MIL-STD-883 Method 2011 sets the minimum pull strength for 25µm gold wire at 3 grams. An 8g population and a 12g population both pass. They do not predict the same things — and treating them as equivalent is how clean-looking pull numbers still send field returns.
Pull strength is often used as a binary gate: the number comes back above the floor, the bond ships. But the mean alone does not describe the process. The distribution shape and the failure mode carry information the mean cannot, and two populations with the same mean can represent very different process states.
What a lower average usually signals
An 8g mean on 25µm gold is typically one of three things. Bond force or ultrasonic energy is running at the low end of the process window — the bonds are forming, but without the full weld area a properly tuned cycle produces. Or the process is drifting, the mean has not crossed the floor yet, but the low-side tail is already growing. Or there is a pad-side surface condition — oxide, contamination, or a thin gold layer — showing up as a softer interface that the wire is bonding to rather than through. Any of those produces a population that passes today and presents at a field return in six to twelve months.
Why the failure mode matters as much as the number
A 12g mean looks healthy on a report. But a 12g mean with a long low-side tail is hiding the same upstream problems as an 8g run — the mean just covers for it. The tail is where the failures live. Break-mode analysis tells the part of the story that the number omits:
- Heel break at the wire. The wire is the weak link. This is the expected failure mode on a correctly bonded part — the bond foot is stronger than the wire body.
- Neck break with intermetallic exposed at the foot. The bond foot is the weak link. A 12g result here masks a bond quality problem; the number passes while the failure mode flags the root cause.
- Lift at the pad interface. The pad surface is the weak link. Cleaning, metallization, or surface preparation is the upstream cause, not the bond cycle itself.
- Wide sigma on an otherwise-passing mean. Process drift in progress — the distribution is spreading before the mean crosses the floor, and the next characterization run will show it.
Logging what the single number hides
Distribution shape and break location on every characterization run — not just minimum and mean — gives a complete picture of where the bond cycle is and where it is headed. When a customer asks what a typical pull strength looks like, the honest answer is a histogram and a break-mode split, not a number. The histogram shows where the population sits, whether it is shifting, and whether the mean is masking a tail that is already out of tolerance.
If your team is debugging recurring field returns and the pull numbers all came back in spec, the population shape and the break-mode distribution are usually where the signal is. A passing mean confirms the process has not crossed the floor. It does not confirm the process is healthy.