According to industry sources, Micron has begun preparing equipment for a new vertically stacked GDDR product and plans to start process testing in the second half of 2026. Early design concepts indicate an initial structure of around four stacked layers, with prototype samples potentially emerging in 2027.

GDDR, traditionally used in video processing and 3D graphics, has seen increasing adoption in specific AI accelerators. Although it offers lower bandwidth than High Bandwidth Memory (HBM), its lower cost has made it attractive for AI inference workloads - an area Micron is believed to be targeting.

Micron’s aim is to apply stacking methods similar to those used in HBM, potentially boosting the performance and capacity of conventional GDDR while still maintaining a cost advantage. Demand is also anticipated from the gaming sector, where high-performance graphics cards continue to drive substantial memory requirements.

While stacked GDDR would not match HBM in raw performance, it could benefit from simpler packaging and greater supply scalability. HBM relies on hundreds of through-silicon vias (TSVs) and complex interposer-based packaging, such as CoWoS, which has contributed to supply constraints. Stacked GDDR, by contrast, could use less complex manufacturing processes, offering a more economical and scalable alternative.

Industry analysts suggest that if stacked GDDR gains adoption in AI inference, a phase that requires deploying tens of thousands of servers, it could challenge an overreliance on HBM. With GDDR priced at roughly 5–10% of HBM per gigabyte, it may offer significant savings for large-scale deployments.

Stacked GDDR has so far remained largely within research environments, with no proven path to mass production. Whether Micron can overcome the technical hurdles remains uncertain. Key challenges include creating reliable stacking methods for GDDR dies, managing thermal and power constraints, and ensuring that production costs remain competitive relative to HBM.

If successful, the effort could give Micron an early lead in a developing market segment as the demand for AI infrastructure accelerates, potentially placing it ahead of competitors such as Samsung Electronics and SK hynix.

By: DocMemory
Copyright © 2023 CST, Inc. All Rights Reserved