Page 9 - Chip Scale Review_November-December_2023-digital

than the barrier as shown in Figure 3a,
while keeping the roughness, Cu dishing
and erosion within the D2W bonding
process requirements. The same bonding
and annealing was repeated. The results
showed that the Cu-Cu bond was fused,
and was further validated with transmission
electron-back-scatter-diffraction (T-EBSD)
portraying a successful Cu diffusion across
the bonding boundary (Figure 3b).

Surface activation, cleanliness,
queue time and electrical yield
Once the interfaces were optimized as
discussed in the section on PVD-CMP
co-optimization, we moved on to study
dielectric surface activation, the impact
of a particle-free surface and queue time.
These requirements are more stringent for
Figure 2: a) A PVD thick barrier-CMP interaction with zoom in focus and an illustration of barrier protrusion, D2W HB than for W2W HB because of
which resulted in b) a Cu-Cu bonding gap due to barrier-CMP interaction (see the XSEM). the need to process dicing and singulated
dies on flexible organic tape or on carrier
CMP optimization [4]. However, when the pads, and in addition to the already reduced wafers with organic adhesive.
same process was applied to D2W HB, it force acting on the bond interfaces from Plasma activation has been demonstrated
resulted in a visible Cu-Cu gap from X-ray grinded and singulated die, was preventing to increase hydrophilicity for dielectrics
secondary emission microscopy (XSEM) the “tacking” of dielectric-dielectric during such as SiO 2 and SiCN [6]. Hydrophilicity
analysis as shown in Figure 2a due to die placement. At post-bond anneal, the is achieved by the presence of the silanol
insufficient force (weight) to form good protruded barrier has higher thermal groups (Si-O-H) on the dielectric surfaces.
tacking (i.e., low force was used to bond expansion than the oxide (Ta: 6.5 10 /K Ion energy is one way to promote the
-6
the samples compared to the forces used in vs. SiO 2 : 0.65 10 /K [5]) and therefore was silanol groups’ attachment to the dielectric
-6
W2W bonding) as a result of thinning and expanding faster, pushing the two bond surface by creating disorders and high-
dice. The failure analysis of a cross section interfaces apart and preventing the Cu energy states on the surfaces. The presence
with transmission electron microscopy pads from making contact and diffusing of these silanol groups is essential, as they
(TEM) at the edge of the Cu bond pad (or inter-diffusing), giving rise to a wider enable initial bonding when dielectric
(Figure 2b), indicated the barrier was gap (20~40nm) than initial dishing. The surfaces from substrates and chiplets are
higher than the dielectric by 1-2nm. One physical vapor deposition (PVD) barrier- brought together even at room temperature.
hypothesis suggested that this barrier CMP process was optimized by using The measurement of the hydrophilicity
protrusion was standing out from the rest a thin barrier and a tuned CMP process of dielectric surfaces is a way to evaluate
of the dielectric at the corner of the bonding to ensure that oxide was always higher the effectiveness of the activation

Figure 3: a) An optimized thin PVD-CMP, with zoom in focus and an illustration of higher dielectric than barrier in nm-scale, which resulted in a Cu-Cu diffusion (see the
XSEM) and confirmation with T-EBSD showing grain growth across the bond boundary.

Chip Scale Review November • December • 2023 [ChipScaleReview.com] 7 7