8.1. Nondestructive Test for Beam-Lead Bonds

Detailed experiments were carried out to evaluate the acoustic emission test (NBS Spec. Publs. 400-12, pp. 31-32; 400-19, pp. 48-50; and 400-25, pp. 46-49) [99] as a nondestrucCtive means for evaluating the quality of beamlead and other types of multiple bonds. The measurement system is capable of detecting acoustic emission signals barely above the average noise level of the preamplifiers and considerably below various system and line transients. Most acoustic emission detector output signals produced in the present experiments were in the range of about 10 to 100 μV and were easily captured by the equipment. Because of the variety of gain adjustments possible (preamplifiers, pulse capturing equipment, and oscilloscope), the vertical scale of most acoustic emission oscillograms has no particular significance. The only important consideration is the signal-to-noise ratio, and this can be easily observed from the photographs. Signal sizes given in the discussion below refer to the detector output.

Pull Tests In order to demonstrate the sensitivity of the method, several beam-lead devices were tested by cutting all but one of the beams and pulling the uncut beam to destruction with an electrolytically etched, 150-um diameter tungsten hook inserted into a dab of silicone rubber on top of the device. Acoustic emission signals picked up by a substrate detector are displayed in figure 52. In the case of a well-bonded beam (fig. 52a), failure occurred by a break at the bond heel. The clipped waveform peaks indicate that the substrate detector output was significantly greater than 1 mV (peak-to-peak) during the initial part of the break. The signal from such breaks generally continues erratically for several times the time interval shown in the figure. In the case of a well-bonded beam with a weak anchor (fig. 52b), failure occurred by peel at the anchor at a force of about 30 mN (1 gf 9.8 mN). The peak-topeak detector output in this case was about 0.3 mV.

=

Pull tests were conducted on several wellbonded devices with weak anchors. A series of short bursts was observed starting at about 15 millinewtons per beam (about half the peel strength of the anchors). Similar well-bonded devices with strong anchors produced no signal until a pull force of about 25 millinewtons per beam was applied; the bursts in this case were longer and higher in amplitude than the devices with weak anchors.

Several groups of commercially available devices from four different manufacturers were tested. Devices from three of these sources produced no acoustic emission signal until they were stressed to about 25 millinewtons per beam as noted above. However, devices from the fourth source produced large bursts of acoustic emission, such as those shown in figure 53a, when they were stressed to only about 10 millinewtons per beam; these bursts increased in number and amplitude with increasing stress. Examination of these devices in a scanning electron microscope following stressing to 10 millinewtons per beam revealed no evidence of mechanical failure. However, examination following stressing at about 25 millinewtons per beam revealed elongation of the beams, separation of the relatively thick titanium layer, peeling of the anchors, separation of the silicon nitride from the beams or silicon, and broken chips of silicon at the anchor as shown in figure 53b. Any of these observed degradations could be responsible for the acoustic emission signals. These problems can be attributed to design deficiencies; they were observed on three different device types from the same source on lots purchased 18 months apart. It should be noted that a normal destructive pull-off [100] or push test (NBS Tech. Note 806, pp. 34-35) would not have revealed any problems with the structures because the beams ultimately broke with forces typical of mechanically strong devices. The consequences of the poor mechanical integrity on device performance or reliability were not established.

In a large series of pull tests on devices bonded to substrates with chrome-inhibited bonding pads (NBS Spec. Publ. 400-25, pp. 4647), it was found that a force of 10 to 15 millinewtons per beam was required to produce acoustic emission from one or two weaklybonded beams on an otherwise well-bonded device. On the other hand, a force of only 5 to 10 millinewtons per bond was often sufficient to produce acoustic emission when all beams were poorly bonded; such devices would pull off completely at forces of 10 to 15 millinewtons per beam.

Push-Down Test on Devices Encapsulated in Silicone Rubber Both well-bonded and poorlybonded devices were encapsulated, except for the top of the chip, with a silicone rubber thinned with xylene [99] and allowed to cure. For testing, the device was mounted on a vacuum chuck and the top of the chip was uniformly pressed downward with an acoustic

inner walls of the shortened sides of the tool are about 25 um larger than the silicon chip on all sides. The rim of the probe was coated with silicone rubber and molded with a deep undercut pattern of the beams as shown in figure 56. In use, the probe is pressed down on top of the beams; the lower portion of the silicone rubber is forced against the substrate and bulges underneath the beam in such a manner as to lift the downward curving portion of the beam. A weak beam will be lifted upward resulting in acoustic emission. If the bugging height is uniform and if the probe is properly aligned to the chip and is perpendicular to the substrate, about 1.5 N can be applied before the device collapses. Most of this force is dissipated by compressing the silicone rubber and only a small amount is actually applied as a lift force to the beams. This probe concept appears to offer the best method of stressing weaklybonded beams. The larger signal appears in the substrate detector, partly because the bond is more closely related to the substrate and partly because the coupling from the beams through the silicone rubber to the probe is relatively poor. However considerably more development is required to increase the operating life of the probe, and to obtain reproducible stressing forces and better coupling to the probe detector. This type of probe is only effective on devices with relatively uniform bugging height and bonds which extend at least 75 to 100 um from the chip. Unlike the other probes employed in these tests, this probe cleanly strips off the free silicon nitride skirt in all areas away from the beam and anchor. The silicon nitride on top of the beams is not disturbed; the effect of the damage which does occur on device performance and reliability has not been determined [99].

