MIL- STD-883F 2004 TEST METHOD STANDARD MICROCIRCUITS.pdf - 第610页
MIL-STD-883F METHOD 5004.11 18 June 2004 20 APPENDIX A ATTACHMENT 1 Example 1 - Qualit y Scenari o : A defect charac teri zation has been perfor med on an LPCVD operati on. The pri mary defec t mechani sm was found to be…
MIL-STD-883F
METHOD 5004.11
18 June 2004
19
APPENDIX A
Screens
- 100% of a population (dice or wafers) is inspected or tested and all material containing targeted defects are
rejected.
Sub-process
- Any number of related process steps leading to an outcome on the wafer. Examples would include poly
interconnect formation (comprised of poly deposition, poly layer lithography, poly etch and resist strip) and contact
formation (dielectric deposition, contact layer lithography, contact etch and resist strip).
Telescoping defects
- Defects which increase in visibility, due to an apparent increase in size, as wafers are processed
through subsequent operations. The increase is a function of the defect being decorated by etches or films, the defect
acting as a nucleation site for subsequent depositions or by the defect creating non-uniform regions in a film or oxide.
Test
- 1) Evaluate (ie: stress and measure) reliability, quality and performance; 2) ensure the defects present do not
affect reliability, quality or performance.
Unexpected failures
- Failures that are not detected, or cannot be predicted, using the manufacturer's standard in-line
inspection and containment plans.
Wafer process
- The materials, equipment, operations and environment necessary to manufacture a product or family of
products. This includes all potential sources of defect generation.
Yield analysis
- The analysis of die yields to determine failure modes and defect mechanisms. This can entail analyzing
low yielding material, average yielding material or high yielding material or combinations of these items. This type of
analysis can be used to validate in-line monitors.
MIL-STD-883F
METHOD 5004.11
18 June 2004
20
APPENDIX A
ATTACHMENT 1
Example 1 - Quality Scenario
:
A defect characterization has been performed on an LPCVD operation. The primary defect mechanism was found to be
particles. These particles were quantified using a laser surface-scanning tool. The results show that the particles fell into
three size distributions: 1) <0.3 microns randomly distributed from wafer to wafer and within a wafer, 2) about 1.0 microns
with a higher density near the pump end of the deposition tube, and 3) greater than 6.0 microns that appeared heavily on
some wafers but did not appear at all on others. The defects in the 1.0 micron or less categories were found to be relatively
small, dark particles when viewed with an optical microscope. The larger particles (>6 microns) appeared as large, black
particles that appeared to be on the wafer surface. A compositional analysis of particles from the three distributions showed
that the first two types (<0.3 microns and about 1.0 microns) were composed of Si and O, essentially the same composition
as the deposited film. The large particles were composed of primarily Fe and Ni.
Wafers containing defects from the smaller size distributions were processed through the subsequent patterning
operations. The 1.0 micron particles were observed to have an affect on the subsequent pattern when they occurred
adjacent to the patterned lines. The <0.3 micron particles had no observable effect. Both defects were characterized using
optical microscopes and an automated pattern inspection system. After resist strip, the 1.0 micron particles were gone, with
only their effects on the patterning operation being visible. The <0.3 micron defects were still observable after resist strip.
After a subsequent LTO deposition, the <0.3 micron particles appeared to "telescope" in size to about 1 micron due to the
conformal nature of the LTO film. The "telescoped" particles had a noticeable effect on the next patterning operation.
Observation of both particle types using an SEM (scanning electron microscope) showed that the 1.0 micron particles
appeared to be incorporated into the film, whereas the <0.3 micron particles appeared to be under the film. This was
consistent with the defect behavior observed during subsequent processing.
The signal from the large particles suggested contamination from a stainless steel source. Observation of the defect with
an SEM showed that the defects were on top of the deposited film. The defects were found to be coming from the unload
arm of the LPCVD system. The unload arm was occasionally striking another piece of the load/unload assembly, generating
metal particles each time it did this.
The characterization of particle defects from this LPCVD operation resulted in the following monitoring plan: 1) The
alignment of the unload arm was found to be most affected by the preventive maintenance procedure performed on the
load/unload assembly once each week. As a result, a bare silicon particle monitor is run after each PM, before any product
wafers can be run on the system. The monitor is set to look for 6 micron and larger particles with the expectation that no
such particles should be present if the unloader is working properly. 2) The source of the 1.0 micron particles is unknown.
What is known is that these defects are always worse near the pump-end of the tube. As a result, the monitor for this
particle source is run at the pump end of the tube, with a door end monitor run simultaneously as a "control". Different
action limits exist for each monitor. 3) The small particles were found to be very difficult to monitor at the LPCVD operation
since they fell into the "noise" caused by limitations in the particle detection equipment. However, they are easily monitored
in a "look-back" fashion after the subsequent patterning operation using the automated pattern inspection system. As a
result, this defect is monitored at the post-patterning inspection step with action limits initiating feedback to the LPCVD
operation.
MIL-STD-883F
METHOD 5004.11
18 June 2004
21
APPENDIX A
ATTACHMENT 1
Example 2 - Reliability Scenario
:
Characterization of particles at a gate oxide preclean operation showed that the particles contributed by the operation tend
to be small (0.2 microns) and vary in concentration from 0.02 d/cm
2
to 0.8 d/cm
2
depending on how heavily the station is
utilized. Defect density increased as the number of wafers processed through the station increased.
Wafers from this operation were selected such that some of them had low defect densities (approximately 0.3 d/cm
2
) and
the remainder had high defect densities (approximately 0.8 d/cm
2
). These wafers were processed through the line and the
die from these wafers subjected to high voltage stress testing. The results of the tests were that the low and moderate
defect density groups showed levels of gate leakage consistent with the historical process baseline. The high defect density
die show gate leakage that was 3 times that of the historical baseline and resulted in barely acceptable failure rates.
As a result of this characterization, a particle monitor was implemented at gate oxide preclean with an upper limit of 0.6
d/cm
2
to allow some safety margin from the gate leakage problems seen at 0.8 d/cm
2
. However, due to resource limitations,
this monitor can only be run once every shift (approximately every 12 hours). It is likely that the movement of material in the
line will lead to the station occasionally exceeding its control limits between monitors. A second preclean station is
scheduled to be installed in about three months. This station will provide enough capacity to prevent wafer-volume related
out-of-control particle conditions at the gate preclean operation. In order to ensure that no material with bad gate oxide is
shipped during the interim period (before the new station comes on-line), a manufacturer imposed screen (high-voltage
stress test) is used on all material processed between a failing monitor and the last known good monitor at this operation.
In order to show that the screen is effective, particle monitors are processed through the station with every lot of wafers.
This test is done for a period of time sufficient to yield multiple lots at various defect densities. Die from each of these lots
are processed through the high-voltage screen. The results show that the screen is 100 percent effective at detecting the
lots with defect densities greater that 0.6 d/cm
2
. The results show a solid correlation between gate oxide preclean defect
densities and gate oxide leakage levels. The screen is then used to augment station particle level data and remains in place
until the second station is installed and qualified.