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 for example, a single diffusion lot involved or a single wafer. If the lots were produced under total process control, very small variations may result between lots. Then, if future lots will come from the same process, the characterization data is valid. However, if future lots will not all come from the same process, then the data is not valid. 5.2.4.1.6.3 Other systematic effects. If the data comes from several sources, it can vary systematically, i.e., nonrandomly, according to the source of the data. Even in one test facility, there may be a systematic dependence on test cell, operator, dosimetry equipment, etc.

If lots come from several plants, there may be a systematic dependence on which plant supplied the parts. This type of effect should not be common; it is mentioned here to make sure that is not entirely overlooked. 5.2.4.1.6.4 Checking for systematic effects. If data is suspect, it can be checked for self consistency by removing all data having a common attribute such as date, test facility, place of manufacture, operator, etc. If there are no systematic effects, such removal should not seriously perturb the computed results. Likewise, when parameter values or stress to failure are ranked, there should be no grouping by attribute except for lot identity.

Another possibility that should be checked is that the devices did not receive some previous radiation of another type. Thus, as an example, ionizing radiation dose data can be obtained from a data bank, but the devices listed may, previously, have been exposed to neutrons. 5.2.4.2 Correcting the data. If the data examination shows that there may be bad data points present then they have to be either corrected or discarded before the data analysis can proceed. 5.2.4.2.1 Correcting bad data. Sometimes an isolated missing, outlying or otherwise faulty measurement may be inferred by interpolation if the parameter varies smoothly with radiation stress or bias.

Whether or not data should be corrected is often a matter of judgment. This documentation should identify where inferred data was used to replace incorrect or missing data. 5.2.4.2.2 Discarding bad data. Often, only a single measurement on a device has to be corrected or discarded. However, if one device has many outlying or other obvious measurements, it may be necessary to eliminate the entire series of measurements on that device and seek an explanation of what went wrong. It may be a case of a mismeasurement or of the measuring equipment itself having an effect on the device.

On the other hand, the device itself may be an outlier and the question is then raised as to how many outliers will be likely in practice, i.e., when devices are procured for system development and production. If the data shows evidence of many apparent mismeasurements, outlying devices, or outlying lots, then all of the data comes into question. When data is discarded, such action should be documented. 5.2.4.3 Recasting the data to the specified conditions. The assumption is now made that the data examination has been successfully completed and that the data to be analyzed is satisfactory in all respects.

One final task now remains before the actual calculation of the end-point limits can be made and that is that the data must be recast into the conditions that will be specified for the lot acceptance tests. Thus, if the lot acceptance test is to be made on a particular parameter, PAR, at a radiation level PHI(SPEC), then the post-radiation values PAR(PHI(SPEC)) must be obtained from the characterization data and they must be obtained for the bias conditions that will also be specified in the procurement document. In general these values will be obtainable by interpolation.

In the case of abrupt failure, defined in 5.1.8.7 as a failure level which cannot be determined by interpolation, the mean value and standard deviation of the fluences to abrupt failure in the characterization data should be used to estimate the radiation level at which the part can be safely used. 5.2.4.4 Examples of data analyses. Depending on the exact nature of the data, the statistical methods that should be used to analyze the data and to calculate the required end-point limits range from relatively simple techniques to some that are still under development. Examples are given below for two cases for which the calculations are relatively straightforward.

All statistical quantities are assumed to obey lognormal probability distributions i.e., the logarithms of the parameter values are normally distributed. This assumption may be considered reasonable because many of the parameters that are of interest for radiation effects in devices have been observed experimentally to obey a lognormal distribution. In any case, extensions of the analysis techniques discussed below can be made to other types of probability distributions. 5.2.4.4.1 Parameters and delta-parameters. For convenience, the analyses are discussed in terms of the value of the parameter in question after irradiation.

In practice, however, it is often useful to measure the changes, or deltas, in a parameter value produced by the irradiation. Again, the extension from the formulas given to the case of parameter deltas is straightforward. It should be noted that lot acceptance tests based on parameter deltas can be somewhat more costly because they require that part identity be maintained and that pre- and post-irradiation parameter values be recorded. They should not be overlooked, however, because, in REPRINTED WITHOUT CHANGE. 31