Letter to the Editor
Response to Editorial: No Failures Found
The relationship modeled by T. W. Williams for fault coverage is D = [1-Y (1-T) ] where: D the defect level, Y the theoretical functional process yield and T the Dear Sir fault coverage of the test program used. There are three basic reason(s) for NFFs.
Using this equation, if the fault grade of the test program 1. Almost everyone is familiar with this reason, but it is 70 per cent, the defect level is projected to be 10โข1 per will be elaborated for completeness. Sometimes a techcent. So if a failure occurs in the 10 per cent of the nician who is given the task of getting equipment in the board not covered then it will pass functional and not its field running again as soon as possible will simply replace application. From our experience, manufacturing believe everything which might be the cause. This is good for they have better test coverage then they actually have the customer because he then has his equipment up and without fault grading their test suites. Another consequence running again as soon as possible, however it is very bad of poor quality tests is consumers of electronic products, for the person who will attempt to troubleshoot the who rely on high availability and maintainability experireturned pieces, as some of them may not really be ence higher cost of ownership and longer repair times. defective. This scenario has existed in every field in which 3. The test run to find the fault may not stimulate the I have worked and is a continuing source of frustration fault into a detectable mode. Precipitation screens are to the person attempting failure analysis. I do not know intended to precipitate a latent defect into a patent defect, what to suggest for this source of NFFs as I think that i.e. one that is detectable (under the correct conditions of the technicians are doing the correct things in getting the stimulation). These screens use very high combined stress customer's equipment up and running as soon as possible. levels intended to accelerate the precipitation process. In 2. The returned units from the field may have faults the case of field returns, the precipitation screen is not that the functional self test software suites (power up, necessary as, if the unit to be tested has a fault that extended and functional) cannot detect. Test systems must caused the field failure, then it has been precipitated have very good fault coverage in order to be effective in already. What is needed here is a detection screen which detecting patent defects and this is frequently not the case.
is designed to detect patent defects, i.e. those which have The actual test coverage can be accurately assessed using already been precipitated. A detection screen typically uses automated fault injection techniques which are available amplitude modulated all axis (six degree of freedom) today from at least one manufacturer of the equipment, broadband vibration, temperature cycling, voltage vari-Proteus Corporation from my home town of Denver, Colations and other stresses simultaneously to detect patent orado USA. The equipment is used to inject non-destrucdefects. It generally takes the correct combination of tive faults in all available locations in a system and the stresses to make a patent defect detectable. This combifault detection software is run to ascertain if the fault has nation might be 8 GRMS of six axis vibration and -38 been detected. A report of faults found and those not degrees Celsius and nominal voltage input and some given found is generated allowing the software author to modify atmospheric pressure and some combination of other the software to gain better coverage. The best screen for stresses, all applied simultaneously. Good detection screens defects in the world will not find many unless very good are critical for any test scheme, especially for field returns. test coverage exists. Most manufacturers are reluctant to A combination of excellent fault coverage and detection send these boards back into the field so they pile up on screens will expose essentially all of the faults in field shelves or are thrown out because of the inventory tax returns, if they exist at all. I have seen situations where liability. In addition, the cost of not resolving this problem field returns showed no faults at all until detection screens is passed on to the customer.
were applied and then essentially all of them showed Test coverage is a critical component to detecting these the faults! types of failures. Most manufactures rely on power up self-test, extended and system functional tests to detect Gregg K. Hobbs President functional failures. These tests are used 90 per cent of the time to troubleshoot board failures by engineering, Hobbs Engineering Corporation 10218 Osceola Court manufacturing, field service, customer servicee and failure analysis. The relationship between fault coverage and Westminster, CO 80030 USA defect level is well documented.