Use of Acoustic Emission in Other Areas Experiments were conducted to determine the feasibility of applying this technique to the evaluation of bond integrity on film carrier and reel systems. Tests on three different types of such systems indicated that the acoustic emission technique can be effectively used for this purpose [99]. In one case, a signal was obtained which could be correlated with the separation of a single solder bump. In a second case, several weak bonds in a gang-bonded device produced an emission burst. In a third case, a bond which was mechanically solid despite the fact that it had a poor visual appearance emitted no bursts. Acoustic emission bursts were also observed when a weakly-bonded chip capacitor in a hybrid circuit was subjected to a downward force of about 2 N. Preliminary experiments suggest that acoustic emission can also be used to supplement the nondestructive wire bond pull test, both to assist in determining the maximum force to be applied and to assure the nondestructive nature of the test. (G. G. Harman)

The experimental set-up used was similar to that employed previously [101] except that since only the envelope of the nodal pattern was monitored there was no need for the two phase-delay pulse generators used in the previous work. The tool used in these experiments was a long 0.828-in. (21.0-mm) tungsten carbide tool with a 4-mil (100-μm) foot length. Also, the capacitor microphone was fitted with a stainless steel taper tip having a 25-mil (0.64-mm) diameter hole. This larger than usual taper tip was chosen for two reasons, both of which enhance its usability in an industrial environment. First, it is quite rugged and less easily damaged. Second, it gives an acceptable signal at low gain and hence is less susceptible to picking up any extraneous noise.

In all the experiments reported here, the node of the unloaded tool was found as follows: the tool was brought into contact with the wire to be bonded and the power dial set

ting was turned down to its lowest position (1.0); the power was then turned on and the microphone was raised and lowered along the length of the tool until a minimum signal was obtained. This point was checked occasionally and was found to be fairly stable.

The first series of experiments employed aluminum pads of good bondability. On a single

substrate, two different groups of aluminum bond pairs were fabricated. The first group, the control group, was made using the following machine settings for both the first and second bonds: power setting, 5.0; bonding time, 50 ms; and bonding force, 0.25 N. The control group bonds were pulled to destruction in order to establish that the powertime-force combination used in its fabrication produced acceptable bonds. The predominant failure mode found in the pull test was heel breakage. The overall mean, X, and sample standard deviation, s, were 9.98 and 0.58 gf (97.8 and 5.7 mN), respectively; these values are indicative of a satisfactory bonding process which is under control.

The second group, the test group, was made with the same settings except that the power setting and bonding time for the second bond were set to zero. The second bond of each bond pair in this group was then made while the envelope of the nodal pattern was monitored. The force and time employed were the same as were used in bonding the control group; groups of eight bonds each were made with power settings of 2.0, 5.0, 7.0, and 10.0.

The results of the pull test are given in table 17. The pull strengths of the bonds made with a power setting of 2.0 are indicative of a bonding process which is out of control.

Typical microphone outputs (signatures) are shown in figure 57. The interesting feature of the signature of bonds made with the low

est power setting is the ringing characteristic at the beginning of the bonding cycle and the gradual tapering off of the tool vibration amplitude. The bonds made with higher power settings showed no evidence of ringing; all these bonds failed at the heel. This suggested that the microphone signal may be more sensitive to weld formation rather than to heel weakening due to excess deformation. It appeared likely that the nodal pattern contains only information concerning the quality of the weld. As the power increases, the interfacial weld strength also increases. However, at the same time the bond heel is further deformed which decreases its strength under the application of the pull test (NBS Spec. Publ. 400-19, p. 51).

Other experiments made with the same tooltransducer-power supply combination yielded similar results. When the experiments were repeated with a tool with a 2-mil (50-μm) footlength similar results were also obtained. When the experiments were repeated with different power supply-transducer combinations, however, no ringing was observed, even for the cases where the bonds failed by lift off.

Since the node rises during the loading of the tool during bonding, the vibrational amplitude at various positions below the node of the unloaded tool was studied at various power settings with the original transducerpower supply combination. If there were any ringing in the unloaded tool below the node it was expected to travel up to the position of the node during bonding where it could be detected by the microphone; in this case the ringing phenomenon would be suspected of being an artifact of the particular equipment being used. If ringing were not there, then there would be some reason to believe that ringing is a true signature of the lift-off failure mode. Ringing was observed 12 and 52 mils (0.30 and 1.32 mm) below the mode for a