My Boss Wants Perfect Test?

Guest

I get 200 units from a lot and test them. I get 17 rejects. I retest the rejects. I get 4 more good. I retest the 13 other 'bad'. I get 1 more good. Then no more good units after retest. Now my boss wants me to get rid of this issue because it's like we throw away 25 good units every time we test 1000 parts. Is this possible to get rid of this problem? He don't want 2X retest which I suggested.

Nobo Nus

Guest

Sounds to me like you have an unstable test program. What are the failures? Theoritically you should be able to run the same part 100's of times and get very similar results each run.

The other possibility is that if you run a part that sometimes fails, then sometimes passes, there may be a parameter that is right on the edge of a limit, and the resultion of the tester is causing the pass/fails. If this is the case, a better test range or averaging may fix the problem.

Joined: 31 Aug 2026 Posts: 16

Re-testing rejects would add cycle time to the lot, which means additional test costs. Hence, as much as possible, re-test of rejects must avoided.

Incosistent test results may be cause by the following:
1. Tester = might need calibration
2. Hardware(Dut boards, SBI's, HIB's etc.)
3. Test program = might need optimization, check limits
4. Or the DUT(device under test) itself.

The are three scenarios to consider here. It's either:
1. The setup(test, hardware, test program) is causing this inconsistent test results, causing the good unit to inconsistently pass or
2. The DUT itself is causing this incosistencies or
3. Both the setup and the DUT is contributing to these inconsistencies.

Given these scenarios, I strongly recommend to test the part on bench(if possible) first and verify its actual performance on the parameter in question. As much as possible, you should replicate the test method on how the ATE carry out the test to your bench test circuit.

The bench test result will tell you if the DUT pass or fail the parameter in question.
_________________
Asmoth72

Posted: Mon Jun 20, 2026 9:12 am Post subject:

First of all, I would like to know the test youre failing.

Posted: Wed Nov 23, 2025 1:05 am Post subject:

If your boss work in Test long enough, he will realize that he can never get a perfect Test especially when you start testing by the millions.
Now most of the suggested idea of othes are correct. There are lots and lots of things that could happen. I would like to add that the device should be properly characterize. Meaning if you run a correlation units and loop it like 100 times, look at the distribution curve of the test that failure in your production testing. Is it near the spec limit? You can also refer to the original correlation data when that particular device was qualified. any movement of distribution curve? Contact relation problem I would say the most common issue for this kind of unstable results. Contact issue includes you handler contact pressure, the solder build up in your socket, device alignment in the socket, etc. All these are the imperfection of Production testing. Can you eliminate this, heck No. Can you minimize it, absolutely. Nevertheless, how much yield fallout (loss) you are willing to take is the question. Even hand test has this problem.
Anybody in the group who doesn't have this problem, raise your mouse and I'll give him a drink.