Department of Defense MIL-HDBK-816 Notice 1 — MIL-HDBK-816 Notice 1 Radiation Hardness Assurance Guidelines Manual PDF

Downloaded from https://www.everyspec.com MIL-HDBK-816 NOTICE 1 5.1.2 Selection of manufacturers. For a selected part type, the potential manufacturers to supply that part for government procurement should be identified. If the part type is already a JAN type or SMD, the respective QPL or QML will serve as a starting point. DSCC and the preparing activities for the military standard procurement system (SPAWAR, and NASA) should be consulted, however, to learn whether additional vendors might be in the process of qualifying for that part type.

If the part in question is to be supplied to a Standard Microcircuit Drawing (SMD) or other government specification, then the list of vendors will need to be obtained either from DSCC or the cognizant project office. 5.1.3 Selection of random test samples. Sample parts for testing should be selected at random. Because of cost, part availability, schedules, etc., it may not always be practical to obtain an adequate set of samples. The procedures discussed under data analysis in 5.2.4, will still apply but will produce increased uncertainties in the results which will reflect the inadequacy of the sample of parts which was used for the characterization.

One way of saving costs might be the use of parts which have passed group A electrical performance tests and alternate visual criteria; this possibility is mentioned for VLSI parts in particular but could have application whenever the cost of sample parts is a limiting factor. It is always advisable that the test parts have gone through burn-in. 5.1.4 Obtaining samples. Once the vendors have been identified, samples of parts should be obtained from each vendor. For each vendor, the samples need to be taken in random fashion from each lot of parts that is being used and the lots need to be selected so that they represent, as accurately as possible, production characteristics.

The production variations which are of concern are: (a) variations over time, (b) variations from one wafer lot to another, (c) variations from one wafer to another in a wafer lot, and (d) variations of the devices from one region of the wafer to another. A few extra samples should be obtained to cover the possibility that some devices may, in later measurements, prove to be either bad devices or outliers which will need to be replaced. 5.1.4.1 Recommended sample size. To meet the concerns listed in the previous section, the following quantities, for each vendor and for each radiation environment, are recommended: a. Three wafer lots taken at least 1 month apart. b.

Five wafers per wafer lot. c. Five devices per wafer taken one from each quadrant and one from the center. A few extra samples should be obtained to cover the possibility that some devices may, in later measurements, prove to be either bad devices or outliers which will need to be replaced. See 5.2.4.1.1 for a discussion of bad devices or outliers. 5.1.4.2 Reduced sample sizes. It should be understood that, in cases where the recommended quantities prove to be impractical because of cost or schedule impacts, the vendor may propose reduced sampling requirements.

Reduced requirements may be acceptable, for example, where data exists showing the radiation response of the part type in question to be stable and predictable over long periods of time or that the part response will be similar to that of another part. 5.1.4.3 Minimum sample sizes. The sample size recommended in the previous section, for each vendor and radiation environment, is a total of 75 devices. If that number of devices cannot be made available, then reduced sample sizes will have to be considered. Roughly speaking, 25 is the minimum sample size which should be used for a characterization measurement, the more parts the better.

For this minimum sample size, the recommendation is that five different lots be sampled and that five samples per lot be taken. A sample of 25 will at least allow a goodness of fit test to determine whether the parts’ behavior belongs to a well defined statistical distribution. If the relevant parameters degrade gracefully with radiation and if the probability distribution is known well enough, then five or even as few as three parts from each lot might be adequate.

Larger sample sizes will reduce uncertainties about the nature of the governing probability distribution and will thereby permit better performance estimates for the parts remaining in the population from which the sample was drawn. Sample sizes which are too small may result in characterization data with large statistical uncertainties. These uncertainties can result in overly conservative decisions about how the part should be used or purchased and may lead to costs which will more than undo the savings produced by the small sample size. No simple prescription for selecting sample size can be given.

Relevant factors, in addition to part cost, which will affect how large the sample size should be (for each vendor), will include the amount of previous knowledge available about the part and the variability of the parts’ performance on a single wafer, within a wafer lot, and from lot to lot. If lot to lot variability is large compared to variability within one lot (a SUPERSEDES PAGE 16 OF MIL-HDBK-816 16