Skip to content
gentic.news — AI News Intelligence Platform
Connecting to the Living Graph…

Listen to today's AI briefing

Daily podcast — 5 min, AI-narrated summary of top stories

POSTECH 10+ Layer Chip Stack Hits 4× Density of 12-Hi HBM
AI ResearchBreakthroughScore: 95

POSTECH 10+ Layer Chip Stack Hits 4× Density of 12-Hi HBM

POSTECH developed 10+ layer chip stacking with 4× HBM density, targeting AI inference memory bottlenecks.

·5d ago·2 min read··27 views·AI-Generated·Report error
Share:
Source: news.google.comvia trendforce_gnWidely Reported
What is POSTECH's new chip-stacking technology and how does it compare to HBM?

POSTECH developed a chip-stacking technology achieving 10+ layers with 4× the density of 12-Hi HBM, potentially enabling higher-bandwidth, lower-latency memory for AI accelerators.

TL;DR

POSTECH achieves stable 10+ layer chip stacking. · 4× density of 12-Hi HBM memory. · Breakthrough for AI inference memory bottlenecks.

POSTECH researchers achieved stable 10+ layer chip stacking with 4× the density of 12-Hi HBM. The breakthrough targets the memory bandwidth bottleneck constraining AI inference accelerators.

Key facts

  • POSTECH achieved 10+ layer chip stacking.
  • 4× density compared to 12-Hi HBM.
  • Uses hybrid bonding (Cu-Cu + oxide fusion).
  • Targets AI inference memory bandwidth bottleneck.
  • No commercialization timeline announced.

POSTECH (Pohang University of Science and Technology) announced a chip-stacking technology that achieves stable operation at 10+ layers, delivering 4× the interconnect density of current 12-Hi HBM (High Bandwidth Memory) stacks. The result was reported by TrendForce on July 10, 2026.

Why It Matters for AI Inference

The 10+ layer stacking targets the memory bandwidth bottleneck in AI inference workloads. Current HBM3E stacks peak at 12 layers (12-Hi) with bandwidth around 1.2 TB/s per stack. POSTECH's approach uses a hybrid bonding technique—combining direct copper-to-copper bonding with oxide-oxide fusion—to reduce layer thickness and improve thermal dissipation. The result is a denser vertical interconnect that could enable higher bandwidth per stack without increasing physical footprint.

Density vs. Commercial Reality

While 4× density over 12-Hi HBM is impressive on paper, the technology remains at the university research stage with no announced commercialization timeline. HBM manufacturers (Samsung, SK hynix, Micron) have already roadmaped 16-Hi and 20-Hi stacks for 2027–2028, though with diminishing returns on density due to thermal and yield challenges. POSTECH's hybrid bonding could leapfrog those roadmaps if scaled, but the gap between lab demo and volume production is wide.

Comparison to Prior Art

The result echoes earlier work by imec and TSMC on hybrid bonding for 3D NAND and logic-on-logic stacking. POSTECH's contribution is specific to DRAM-like memory stacking, where layer count above 8 has historically caused thermal runaway and signal integrity degradation. The team did not disclose the exact layer count beyond "10+" or the specific test chip yields.

What to watch

Micron to begin mass production of 12-Hi HBM3E, next-gen HBM4 ships in ...

Watch for POSTECH to publish a peer-reviewed paper with exact layer count, yield data, and thermal measurements. Also monitor whether Samsung or SK hynix license the hybrid bonding technique for their 2028 HBM4E roadmaps.


Source: news.google.com


Source: gentic.news · · author= · citation.json

AI-assisted reporting. Generated by gentic.news from multiple verified sources, fact-checked against the Living Graph of 4,300+ entities. Edited by Ala SMITH.

Following this story?

Get a weekly digest with AI predictions, trends, and analysis — free.

AI Analysis

POSTECH's result is a credible lab demonstration, but the gap to commercial HBM is large. HBM manufacturers already plan 16-Hi and 20-Hi stacks using existing TSV technology; POSTECH's hybrid bonding offers density gains but introduces new thermal and reliability unknowns. The 4× density claim is relative to 12-Hi HBM, which is already being superseded by 16-Hi products in sampling. The real test will be whether POSTECH can demonstrate >90% yield on a test vehicle with actual DRAM dies, not just dummy layers. Until then, this is a promising research direction, not a product threat.
Compare side-by-side
POSTECH vs TrendForce
Enjoyed this article?
Share:

AI Toolslive

Five one-click lenses on this article. Cached for 24h.

Pick a tool above to generate an instant lens on this article.

Related Articles

From the lab

The framework underneath this story

Every article on this site sits on top of one engine and one framework — both built by the lab.

More in AI Research

View all