Hi-Rel Laboratories
Report Number FR-32271
April 30, 2002
Part Number RT54SX16
Submitted By: John R. Devaney
Approved By: Trevor A. Devaney
Background
Four samples of a Field Programmable Gate Array (FPGA) manufactured by Actel were submitted for evaluation of the ball bond interface to the die pad metallization.Previous analytical efforts at Goddard, reported in Q20101EV and Q20024FA,had noted various anomalous conditions with respect to the gold ball to die pad metallization reactions.
These anomalies were specifically:
- discoloration of the gold bond surface due to thin films of organics bleeding out of the grain boundaries of the ball
- intermetallic halos around the bonds
- voiding of the pad aluminum adjacent to the bonds and
- bond lifts experienced at bond pull testing.
Subsequent evaluations of other types and date codes of Actel devices found that the problem, bond lifts, was unique to the RT54SX16 devices but was experienced in only the 9937 and 9901 date codes not the 0031 dated coded devices.Not all Actel devices utilize gold bond wire,some are wire bonded with aluminum preventing the formation of Au-Al intermetallics at the die pad interface.
The investigation at Hi-Rel Laboratories was specifically to characterize the intermetallics in the bond to pad metal interface.
Analysis Methodology
To characterize the bonds a combination of Bond Pull Testing ,Scanning Electron Microscopy, Metallography and Energy Dispersive X-ray Spectroscopy were utilized.
Sample #20-RT54SX16-date code 9901 which had exhibited bond lifts during testing at Goddard as a reference "bad " sample and Sample #14-RT54SX32-date code 0001 which did not exhibit bond lifts as a reference "good " sample..
Sample #20 Analysis:
Bond Pull per Mil Std 883-Method 2011.7 was performed on all the wires on one side of the die. Twenty-eight (28)wires were pulled with lifts occurring on eleven (11)of the 28.There was no clear cut distribution of failures to upper or lower package bond pads.All the lifts occurred at the die bond pad interface but none failed the minimum pull strength of 2.5 grams. Typical lift values were about 3.5 grams versus a wire break value of 6 grams.
SEM examination of the bond pad lift sites and the underside of the lifted gold balls was performed and typical micrographs obtained prior to metallographic cross-sectioning of intact bonds.
SEM examination of the lift sites found that the residue on the bond pad surface was composed of gold-aluminum intermetallics, Figure 1A. Examination of the underside (mating)surface of the gold ball found a cavity in the ball which replicated the intermetallic residue on the bond pad,Figure 1B.
Sample #14
Bond Pull per Mil Std 883-Method 2011.7 was performed on 28 wires and none of the bonds lifted. To further characterize the strength of the interface, ten wires were subjected to a bond "pluck" test with an Ophthalmic scalpel. None of the balls separated cleanly from the pad at the intermetallic interface! In all instances a portion of the ball gold was attached to the to the intermetallic residue on the pad although in some bonds this was relatively isolated and spotty. In several other pads, the pad metal pulled away from the die surface with the gold ball. SEM examination and documentation of the pad "pluck " sites was performed prior to cross-sectioning of other intact bonds along the same side of the die.
Evaluation of the data obtained from examination of the "pluck " site and cross-sections found several features which were different from the #20 device.
From an external point of view, the bond on the good device #14 "looks" the same as on the "bad" device, #20, i.e.both show a well developed intermetallic halo forming around the edge of the ball. After plucking, separation occurred between the ball and the intermetallic for a short distance in from the perimeter of the halo on the good device but this terminated where gold from the ball was still attached to surface of the intermetallic indicating crack formation had progressed no further across the bond zone. This was quite different than on the bad device where the gold ball separated cleanly from the intermetallic residue on the pad surface.
CONCLUSION
Analysis of this type failure over the last 40 years has found that failures of this type can,after a lengthy scientific exercise consisting of many samples,be traced to trace level contaminants on the surface of the aluminum film which interfere with the uniform formation of intermetallic phases.These contaminants can be in the part per million range or films in the angstroms in thickness. These are both beneath the detection limits of energy dispersive spectrometry.
Cross-sections of the samples indicate that both good and bad devices exhibited about the same degree of intermetallic thickness which in itself indicates they both have experienced similar thermal histories. Therefore the failures must be attributed to trace level contaminants on the pad surface at the time of bonding.This is supported by the fact that other lot date codes and similar product lines did not experience this mechanism.An additional factor which may exacerbate the problem could be the presence of the large number of metal 3 to metal 2 vias in the bond pad metal.These could act as traps for contaminants during processing.
From a reliability perspective, these parts are at risk since the Kirkendall void formation and subsequent weakening of the bonds will only progress with time and temperature.
Figure: 1A
Sample: #20-date code 9901
Subject: SEM photo of typical pad lift site. Note the intermetallics which have formed in the probe mark metal are similar to those on the undamaged pad metal which indicates that the problem is not associated with surface contamination.
