etch-sequencing-induced dielectric defects in high-volume manufacturing
Agilent Technologies; and Abhay Ramrao Deshmukh, National
manufacturers often employ unique practices in their fab operations
resulting from their operational methodology, tool configurations, and
known process sensitivities. To achieve effective tool utilization,
they often configure their dry-etch tools for multiple processes, such
as oxide, nitride, and spin-on-glass (SOG) dielectric etch. The interaction
between these varied processes and their residual effects on chamber
conditions can change process outputs and generate unique defects. Because
the information available in traditional books and best-known methods
generally does not help solve unconventional issues, fabs must resort
to extensive investigations and experimentation to formulate solutions.
article presents a case study from National Semiconductor's fab in Arlington,
TX, in which a residual contaminant from an upstream dry-etch process
interacted with incoming material to produce localized spongy-looking
SOG defects in a downstream interlevel dielectric (ILD). The article
discusses the origin of this unique defect, the process conditions that
affect it, and various methodologies that can be used to monitor, control,
and eliminate it. Finally, the article analyzes the defect's effect
on various electrical test parameters and die yield.
>0.5-µm devices, SOG is used extensively to perform planarization
processes.1 Five steps are performed to achieve a planar
ILD surface: deposition of a base dielectric, SOG deposition, SOG etch
back, SOG cure, and deposition of a cap dielectric. Two types of dielectric
materials are typically used in ILD processes: doped and undoped ones.
Doped oxides can be obtained by adding appropriate doping gases, such
as phosphine (PH3) or diborane (B2H6).
The ILD scheme under discussion here used phosphorus-doped tetraethylorthosilicate
this case, an intermittent yield loss in a few lots was investigated
and found to be related to metal bridging. The scanning electron microscope
(SEM) cross-section images in Figure
1 show a bad wafer with metal bridging and a known good one without
metal bridging. In Figure 1a, the image indicates that the bridging
was caused by unetched metal steps, which in turn were caused by spongy
and void-filled defects in the cap dielectric. It can be seen that these
spongy cap-dielectric defects were localized—in other words, they
appeared only over the SOG pockets. In contrast, no defects could be
observed over the non-SOG areas. After a detailed analysis, a correlation
was drawn between the occurrence of the defects and the sequencing of
defect-bearing lots during the SOG etch-back step through the plasma
etch tools. These lots were preceded by bond-pad etch lots.
the bond-pad etch step, especially for big bond pads, large quantities
of residual aluminum and titanium contaminant are sputtered from the
underlying metal layer and deposited onto the etch-chamber walls. This
contaminant is resputtered during subsequent etch processes. In the
of the SOG etch-back step at National, it was thought that phosphorus
present in the underlying base dielectric aggressively gettered that
residual aluminum, depositing a thin polymeric film over the SOG pockets
on the wafer. Outgassing from this polymeric film during cap dielectric
deposition caused the cap dielectric to become spongy, as shown in Figure
1a. The intermittent nature of the problem could be explained by the
fact that the metal steps resulting from the spongy SOG issue were sometimes
small, and metal overetch was sufficient to etch it completely.
the Spongy Defects
eliminate the spongy defects, a standard problem-solving procedure was
performed in which affected and unaffected wafers were grouped and contrasted
and failure theories were formulated.
Content. The engineers observed that only the dielectric deposited
over the SOG pockets was affected and that there was no commonality
among the affected lots after the SOG cure step. Hence, their initial
failure theories focused on the residual solvent content of the SOG.
Efforts were made to recreate the defect problem by increasing the solvent
content, limiting the duration of the SOG cure step, and skipping the
SOG bake process altogether. However, these measures failed to reproduce
the problem, indicating that it was not related to the residual solvent
content of the SOG.
Sequencing. After performing the solvent-content experiments,
the engineers investigated the product mix by determining when problem
lots had been processed relative to etch-chamber clean cycles and by
analyzing the lots and processes that were run before and after the
problem lots. Based on data from those tests, the investigators observed
that there was no relationship between the production of problem lots
and etch-chamber cleans. However, all the problem lots had been run
through the SOG etch-back step immediately following the bond-pad etch
process. This effect was illustrated clearly when the circuit yield
for lots preceded by bond-pad etch was compared with the circuit yield
for lots preceded by other processes. A comparison between the two groups,
illustrated in Figure 2, indicates that the average circuit yield for
the lots preceded by bond-pad etch was approximately 5% lower than that
for the lots preceded by other processes.
