Jukan@jukan05
Micron to Attempt Industry-First HBM-Style GDDR Stacking
U.S.-based Micron is attempting to stack graphics DRAM (GDDR) for the first time in the industry. The move aims to boost performance by stacking GDDR in a manner similar to high-bandwidth memory (HBM), as a response to the diversifying demand for AI memory. Attention is on whether this could fire the starting gun on a new memory stacking competition.
According to industry sources on the 30th, Micron has begun developing a new product that vertically stacks GDDR. It has been confirmed that the company plans to have the relevant equipment in place and begin process testing by the second half of this year.
An industry source with direct knowledge of the matter said, "An initial stack of around four layers of GDDR is expected," adding that "samples could emerge as early as next year." Given that specifications including the number of layers are already under discussion, the effort is likely in response to requests from customers such as AI accelerator makers.
GDDR is a type of memory specialized for video and 3D graphics. It has traditionally been used primarily in graphics cards and gaming devices, but has recently been adopted in some AI accelerators as well. While slower in bandwidth compared to HBM, its cost competitiveness has driven growing demand for specific use cases such as inference workloads.
Micron's move is analyzed as targeting this trend. The attempt is to stack GDDR like HBM to develop a product that, while slower than HBM, would be faster and offer greater capacity than conventional GDDR.
As the AI market has expanded recently, a variety of products are being developed according to use case and price point, and memory demand is diversifying accordingly. A prime example is NVIDIA's use of SRAM in its inference-dedicated chips.
Stacked GDDR products are expected to occupy the market segment between HBM and conventional (non-stacked) GDDR. Significant demand is also anticipated in high-performance gaming graphics cards, a segment that continues to see sustained growth.
If Micron succeeds in stacking GDDR, it could seize first-mover advantage in the market. Although it is currently a niche segment, the company appears to have judged that the proliferation of AI gives it meaningful growth potential, prompting it to pursue full-scale commercialization. The strategy is to preemptively target the stacked GDDR market ahead of Samsung Electronics and SK Hynix to secure a competitive edge.
Stacked GDDR has so far only been explored at the level of early-stage R&D and academic papers. With no prior precedent for mass production, whether Micron can clear the technical hurdles remains a key question. Challenges cited include the method of inter-die stacking, as well as power consumption and thermal management. Cost control through the stacking process is also essential, as the product must maintain a compelling performance-to-price ratio relative to HBM to secure market viability — meaning the utility of stacked GDDR must be maximized.
An industry source said, "Until now, commercialization had not been actively pursued not only due to technical issues but also because the market positioning of stacked GDDR was unclear," adding that "as the need for stacked GDDR emerges amid changes in the AI memory market, Micron's challenge could serve as the starting gun for a new memory stacking competition."
$MU
