SK hynix Produces HBM4 Faster than JEDEC Specs Entering Mass Production

This might seem trivial, but SK hynix says that its HBM4 has hit new milestones. HBM4 is the next generation of high-bandwidth memory that is destined for next-generation AI accelerators. More memory throughput often means that the compute elements in those AI accelerators are better fed with data and therefore the entire AI chip and cluster can operate faster.

SK hynix Produces HBM4 Faster than JEDEC Specs Before Mass Production

The standard JEDEC (Joint Electron Device Engineering Council) of HBM4 is a 2048-bit wide interface with 8Gbps of throughput per pin. SK hynix says that it is implementing “over 10Gbps” per pin, which is at least 25% faster.

It also is using its 1bnm 10nm process and Mass Reflow Molded Underfill to help minimize production risk. MR-MUF uses liquid injected between chips instead of thin films that are layers between them. SK hynix says that process combination helped bring these new chips to the point that they can be mass produced.

Part of SK hynix’s announcement is that it has completed development and is readying to ship. This is also important because Micron said it would start shipping HBM4 but was advertising JEDEC specs at 2TB/s.

Final Words

This may seem trivial at first, but it is a big deal. Instead of just planning for JEDEC speeds, SK hynix is going to let customers go faster. For example, when DDR5 was first launched in servers we had DDR5-4800. This is more like having DDR5-6000 as an officially supported speed. That is not a perfect example because DDR5 is slotted and can be upgraded, but it at least gives some sense of this type of announcement and its impact. If you are a company like NVIDIA, you have customers that are willing to pay a premium for a premium product constructed with faster HBM4. This is great work by the SK hynix team.