of the destructive wirebond pull test and the ball-bond shear test. Figure 4.16 illustrates the improvement that can be achieved in the strength of the ball-bond pad interface by using the ball shear test (instead of the wirebond pull test) to optimize the bonding machine parameters [11]. This particular sample set was thermosonically bonded gold wire on aluminum metallized silicon.
0 10 20 30 40 50 60 70 80
0 2 4 6 8 10 12 14
Pull Strength, grams (force)
Number of Occurrences
NDPT, Mean: 8.74, SD: 1.62, n: 579
(a)
0 10 20 30 40 50 60 70 80
0 2 4 6 8 10 12 14
Pull Strength, grams (force)
Number of Occurrences
No NDPT, Mean: 8.37, SD: 1.93, n:
714
(b)
Fig. 4.15 Wire Bond pull strength histograms for 25mm diameter thermosonically bonded gold wire on gold thin-film metallization on highly polished alumina ceramic. (a) after burn-in without NDPT; (b) after burn-in with NDPT applied post bonding. NDPT limit was 3 g (force)
As mentioned above, the most common gauge of wirebond strength, and hence quality has been mechanical testing, i.e., the wirebond pull test and the ball-bond shear test. Improvements in wirebond technology have caused both tests to have limitations. The pull test requires a hook to be placed under a wire, which is very difficult in situations where the wires are closely space without damaging adjacent wires. For successful NDPT, wires should be spaced at least two to three hook lengths apart. There is also the difficulty of applying a consistent force to the bond Table 4.7 A comparison of areas of applicability between the wirebond pull test (ASTM Standard Test Method F458-84), and the ball bond shear test (ASTM Standard Test Method F1269-89)
Area of Applicability Wirebond Pull Test Ball Bond Shear Test
Module geometry Yes No
Wirebond geometry Yes No
Wire quality, defects, etc. Yes Noa
Second Bond Yesb No
Bonding machine set-up, optimization, etc. Noc Yes
Process development Noc Yes
Substrate, bonding pad quality Noc Yes
aSensitivity to contamination, insensitive to mechanical defects
bExtremely dependent on geometry
cInsensitive unless the effect is catastrophic
0 5 10 15 20 25
9.5 15.5 21.5 27.5 33.5 39.5 45.5 51.5 57.5 63.5 Shear Strength, grams (force)
Number of Occurrences
Mean: 30.60, SD: 6.06, n: 175 Mean: 45.72, SD: 5.28, n: 164
A B
Fig. 4.16 Histograms of gold thermosonic ball bond shear strengths for bonds placed on aluminum metallization (over silicon). Histogram A are the shear test results after the bonding machine was set up using the wirebond pull test. Histogram B are the shear test results after the bonding machine was optimized using the ball shear test
interface, since the tensile and shear forces on the bond vary with the wire length and the hook position along the wire [38]. The ball shear test requires that a ram (wedge-shaped tool with a flat or slightly curved face) be placed on the major diameter of the ball. If the ball is low profile or flat such as those encountered in fine pitch wire bonding (Section 4.9), or thermocompression wire bonding, the ram can easily ride up over the ball. With close spaced bonds (40–50mm or less separation) the ram can run into adjacent bonds causing damage.
Non-destructive ball shear (NDBS) is possible in direct analogy with NDPT.
In NDBS the ram loads the ball to a preset value (nominally a fraction of the envision shearing strength of the device or system under test) and if a failure does not occur the ram is retracted and moved to the next ball bond. Experiments [12]
have shown that NDBS does not affect ultimate destructive shear value, at least with balls formed from 25.4 mm diameter wire with reasonable ball diameter (greater than or equal to 2.5D) There is information in the ASTM Standard (F1269-06) that can be used to set the NDBS limit for various bonding situations.
Other studies have shown that the ball can be loaded non-destructively up to 50–60% of its shearing strength without influencing the destructive shear.
Mechanical testing also tends to be time consuming and more importantly destructive. Even in non-destructive modes (see above) wires are deformed and ball edges flatten in the case of non-destructive shear testing [11], thus giving rise to concerns about future product reliability. Hence most people recommend the mechanical testing of product on a lot sample basis only and, of course for the set-up of wire bonding machines.
