SK hynix's MR-MUF packaging stacked 12 HBM layers as thin as 8 and won the AI memory lead

High-bandwidth memory (HBM) stacks DRAM dies vertically and connects them, giving AI accelerators the speed and capacity they need. Capacity rises with layers, but every extra layer worsens two problems: the stack warps, and heat cannot escape — both of which break yield.

SK hynix built its own answer, mass reflow molded underfill (MR-MUF): instead of laying a film between every die like rivals, it injects liquid protective material that hardens in one reflow step, cutting stacking pressure and controlling warpage while dissipating heat well.

In September 2024 SK hynix began volume production of the world's first 12-layer HBM3E — 36GB, same thickness as the 8-layer product. Thinning each DRAM 40% and applying Advanced MR-MUF raised capacity 50% and cut heat 10% at 9.6Gbps, with yield around 80%, the industry's best.

One-step die fill lets stacks grow without extra thickness

One-step underfill beats per-layer film for speed and heat

Warpage control makes tall stacks manufacturable

Thinner dies buy capacity without extra thickness

First-mover supply to NVIDIA locked in the lead

In a stacking business the packaging process is the moat: whoever controls warpage and heat decides how many layers — and how much value — a product can carry, and the lead compounds fast.

A month after the 12-layer launch SK hynix announced 16-layer HBM3E (48GB) on the same Advanced MR-MUF process, and by 2024 it dominated HBM supply to NVIDIA while Samsung and Micron lagged on yield. The process became the template for HBM4-era hybrid bonding work.