AN1231 Motorola / Freescale Semiconductor, AN1231 Datasheet - Page 15

• Manufacturer Part Number: AN1231
• Description: Plastic Ball Grid Array (PBGA)
• Manufacturer: Motorola / Freescale Semiconductor

rough first–order estimate and could give very erroneous re-
sults, but it may be used for a lack of any other more in–
depth analysis (such as nonlinear finite element modeling). It
is also prudent to obtain an actual acceleration factor from
two different testing conditions to verify the validity of equa-
tion 8 before its use in predicting field cycles to failure.
parameter have been determined, test data may be extrapo-
lated to determine cycles to failure for a much larger sample
size such as a population in the field. The acceleration factor
is multiplied by the percentage failed in accelerated thermal
cycling to determine the percentage failed in the field as
follows:
ply be calculated by substituting the desired reliability (i.e.,
fraction failed), the shape parameter and the field scale pa-
rameter and solving for time (in cycles) in either equations (1)
or (2) above. It also has to be assumed that the field shape
parameter ( or ) is the same as that calculated from test-
ing. This then assumes that the failure mode is the same in
both the field and during accelerated testing since the shape
parameter is also an indicator of failure mode. For the Wei-
bull distribution, solving equation (1) for time yields:
are desired. It is commonly desirable to know the time at
which 1,000 devices per million (ppm) would be predicted to
fail. This 1,000 ppm corresponds to an R of 0.999 (R = 1 –
1,000/1,000,000=1 – fraction failed). Substituting this R as
well as a previously calculated value of
known
N 0.1% of 4685 cycles. Since the Lognormal equation (2)
cannot readily be solved for time due to its complexity, an
iterative process (with the aid of a spreadsheet that has the
erf function) can be performed to determine the N 0.1% .
MOTOROLA FAST SRAM
Once again, such an equation should be used as a very
After the acceleration factor, scale parameter and shape
Then the time for any percentage to fail in the field can sim-
It must be determined to what reliability cycles to failure
into equation (10) yields a time to fail 1,000 ppm or
N xx%,f = AF
t =
Reliability
Reliability
0.999999
0.99999
0.9999
0.999
0.368
0.99
0.84
(R)
0.9
0.5

Log Normal Distributions (0 to 100 C Thermal Cycling, 20 Minute Cycle)
{– 1n [R(t)]} (1/ )
Table 5. 225 Pin PBGA Reliability Predictions Using the Weibull and

N xx%,ATC
Percentage
Percentage
Failed (%)
0.0001
0.001
0.01
10.0
16.0
50.0
63.2
0.1
1.0
(N 63.2% ) and
Devices Failed
Devices Failed
(10)
Per Million
(9)
100,000
160,000
500,000
632,121
10,000
1000
100
10
1
Alternately, statistical software with Lognormal capabilities
may be used. For this example it was determined to be 5659
cycles. This is slightly higher than what was predicted using
the Weibull distribution. This is usually the case, as the
Weibull distribution is a more conservative predictor than the
Lognormal. However, the Lognormal traditionally results in a
better correlation coefficient. The distribution that is used
should be whatever the user is most comfortable and has the
most experience or history using. Predictions to any given
reliability can likewise be made from the two distributions. To
illustrate this and to further compare the Weibull and
Lognormal distributions, Table 5 shows a range of reliabilities
calculated from using the data in Table 4.
tive cycles to failure values for small percentages of a total
population, data to a desired confidence interval should be
used. In the two reliability plots above (Figures 12 and 13)
this would mean using the lines forming the 90% confidence
interval (or whatever confidence level was desired) as
opposed the scale and shape parameters determined from
the best fit of the data. It is only practical to consider the
lower confidence limit since using the upper limit of the
expected cycles to failure is not useful or prudent for field
failure prediction. Table 6 compares the predicted cycles to
failure from the best fit line versus those predicted using the
upper and lower 95% confidence limits.
PBGA Thermal Cycling Data
119, 225, and 361 pin PBGAs while testing of other configu-
rations is ongoing. Additionally, several other companies
have thermal cycling testing either underway or completed.
Two of those companies are AT&T and Compaq and their
published data, along with a sampling of Motorola data are
presented in Table 7. Also listed are Motorola data on two
leaded devices, the 68 PLCC and the 208 PQFP, both with
copper leadframes. The Motorola PBGA data shown shaded
in Table 7 represents data that was used as example data for
thermal cycling statistics in the previous section.
It should be noted that to extrapolate the most conserva-
Motorola has thermal cycled several configurations of 72,
Predicted Cycles to Failure Using:
Weibull
2758
3291
3926
4685
5592
6696
6960
7737
7958
Log Normal
4794
5035
5317
5659
6098
6739
6925
7628
7878
AN1231
15