A new method for wirebond testing has been developed to address the mechanical test limitations [73]. The technique uses a laser to generate an ultrasonic pulse which is passed through the bond interface and detected nearby. The test is non-destructive, fast, and appears to detect bond interface anomalies. The ultrasonic wave train is thermoelastically generated by a sub- nanosecond laser pulse hitting the top of the ball or wedge bond. The ultrasonic wave travels through the ball or wedge bond and the bond interface onto the surface of the IC. The ultrasonic wave is then detected on the surface of the integrated circuit by a laser interferometer that measures changes in the surface height. This surface displacement versus time data is then numerically con- verted to power versus frequency data, or Power Spectral Density (PSD). The laser ultrasonic bond testing has several potential advantages over the standard mechanical tests: (1) it is non-contact and (2) it is non-destructive. All devices produced can be tested, so quality data does not have to be inferred from a lot sample. In addition, the equipment is controlled by computer so the potential exists to fully implement the test for high production rates when attached to a wirebonder for real-time bond assessment.
A schematic representation of the test configuration is shown in Fig. 4.17.
Figure 4.18 presents displacement versus time curves recorded by the inter- ferometric detection system. The numerical analysis results for a representa- tive sample (bond aged 48 h at 2508C) are shown in Fig. 4.19. The dotted spectrum is the result of applying standard Fast Fourier Transform analysis
methods to the displacement versus time curve to extract the PSD. Further analysis using an autoregressive covariance-based technique produced the solid line shown in Fig. 4.19. The covariance method clearly shows a reso- nance response at 14.5 MHz. Applying this method to the other samples produced the data shown in Table 4.8. Table 4.8 presents the fundamental peak frequency and power levels for the aged samples along with shear strength data from bonds of the same population. Details of these results along with complete description of the method can be found in the paper by Romenesko, et al., [73].
½ Wave Plate
½ Wave Plate Laser
Trap Photo Detector Calibration
Video Camera
Detector Laser
Focus Lens
Beam Expander
Photo Detector
Focus Lens
¼ Wave Plate
Photo Detector Polarized
Beam Splitter Reflector
Objective Lens
Part Transport
Stage Beam
Splitter Micrometer Adjustment
Pulse Laser
Reflector
Piezo-Electric Oscillator
(a)
Microchip
Detector LaserPulse
Laser Bond Wire Ball
Bond Bond
Pad
(b) (c)
Fig. 4.17 Laser-induced ultrasonic energy wirebond evaluation system. (a) optical system schematic; (b) schematic representation of placement of the excitation and detection laser beams relative to the wirebond; (c) a photomicrograph showing the location of the excitation laser (cross hairs on top of ball bond) and the detection laser (white dot on right)
The laser ultrasonic bond evaluation has correlated a shift in the ultrasonic frequency spectrum with both bond aging and intermetallic growth. The ultra- sonic wave detected was shown to be a true surface wave and thus, non-dispersive in nature. Results proving the ultrasonic wave is a surface wave are given in
Fig. 4.19 Comparison of the power spectral density (PSD) resulting from the fast fourier transform (FFT) and the auto regressive (covariance based) numerical methods
0.000 0.005 0.010 0.015 0.020 0.025 0.030
100 300 500 700 900
Time in nanoseconds
Amplitude
Surface-No Bond
As Bonded
Aged 96 hrs @ 200°C
Aged 48 hrs @ 250°C
Fig. 4.18 Displacement amplitude vs. time for bonds with different aging conditions. ‘‘No bond’’ illustrates noise level after bond pad surface is pulsed with a laser. Traces represent averages of at least seven individual trials and have been offset in amplitude for clarity
Fig. 4.20. This means that the detected frequency shifts cannot be attributable to spectral changes due to dispersion as the detection point is moved farther away from the bond pad. In addition, no significant directional dependence of the spectrum was found – again indicating that the measurements are insensitive to the detector location relative to the crystal axes of the semiconductor.
Table 4.8 The effect of thermal aging on the power spectral density (PSD) behavior of typical thermosonic ball bonds made to various metallizations on silicon
Sample Frequency, MHz Power, dB Shear strength* grams (force) Sample 1: A1-1%Si
Substrate 18.5 56.0
As bonded 16.5 39.0 51.71.8
Aged: 96 h @ 2008C 13.5 44.5 60.72.6
Sample 2: A1-1%+0.5%Cu
Substrate 19.5 57.0
As bonded 16.5 45.5 54.32.5
Aged: 48 h @ 2508C 14.5 46.5 57.62.2
*Shear strength obtained from other samples in the same sample population.
Distance (mil)
Fig. 4.20 Results of arrival time measurement with distance. Waveforms are arranged on edge and spaced by the distance to the detector, showing arrival time to be linear with distance.
Vertical axis is displacement amplitude