4E Memory

SK Hynix is considering changing its strategy for the production of its upcoming HBM4E memory. According to reports from the South Korea Herald, the South Korean company is looking to diversify its network of suppliers and adopt a dual-supply model, where Intel Foundry would compete directly with TSMC to manufacture the base dies of these high-speed modules.

Unlike previous generations, HBM4E memory includes a base layer with custom logic, on top of which traditional DRAM dies are stacked. This architecture makes it possible to integrate memory controllers and PHY circuits directly into the base die, instead of placing them in the main processor.

Intel Foundry Could Challenge TSMC for SK Hynix’s HBM4E Memory Production

With this design, manufacturers of artificial intelligence accelerators can free up valuable space on their computing chips to add more processing units, reducing latency and greatly increasing overall performance.

Up to the HBM3E generation, SK Hynix manufactured its own base dies using its proprietary 10 nm-class nodes. However, for HBM4, TSMC’s 12 nm node was used, and with HBM4E, industry requirements demand even more advanced lithography. This is where Intel’s foundry division comes into play. Although cutting-edge technologies such as Intel 18A are a possible option, financial analysis suggests that using mature and efficient nodes such as Intel 3 and Intel 4 makes much more sense right now.

This move is excellent news for Intel Foundry’s ambitions. At the Hot Chips event, it was revealed that SK Hynix has already validated its designs to work with Intel’s EMIB-T packaging technology. In addition, the use of through-silicon vias (TSVs) will allow vertical power distribution in the memory stack, making this partnership extremely important for advancing artificial intelligence infrastructure.