Wire Bond Lifts due to Contaminated Bond Pads
August 15, 2002
Failure analysis was performed on two RT54SX16-1CQ208B Field Programmable Gate Array (FPGA) devices, D/C 9937, after one of the devices exceeded the manufacturer’s recommended limits for fall time. Bond lifts, intermetallic "halo's" and dark streaks on the bond wire were found. Scanning Electron Microscope (SEM) inspection of the bond wires showed that the dark streaks followed the contour of the gold wires and ball bonds appearing like watermelon stripes. The streaks could have been caused by contamination in the capillary tube during wire bonding, and transferred to the gold bulk material during ‘flame-off’. Auger Electron Scans (AES) found the presence of oxygen and carbon in the areas where the gold is streaked, indicating hydrocarbon contamination.
The 1 mil gold wires used to make the bonds were pulled on one side of the upper tier of a two-tiered ceramic package. The average pull strength was 4.84 grams-force, with a low of 2.6 grams-force. Eighteen of twenty-eight wires pulled lifted at the ball/pad interface.
Further inspection of the intermetallic on the bond pad showed the presence of titanium. The vias under the aluminum bond pads are comprised of tungsten and titanium and are called plugs. In crossection, these plugs showed depletion of metal at their centers (Pic 1-1). It is not currently known if the depletion of the metal is due to intermetallic formation, or if it occurred during wafer processing. These hollow plugs may be the cause of the “dippled” appearance of the bond pad surface. These dimples can retain contamination if they are not adequately cleaned leading to accelerated growth of intermetallics and the creation of “Horsting Voids". Focused Ion Beam (FIB) cross-sections and AES analysis confirmed the presence of carbon-bearing contamination in the wells of the bond pad.
Thirty-three devices from the same manufacturer, representing approximately 26 wafer lots and 24 packaging lots were tested for bond strength. The lot date codes were between 9543 and 0151. Only the parts with lot date codes 9849, 9901 and 9937 had bond lifts. We do not know yet when this bond pad design came into use by this manufacturer however it was still found to be used on parts made in 2001 and packaged in 2002 (See the complete report at Q20101EV.doc)
Pic 1-1: Crossection of the ball bond with intermetallic consuming all of the gold that welded into the plug cavities and forming a halo of intermetallic material around the periphery of the bond (thereby filling adjacent plug cavities). The arrow on the left points to a line of voids at the diffusion front. The arrow to the right points to the intermetallic halo.
Pic 1-2: Increased magnification of a bond pad dimple nearly completely filled with intermetallic.
A burn-in study was done to look at two additional lots which were representative of devices intended for use on flight hardware. Fifteen FPGA microcircuits, consisting of five groups of three parts with different date codes, were used in this study. One group was placed aside as a control set, and did not receive any burn-in testing. The other four groups were placed in a burn-in oven at ambient atmosphere and +150°C. The scheduled and actual burn-in times for each of the five groups are shown in the Table 1. on the right.
Group Scheduled Actual
Group 1 0 hrs 0 hrs
Group 2 168 hrs 162 hrs
Group 3 336 hrs 328 hrs
Group 4 504 hrs 523 hrs
Group 5 1000 hrs 976 hrs
Each test group in Table 1. above consisted of three parts, each with the same three date codes. The parts were serialized for the test with the following suffixes:
Suffix P/N Date Code
A RT54SX16 9849*
B RT54SX32S 0151
C RT54SX72S 0151
*note that this part is from one of the lots which had bond lifts in the prior examination: known bad*
The study resulted with increasing numbers of bond lifts over burn-in time (even ignoring the 3A unit in Group 3, see Table 3 below). The lowest bond strength measured was 0.6 g which would not result in a catastrophic failure in space. A worst case acceleration factor of 70 was calculated from an assumed operating temperature of 50 degrees C, and a thermal activation energy between aluminum and gold of 0.5 eV. The burn-in duration of 976 hours results in a a useful life of 68,320 hours or 7.8 years.
It is interesting to note that in this case, it was not possible to predict bond strength from the physical appearance of the ball bond. All of the balls in the "A" devices appeared to be undercompressed while all of the balls in the "B"and "C" devices were overcompressed. All of the bonds were accompanied by intermetallic halo's. (See the complete report at Q20166TEST.doc)
Group S/N P/N D/C No. Wires No. Lifts
1 1A RT54SX16 9849 171 0
1 1B RT54SX32S 0151 215 0
1 1C RT54SX72S 0151 145 1
2 2A RT54SX16 9849 171 0
2 2B RT54SX32S 0151 215 0
2 2C RT54SX72S 0151 145 1
3 3A RT54SX16 9849 172 23
3 3B RT54SX32S 0151 177 0
3 3C RT54SX72S 0151 145 0
4 4A RT54SX16 9849 171 2
4 4B RT54SX32S 0151 215 0
4 4C RT54SX72S 0151 145 1
5 5A RT54SX16 9849 171 4
5 5B RT54SX32S 0151 215 1
5 5C RT54SX72S 0151 145 0