SemiconductorsFEATURE

How Advanced Packaging and Chiplets Are Reassembling Compute as Single-Die Scaling Slows

林紀旭 James LinEditor-in-Chief
Published · Updated
TSMC, Intel, and the UCIe Consortium each detail distinct paths to multi-die integration: TSMC's CoWoS interposer exceeds 1,700mm² to pack an SoC with more than four HBM cubes, Intel's EMIB bump pitch scales from 55 to 45 micron, its Foveros Direct 3D copper bonding pitch shrinks from 9 to 3 micron, and UCIe pursues an open chiplet interconnect standard.

How does TSMC's CoWoS platform integrate multiple dies through a large interposer?

TSMC's CoWoS platform uses an interposer larger than 2x the standard reticle size, roughly 1,700mm², to integrate a leading SoC chip with more than four HBM2/HBM2E memory cubes in a single packageCITE:E1. TSMC describes CoWoS as a "wafer level system integration platform" that offers a range of interposer sizes, HBM cube counts, and package sizesCITE:E2.

How does Intel's EMIB bump pitch scaling support high-bandwidth chiplet interconnects?

Intel's second-generation EMIB technology scales bump pitch from 55 micron to 45 micron, enabling high-bandwidth connectivity with either Foveros Direct 3D chiplets or multiple I/O chipletsCITE:E3. Intel states this lets Foundry customers connect chiplets across a package rather than relying on a single monolithic dieCITE:E3.

How does Intel's Foveros Direct 3D copper bonding pitch scaling raise stacked-chiplet interconnect density?

Intel's first generation of Foveros Direct 3D uses copper-to-copper bonding at a 9-micron pitch, and the second generation shrinks that pitch to 3 micronsCITE:E4. This pitch reduction applies to direct 3D die-on-die stacking, distinct from the lateral, interposer-based connections used in EMIB and CoWoSCITE:E4.

How does the UCIe standard drive an open chiplet ecosystem?

The UCIe Consortium states its standard is built to establish "an open ecosystem of chiplets for on-package innovations"CITE:E5. This positions UCIe as a common interconnect specification that packaging platforms such as CoWoS and EMIB/Foveros Direct can build toward, rather than each relying solely on proprietary connectionsCITE:E5.

Packaging metrics compared

TechnologyEntityMetricValue
CoWoS interposerTSMCSize vs. standard reticle>2x reticle (~1,700mm²)
CoWoS packageTSMCHBM cubes integratedMore than 4 HBM2/HBM2E cubes
EMIB, 2nd generationIntelBump pitch55 micron → 45 micron
Foveros Direct 3D, 1st→2nd genIntelCopper bonding pitch9 micron → 3 micron

Across the three sources, two distinct geometries emerge for integrating multiple dies once single-die scaling slows. TSMC's CoWoS scales laterally, expanding interposer area to ~1,700mm² to fit an SoC alongside four or more HBM cubesCITE:E1CITE:E2. Intel scales both laterally and vertically: EMIB narrows bump pitch from 55 to 45 micron for chiplet-to-chiplet links on a substrateCITE:E3, while Foveros Direct 3D narrows copper bonding pitch from 9 to 3 micron for direct die-on-die stackingCITE:E4. UCIe's stated goal of an open chiplet ecosystem points toward interconnects that could, in principle, span across such platform-specific approachesCITE:E5.

📊 Evidence

FAQ

How does TSMC's CoWoS platform integrate multiple dies through a large interposer?

TSMC's CoWoS platform uses an interposer larger than 2x the standard reticle size, roughly 1,700mm², to integrate a leading SoC chip with more than four HBM2/HB…

How does Intel's EMIB bump pitch scaling support high-bandwidth chiplet interconnects?

Intel's second-generation EMIB technology scales bump pitch from 55 micron to 45 micron, enabling high-bandwidth connectivity with either Foveros Direct 3D chip…

How does Intel's Foveros Direct 3D copper bonding pitch scaling raise stacked-chiplet interconnect density?

Intel's first generation of Foveros Direct 3D uses copper-to-copper bonding at a 9-micron pitch, and the second generation shrinks that pitch to 3 micronsCITE:E…

How does the UCIe standard drive an open chiplet ecosystem?

The UCIe Consortium states its standard is built to establish "an open ecosystem of chiplets for on-package innovations"CITE:E5.

📎 Sources

  1. tsmc.com
  2. intel.com
  3. uciexpress.org

Related data

Author's Take林紀旭 James Lin

The two vendors are optimizing different axes of the same problem. TSMC's CoWoS scales by area — a ~1,700mm² interposer fitting an SoC plus four-plus HBM cubes — which favors breadth of memory bandwidth per package. Intel is scaling by pitch: EMIB's bump pitch moving from 55 to 45 micron and Foveros Direct 3D's copper bonding pitch moving from 9 to 3 micron both push interconnect density rather than package area. That 3-micron Foveros Direct pitch is the number worth tracking — it is a substantially finer bond pitch than the EMIB bump pitch, and whether it ships in volume will indicate whether direct die-on-die stacking or interposer-based lateral integration becomes the denser path for multi-chiplet designs. Separately, UCIe's open-ecosystem goal only matters in practice if chiplets built to it can actually land inside CoWoS- or EMIB/Foveros-based packages — that cross-platform adoption is the second thing to watch.

林紀旭 James LinEditor-in-Chief

Related

BRIEF

GreenTrans Unveils GT5X, GT3X Quadruped Robots, Targets 100% Taiwan-Made Content by 2027

GreenTrans (綠捷), the robotics subsidiary of China Motor (中華車), unveiled quadruped robots GT5X and GT3X at SEMICON Taiwan 2026, targeting 100% Taiwan-made content by 2027. The robots combine an in-house-designed control unit and battery management system, NVIDIA's Jetson Orin and Isaac Lab platforms, and a new LFP battery developed with Formosa Smart Energy (台塑新智能), while GreenTrans's inspection robots are already deployed in semiconductor fabs.

EffectStory 編輯部 ·
BRIEF

Nvidia Confirms $12.93 Billion Acquisition of Hugging Face

Nvidia confirmed on September 3, 2026 that it agreed to buy Hugging Face for $12.93 billion, exactly $12,930,300,000, gaining the open-source AI hosting platform used by over 18 million developers. CEO Jensen Huang pledged the platform will stay open, with no Nvidia compute required to build on or deploy through it.

EffectStory 編輯部 ·
BRIEF

NVIDIA to Subscribe US$3.5 Billion of MediaTek's Record US$3.9 Billion Convertible Bond

NVIDIA will subscribe US$3.5 billion of MediaTek's US$3.9 billion offshore convertible bond offering, the largest such issuance in Taiwan's capital market history, deepening cooperation in AI infrastructure, edge AI computing, and automotive platforms while marking NVIDIA's first major investment in a Taiwanese company.

EffectStory 編輯部 ·