Page 106 - Wire Bonding in Microelectronics

W ir e Bond Testing 85

The peel test has been used for years and is completely described
in MIL-STD-883G/H, Method 2011.7. This test method is short and
reproduced here to further clarify the above discussion:
“3.1.1: Test condition A—Bond peel. The lead or terminal and the device
package shall be gripped or clamped in such a manner that a peeling
stress is exerted with the specified angle between the lead or terminal
and the board or substrate. Unless otherwise specified, an angle of 90
degrees shall be used. When a failure occurs, the force causing the fail-
ure and the failure category shall be recorded. 3.1.2 Test condition C—
Wire pull (single bond). This test is normally employed for internal
bonds at the die or substrate and the lead frame of microelectronic
devices. The wire connecting the die or substrate shall be cut so as to
provide two ends accessible for pull test. In the case of short wire runs,
it may be necessary to cut the wire close to one termination in order to
allow pull test at the opposite termination. The wire shall be gripped in
a suitable (tweezer-like) device and simple pulling action applied to the
wire or to the device (with the wire clamped) in such a manner that the
force is applied approximately normal to the surface of the die or sub-
strate. When a failure occurs, the force causing the failure and the failure
category shall be recorded.”

4.2.3 Failure Predictions That Are Based on
Pull Test Data Must Have Confirmed Normality
–
Many organizations use the pull test average (x) range (r), and stan-
dard deviation (σ) distribution charts for production control purposes.
The data are plotted continuously giving a running chart for SPC. In
addition, the operators usually record the bond failure modes. The
latter are helpful when bonding problems are encountered. Modern
pull testers can store the failure mode and pull force and can generate
the statistics, which are often coupled directly into a PC or other
larger computer. From these data, predictions are often made as to the
number of pull test (force values below the specified control limit)
failures to be expected each day. LTPD and confidence levels are also
calculated.
One might assume that such predictions of pull-test failures could
be calculated from the distribution curve as defined by the average
and standard deviation of the test values. However, because there are
usually several test failure modes [e.g., bond lift from either end, wire
break in the span, heel break, neck break (above ball), cratering, and
metallization failure], the distribution is often not normal. When this
occurs, such predictions may be invalid. Before attempting to make
such calculations, a test for normality, such as the Chi Square statis-
tic, should be used on the data. Frequently, when using non-normal
data, the predicted number of test failures can appear higher (or lower)
than the number that actually occurs. For example, in one case,