2: Circuit yield from bad lots with spongy defects (in which the
SOG etch-back step was preceded by bond-pad etch) and circuit yield
from good lots (in which SOG etch back was not preceded by bond-pad
trend investigated by the engineers was wafer quality at the beginning
and end of a lot. After each successive SOG etch-back wafer was processed
following bond-pad etch, chamber seasoning resulted in a decrease in
the residual aluminum level in the etch chamber. This phenomenon was
reflected in decreasing yield loss as wafer processing continued. Figure
3 compares the normalized yield for the first half versus the second
half of a lot that had undergone the SOG etch-back process following
the bond-pad etch process. Yields from the second group were demonstratively
better than those from the first group.
3: Comparison between the normalized yield for the first half versus
the second half of a lot that had undergone the SOG etch-back process
following the bond-pad etch process.
secondary ion mass spectroscopy was used to determine the aluminum and
titanium concentrations. Data from a wafer that had undergone the SOG
etch-back step following bond-pad etch were compared with data from
a control wafer that did not undergo bond-pad etch. The aluminum level
on the experimental wafer was 35 to 45 times higher than that on the
control wafer, while titanium intensity was 130 to 170 times higher
and titanium oxide intensity was 80 to 100 times higher.
Film Composition. First-level dielectric films use phosphorus-doped
materials because phosphorus plays an important part in the gettering
of aluminum. In contrast, at the intermetal levels, undoped dielectrics
are commonly used. To determine the effects of doped and undoped dielectric
materials on the formation of the spongy defects, the investigators
compared cross-sectioned SOG pockets from wafers with a doped or undoped
base dielectric that had been processed after the bond-pad etch process.
The spongy defects could be seen only on the wafers with a doped base
Process Monitoring and Control
optimize in-line process monitoring, various inspections were performed
during the process sequence. During one such inspection, a subtle shift
in the carbon monoxide (CO) signal was observed, as illustrated in Figure
4. The figure compares a typical CO signal obtained from a plasma during
the SOG etch-back step with the CO signal obtained from a wafer processed
after the bond-pad etch process. The typical SOG etch-back signal rises
steadily as more SOG is etched away and a higher CO signal is obtained
from the underlying base dielectric than from the SOG. The signal flattens
when most of the signal is obtained from the base dielectric. However,
when SOG etch back is performed after bond-pad etch, the CO signal rises
quickly, with an initial kink followed by a relatively longer flat region.
This response indicates that most of the signal is obtained from the
base dielectric, not the SOG. The engineers theorized that the SOG surface
experiences a polymerizing reaction at that moment.
4: Comparison between the monitored CO signal from wafers that underwent
the etch-back step following bond-pad etch and control wafers that
did not undergo bond-pad etch.
another experiment, the effect of when a wafer in the lot was processed
was captured by monitoring the CO signal. Figure 5 compares the CO signal
from the first two and the last two wafers of a lot that underwent SOG
etch-back following bond-pad etch. It shows that as the residual aluminum
in the chamber was depleted, the CO signal began to resemble a typical
SOG etch-back signal from a good lot.
5: Comparison between the CO signal from the first two and the last
two wafers of a SOG etch-back lot that was processed after bond-pad
article described an investigation in which ILD deposited over SOG pockets
resulted in spongy defects. Experiments were performed to test a range
of theories and uncover the root cause of the failure. The failure mechanism
involved a complex interaction among the residual aluminum in an oxide
etch chamber, the phosphorus content of underlying dielectric films,
and SOG. After the root cause of the problem was pinpointed, the investigators
performed in-line monitoring to control it. Interim control over the
defect problem was obtained by segregating the tools that perform both
SOG etch-back and bond-pad etch. In addition, the engineers investigated
the use of undoped base dielectric to help solve the problem.
work illustrates the need for caution when increasing tool utilization
by using a single tool to perform different etch processes. It also
highlights the increasing importance of analyzing the product mix to
solve process problems and reduce yield-limiting failures.
work was performed at National Semiconductor's Arlington, TX, manufacturing
facility. The technical input and encouragement from Glen Rentschler,
Chara Mathur, Steven Watson, Amber Longstreet, John Devlin, and Nihar
Kanungo are greatly appreciated.
Wilson, CJ Tracy, and JL Freeman Jr., eds., Handbook of Multilevel
Metallization for Intergrated Circuits: Materials, Technology, and Applications
(Westwood, NJ: William Andrew/Noyes, 1993), 382–402.
Doad is an R&D engineer at Agilent Technologies (Fort Collins,
CO). He has more than nine years of experience in various areas of semiconductor
processing. Before joining the company, he held a process engineering
position at National Semiconductor. He received an MS in chemical engineering
from Arizona State University in Tempe. (Doad can be reached at 970/288-4903
Ramrao Deshmukh is a process engineer at National Semiconductor
in Arlington, TX. He has worked at the company for four years in the
areas of PVD systems and SOG planarization. He received an MS in materials
sciences from the South Dakota School of Mines and Technology in Rapid
City. (Deshmukh can be reached at 817/705-7345 or email@example.com.)