Under contract with Signetics, Mr. Thomas J. Chaney of WashingtonUniversity, St. Louis tested a set of nineteen 74F786 samples(packages) to determine the metastable state recovery statistics forthe circuits. The tests were conducted using a procedure describedin a paper entitled “Characterization and Scaling of MOS Flip-FlopPerformance”, (section IV), by T. Chaney and F. Rosenberger,presented at the CalTech Conference on VLSI, January 1979. Thegeneral test procedure was to test all 19 packages under onecondition, then test the best, worst, and an average package in moredetail. According to Mr. Chaney, the test results from the 19packages formed one of the tightest groupings that he has everseen. As the parts were numbered, package No. 7 had the fastestresolving times, No. 11 produced some of the slowest resolvingtimes, and No. 1 had resolving times near the middle of the testresults. This ranking of the test results from 3 packages remainedthe same throughout the balance of the test program, whichsupports the complete testing of only 3 packages. In general, thepoorest performance resulted when the packages were heated tonear 75°C with VCC = 4.5VDC and the best performance resultedwhen the packages were cooled to near 0°C with VCC = 5.5VDC.The variation within one package caused by the temperature andVCC changes was greater than the variation from package topackage. It must be noted that none of the packages tested evenapproached the data sheet input to output worst case propagationdelay of 10.5ns. All the packages tested for a single active output,had propagation delays of about 6ns. Typically, the parts with longerpropagation delays also have slower resolving times. Thus, onewould expect that the delay time needed to have only one failure in32 years using a 10ns propagation delay part would be much longerthan a value derived from just adding 10–6 = 4ns to the abovecalculations. thus it appears that the poorest performance measuredin this study should be considered a measurement at the edge of thetypical range for 74F786 parts.It must also be noted that the tight grouping of this set of packagesmeans that, when comparing differences between these test results,the measured error, as outlined in “Measured Flip-Flop Responsesto Marginal Triggering”, IEETC, December 1983, is significant. Thisis illustrated in association with Table 5.