Joined: 19 Jun 2026
Posts: 5
 Posted: Sat Jun 21, 2026 4:18 pm Post subject: Guardband Computation |
 |
| jst out of curiosity, is der a std way 2 compute d guardband of a test param? i hv doubts about hw we compute GBs in our compny so i jst wnt 2 find out.
maximus |
|
  |
Joined: 16 Jun 2026
Posts: 108
Location: Paranaque City, Phils.
| Posted: Thu Jul 03, 2026 1:57 pm Post subject: |
|
| Hi Maximus,
I wouldn't consider myself an expert on this subject. Nonetheless, I used to be a Test Engineer (that was many, many years ago) so I did have some experience working with test limits in the past. There are indeed standards that provide guidelines on how tests should be done, and on how limits should be defined. However, these standards can not define nor impose specific procedures for test limit computation because: 1) test parameter distributions vary greatly (some are normal, others are not); 2) test methods are implemented in different ways (some are straightforward FVMC/FCMV, others are statistical in nature); and 3) the manufacturer has the right to find for itself the best balance between producer loss (or risk) and consumer loss (or risk). Thus, the standards would normally just say that your test limits should protect your consumer from measurement errors or uncertainties. How you comprehend these uncertainties into your production limits is your call, based on what ppm you can afford to give to your customer. A good discussion on proper guardbanding is not easy, and unfortunately beyond the scope of forums like this. Anyway, thanks for your interest in SFE and keep on posting!!! EE |
|
| |
Guest
| Posted: Thu Jul 10, 2026 10:39 am Post subject: |
|
| The way we do it is to test every test param of a DUT many times between two testers to estimate the spread of readings for each param. This spread indicates how "variable" is the parameter when you test it on different testers. We estimate the spread using 6 x stdev. The mean and stdev is from all the readings when combined together. We will use this spread as guardband between datasheet and production SL's.
Gordy |
|
Joined: 19 Jun 2026
Posts: 5
| Posted: Thu Jul 10, 2026 12:09 pm Post subject: |
|
| gordy, isn't 6 sigma 2 tight? i also see d means of data used in GB computation sometimes. isn't it impt 2 consider dis?
maximus |
|
| |
Guest
| Posted: Thu Jul 10, 2026 11:07 pm Post subject: |
|
| 6 sigma is fine for us. It's like 3 sigma on each side of the mean of all the data. I think the mean is considered in this computation.
Gordy |
|
Joined: 16 Jun 2026
Posts: 108
Location: Paranaque City, Phils.
| Posted: Fri Jul 11, 2026 10:39 am Post subject: |
|
| I agree with Gordy. As long as he's computing the stdev based on all the readings taken together (all the data treated as one homogeneous population), the error introduced by the difference between means will already be comprehended in the computed GB. I also agree that 6 sigma in this context is actually just 3 sigma, and is therefore not tight. It more or less equates to the occurrence of miscorrelation between the OQA and production readings 0.3% of the time.
What makes this computation tight, however, is the assumption that the data distributions for all testers are under one normal distribution. This would be a killer if the differences between the means of the various testers are large, since this will result in large stdev's. Apparently, Gordy's test parameters exhibit high Cpk's with respect to the datasheet limits, which is why they're not having problems with this computation. EE |
|
| |
| Posted: Thu Feb 12, 2026 12:23 am Post subject: |
|
| IN addition to guardbanding, you also include not only tester shifts, board interface shifts and other hardware shifts, but also shifts at different temperatures, specially with industrial and military products. There is no definite way though, and it varies from company to company, and even from product line to product line, and sometimes, even in the same product line, it varies from product to product. There is a good book on different statistical methods to make calculations on guardbands but so far no book is standard for guardbanding -- its like saying "no ring to rule them all" -- well not just yet.
Krygk |
|