ECTC 2026 | ASE | Characterization of Bonding Behavior and Void Formation in Chip-on-Wafer Hybrid Bonding

2 days ago

5 min read

The Achilles' heel of CoW (chip-on-wafer) hybrid bonding is "interfacial voids": voids trapped at the interconnect interface cause electrical opens and wipe out functional yield. At ECTC 2026, ASE presented a low-cost, purely mechanical fix: tuning the initial "bonding-front angle". When the bonding wave advances in order from the die center to the edge, gas is pushed out smoothly; multi-point or disordered contact instead seals air in the center. Testing four angles A<B<C<D: at angle A (baseline) voids are trapped at the die center, as the angle increases they migrate toward the edge, and angle D is completely void-free. On 8×12×0.05mm TEOS oxide dies, yield jumped from 2.8% to 99%. The method is very robust on thin dies: 0.03mm ultra-thin dies warp as much as 102.4µm (41.3µm at 0.05mm, 14.7µm at 0.10mm), and angle D achieved zero voids at all three thicknesses. Pushed further to patterned wafers (pitch down to 6µm) and SiCN and organic-on-TEOS heterogeneous interfaces, CSAM yield was consistently 95–99%, with FIB cross-sections showing no delamination or micro-voids. In one sentence: CoW hybrid bonding yield is first gated by the mechanical knob of the "bonding-front angle"; set it right and the process window widens dramatically.

1. Background: The Yield Bottleneck in CoW Hybrid Bonding Is Interfacial Voids

This paper comes from ASE and was presented at the 2026 IEEE 76th ECTC. CoW hybrid bonding forms dielectric and metal bonds simultaneously with excellent electrical performance, and is one of the mainstream architectures for ultra-high-density 3D-ICs in HPC. But interfacial voids are a key bottleneck: trapped at the interconnect interface, they cause electrical opens and wipe out functional yield. Plasma activation of the surface to increase hydrophilicity and bonding energy is necessary, but the physical boundary conditions of initial contact are equally decisive: the bonding wave must advance in an orderly, continuous way to expel gas; any disordered or multi-point contact traps air and turns it into persistent voids.

For the context of packaging integration and 3D stacking, see CPO Is Won in Packaging, Not Optics: John Lau on PIC/EIC Heterogeneous Integration; for why the industry is moving toward fine pitch and heterogeneous integration, see The Three-Stage Evolution of CPO: From Scale-Out to Scale-Up.

Table I: CSAM for four bonding-front angles A→D. At angle A voids sit at the center; as the angle increases they migrate to the edge; angle D is completely void-free | Source: ASE, ECTC 2026 - Table I

2. The Core Question

How can interfacial voids in CoW hybrid bonding be eliminated at low cost, and can the fix generalize across die size, thickness, pitch and material? The answer: set the initial bonding-front angle to D so the wave advances in order from center to edge and expels the gas. Yield on 8×12×0.05mm dies goes from 2.8% to 99%, and stays at 95–99% on ultra-thin dies with 102.4µm warpage, 6µm pitch and organic interfaces.

3. Key Figures and Tables, One by One

3.1 Die Preparation: Plasma Dicing for Clean Surfaces (Fig. 1)

The top dies are plasma-diced: a water-soluble protective layer (PL) is spin-coated to prevent contamination from laser debris and plasma bombardment → laser grooving (LG) defines the dicing streets, removes the PL and dielectric and exposes silicon → a Bosch process high-aspect-ratio plasma dry etch completes singulation → a DI water rinse removes the PL, leaving a clean, residue-free surface. The edges are free of re-cast layers and particle contamination that would disturb the bonding wave. Bonding uses a Shibaura TFC-6500 (1µm 3σ alignment), with the initial contact angle tuned by mechanically controlling the bond head's orientation and motion.

3.2 Bonding-Front Angle: Voids Migrate from Center to Edge, Then Disappear (Table I, Fig. 2)

Four angles: A (baseline, untuned) < B < C < D. At angle A voids are trapped near the die center; increasing the angle from B to C clearly moves the voids toward the edge; angle D is completely void-free, as the steeper contact angle lets the bonding front expel gas more effectively. On 8×12×0.05mm TEOS dies, angle A yields only 2.8% because of dense voids, while angle D yields 99%.

3.3 Robustness Across Thickness, Pitch and Material (Tables II/III, Figs. 3/4)

Thickness ↔ warpage: 0.03mm = 102.4µm, 0.05mm = 41.3µm, 0.10mm = 14.7µm (the thinner, the more warpage); angle D was void-free at all three thicknesses, showing it works on ultra-thin dies and widens the process window. Patterned wafers: four TVs with pitch shrinking from 15µm to 6µm, CSAM yields of 96%/99%/95%/99%, and FIB cross-sections with no delamination or micro-voids. Heterogeneous materials: organic-on-TEOS (8×12×0.10mm) yield 99%, showing the parameters transfer from inorganic dielectrics to organic interfaces.

[Image] Which one: Characterization of Bonding Behavior..., Table III | Caption: Table III: Patterned TV specifications and yield. Pitch 15→6µm, CSAM yield 95–99% | Source: ASE, ECTC 2026 - Table III

4. Technical Highlights

First, a low-cost, purely mechanical fix takes CoW yield from 2.8% to 99%: no new materials or chemistry, just tuning the initial bonding-front angle (A→D) so the wave advances in order from center to edge and expels the gas. The voids first migrate from the center to the edge, then disappear.

Second, proven to generalize across size, thickness, pitch and material: 95–99% on 0.03mm ultra-thin dies (102.4µm warpage), 6µm fine-pitch patterns, and SiCN and organic-on-TEOS heterogeneous interfaces, with FIB cross-sections showing no delamination or micro-voids, greatly widening the process window.

5. Industry Connections

Distance to market: maturity is at the "process-parameter experimental validation" stage, with TVs covering multiple sizes, thicknesses, patterned wafers and heterogeneous materials, verified by CSAM and FIB. The bonding-front angle is a mechanical knob that existing tools can already adjust, so the barrier to adoption is low and it is close to volume production. Beneficiaries: OSATs doing CoW HB (here, ASE) and their customers, plus the plasma dicing and bonding equipment supply chain. A sober view: this study focuses on yield (the void-free ratio from CSAM) and does not go deep into long-term electrical or reliability data; the optimal angle must be recalibrated by die size and warpage. What it fills in is "how to eliminate central voids in CoW hybrid bonding with the low-cost knob of initial contact angle."

6. Conclusion

CoW hybrid bonding yield is first gated by the "initial bonding-front angle." ASE set the angle to D so the wave advances in order from center to edge, lifting yield on 8×12×0.05mm dies from 2.8% to 99%, and holding 95–99% on ultra-thin dies with 102.4µm warpage, 6µm pitch and organic interfaces. What to watch: this is a low-cost, purely mechanical fix that generalizes across materials and sizes, greatly widening the process window for high-density 3D integration.

References

Zhao-Ze Jiang, Chih-Jing Hsu, Chen-Hung Lee, Che-Ming Hsu, Jen-Chieh Kao, Yung-I Yeh, Alexis Angelo Garcia, Po Hsiang Wang, "Characterization of Bonding Behavior and Void Formation in Chip-on-Wafer Hybrid Bonding," 2026 IEEE 76th ECTC. Advanced Semiconductor Engineering (ASE) Group.