Figure: 1B
Sample: #20-date code 9901
Subject: SEM photo of ball surface which mates to the pad residue shown in Figure 1A.Halo seen in SEM "top" shots of intact bonds extends beyond region identified as void interface at edge of intermetallics.
Subject: Optical micrograph of a cross-sectioned "good" pulled bond shows both the intermetallic and complex pad structure consisting of vias beneath Metal 3 which connect to Metal 2. Note that the intermetallics appear to have consumed the aluminum in the vias.
Subject: SEM micrograph of a sectioned bond pad in the same row as the one shown in Figure 2.A.The various structures are identified. Note the minimal aluminum in the vias almost exposing the titanium nitride layer on top of Metal 2.
Subject: SEM of bond cross-section. Note the crack developing between the pad intermetallics and the ball gold. This bond was pulled and the wire broke so this bond has been stressed to about 6 grams.
Figure: 3B
Sample: #20 -date code 9901
Subject: Close-up of the right hand side of the bond shown in Figure 3A. This image clearly shows the intermetallic reaction into the vias and the separation crack between two different intermetallic phases, both gold rich.
Figure: 4A
Sample: #20 -date code 9901
Subject: Compare to bond image in Figure 3B. This is the same region but in the SEM Backscatter mode which accentuates differences in atomic density, i.e.gold @79 or tungsten @74 will be much brighter than either aluminum @13 or silicon @14 or silicon dioxide.
Figure: 4B
Sample: #20 -date code 9901
Subject: Backscatter image of the right hand side of the bond shown in Figure 4A with the various regions identified. Note that the crack is developing between two different gold rich intermetallics.Note also the gold has reacted down the sidewalls to the bottom of the vias.
Figure: 5A
Sample: #20 -date code 9901
Subject: SEM image of another typical bond. This bond has also been pulled to wire break but there is no evidence of separation in the bond zone although a void line is forming.
Figure: 5B
Sample: #20 -date code 9901
Subject: Close-up of the right side of the bond cross-section shown in Figure 5A. Various regions and structures have been annotated.
Figure: 6A
Sample: #14 -date code 0001
Subject: SEM image of "plucked" bond. Note that gold has pulled out of the ball and is attached to the intermetallic thru the entire center section of the bond.
Figure: 6B
Sample: #14 -date code 0001
Subject: Close-up of the lower center portion of the bond pad shown in Figure 6A. Note region "intermetallics surface" where separation occurred and no unreacted gold is attached.
Figure: 7A
Sample: #14 -date code 0001
Subject: SEM image of another "plucked" bond. Note that the un-reacted gold separated from the intermetallics surface over the upper region of the interface.
Figure: 7B
Sample: #14 -date code 0001
Subject: Enlargement of the upper region of the bond footprint shown in Figure 7A. Note the island of isolated gold adhering to the intermetallics in the upper right corner of the picture which indicates the incomplete void formation between intermetallics
Figure: 7C
Sample: #14 -date code 0001
Subject: Extreme enlargement of the surface of the intermetallics showing minute regions of ductile rupture.
Figure: 8A
Sample: #14 -date code 0001
Subject: Plucked bond, in this instance the scalpel tip applied sufficient force to rip a portion of the pad metal off of the die surface. Voiding has occurred around the bond periphery but the center section was still intact.:
Figure: 8B
Sample: #14 -date code 0001
Subject: Pad from which the ball in figure 8A was plucked. Note that the center section of the bond reaction zone was not voided retaining sufficient strength to cause the pad metal to lift with the bond.
Figure: 8C
Sample: #14 -date code 0001
Subject: Note along the left side of the bond periphery,voiding has occurred allowing the bond to lift away from the intermetallic surface but over 80%of the bond zone has remained intact.
Figure: 9A
Sample: #20 -date code 9901
Subject: Cross-section of another intact bond on sample #20. Note the non-uniform and erratic intermetallic reaction across the entire bond interface.
Figure: 9B
Sample: #14 -date code 0001
Subject: Another intact bond on sample #14. Note that a peripheral void has formed as well as a crack has initiated on the right side of the bond but the center section is fully reacted and uniform as compared to the bond shown in Figure 9A.
Figure: 10A
Sample: #20 -date code 9901
Subject: Enlargement of left side of bond shown in Figure 9A. Note non-uniformity of intermetallics in reaction zone and line of voids forming between two different intermetallic phases.
Figure: 10B
Sample: #14 -date code 0001
Subject: Enlargement of left side of bond shown in Figure 9B. Note that aside from the peripheral void at the bond-halo interface the inter-metallics in the bond zone proper are uniform combination of gold rich phases.
Figure: 11A
Sample: #20 -date code 9901
Subject: EDS analysis was performed at 5KV in the standardless quantitative mode for the various reaction regions in a bond with a crack developing at the right edge. The various intermetallics present are indicated in the photograph. No AuAl2 (purple phase)was detected.
Home
Last Revised: May 07, 2002
Digital Engineering Institute
Web Grunt: Richard Katz