Samsung’s HBM4 Logic Die Reaches Stability for Mass Production

While Samsung is going through tough times in its HBM business, it aims to bounce back with the upcoming HBM4 chip. The good thing is that the company appears to be on the right track. Its HBM4 4nm logic die has reportedly achieved a yield of over 90%. This is a very promising result for stable mass production, which should attract major clients around the world.

Samsung makes advancements on HBM4 ahead of mass production

Samsung has adopted an in-house 4nm process for the logic die of HBM4. The logic die sits at the bottom of the HBM, which consists of several stacked memory chips (DRAMs). It is a key component that connects the stacked memory to the GPU while also controlling power and data signals.

A few months ago, a report revealed that Samsung achieved over 40% yield in HBM4 logic die tests. The company has now made substantial progress with a yield of over 90%, according to a report from Chosun. This also means the production process has become stable enough for mass production.

“Samsung Electronics applied a fine process to the HBM4 logic die ahead of its competitors in order to make up for its slump in the previous generation,” said a semiconductor industry insider. “Although there is still market verification left, it is true that the importance of foundry process technology will be highlighted in the upcoming next-generation HBM market.”

As of now, Samsung’s memory division is preparing HBM4 samples for customers like Nvidia and AMD. The goal is to clear all the necessary quality tests as quickly as possible, paving the way to secure deals. This could help the company gain a foothold in the growing AI memory market.

It is worth noting that achieving a high yield on the logic die alone is not enough. Samsung needs to obtain promising yields for the 1c DRAM for smooth mass production. Hopefully, the company can achieve stable yields for both key components of HBM4 chips. The firm plans to begin mass production of 12-layer HBM4 chips sometime in 2026.