ùso, you asked for it!
Personally, and from experience, I rather assure solder joint reliability
using an adequate modeling approach, taking into account both the
physics-of-failure and the statistical failure distribution. The reason is
that it is no more error-prone than testing and much less expensive and
time-consuming. The time and cost factors of testing has led to test
short-cut, with sometimes catastrophic consequences.
Using a 'Design for Reliability (DfR)'-approach, of course, requires a
reasonably good model. For near-eutectic Sn/Pb solders and their derivatives,
we have the data on which a number of modeling approaches used in the
industry are based. But for many of the newer soldering alloys, including all
of the lead-free solders, we do not have sufficient data to determine a
fatigue reliability model. Thus, until such time, any modeling requires
assuming that the solder in question behaves similar to Sn/Pb and multiplying
some safety (actually ignorance) factor (>=2xlife to acceptable failure
probability).
For the new solders, side-by side cyclic testing with eutectic (or 60/40)
Sn/Pb solder from -20<->+100C (125C if substrate glass transition temperature
is at least 150C) with 15 minute dwells at each temperature extreme (24
cycles/day) with conventional chambers (or 5 minute dwells with chambers
modified to provide temperature uniformity in the whole test volume; ~100
cycles/day) with at least 32 equal continuity daisy-chains monitored with an
Anatech Event Detector for both solders, is necessary. This will give an
adequate data base to draw valid comparisons with near-eutectic Sn/Pb
solders, and, to me more importantly, from which a fatigue reliability model
can be derived.
Tests that are sufficient to ensure "reasonable" reliability will be totally
dependent on the product design together with its application. But to design
such a test, you have to have an appropriate fatigue reliability model from
which a valid acceleration factor for the test vis-a-vis the use conditions
can be determined. Electronic application vary from easily met reliability
reqirements for consumer products (~1,000 cycles (3 yrs) @ delta-T's ~10C &
10% acceltable failure probability) to difficult to meet requirements for
low-earth-orbit satellites (~90,000 cycles (10 yrs) @delta-T's ~35C & 0.001%)
or automotive-under-hood equipment (~10,000 cycles (~3 yrs) @ delta-T's ~100C
& 0.1%).

Werner Engelmaier
Engelmaier Associates, L.C.
Electronic Packaging, Interconnection and Reliability Consulting
7 Jasmine Run
Ormond Beach, FL  32174  USA
Phone: 904-437-8747, Fax: 904-437-8737
E-mail: [log in to unmask], Website: www.engelmaier.com