An interlaboratory evaluation of the destructive, double-bond pull test is being conducted in cooperation with ASTM Committee F-1 on Electronics. In addition, a twolaboratory comparison of the bond pull test with the bond shear test was conducted to determine their significance for evaluating aluminum-aluminum ultrasonic bonds. The shear test was found not to be sensitive to the weakest part of the connection, the heel of the first bond. The pull test, on the other hand, is particularly sensitive to the weakening of the bond heel.

spectrometer, being conducted in cooperation with ASTM Committee F-1 on Electronics, was further studied. It was decided that the interlaboratory test should be deferred until test specimens which are less susceptible to helium absorption can be obtained.

A study has been initiated at MartinMarietta Aerospace to derive a quantitative relationship between leak size in hermetic packages and moisture infusion. The initial phase of the work involved the calibration of an in-situ moisture sensor, development of microventing procedures, and selection of a moisture sensitive integrated circuit.

Device Inspection and Test Work on the automated scanning low energy electron probe at the Naval Research Laboratory continued with assembly of the system and demonstration of its ability to observe variations in dielectric uniformity on an oxidized silicon wafer.

To facilitate application of the scanning electron microscope operating in the electron beam induced current (EBIC) mode for the diagnostic examination of silicon devices, a procedure was developed for predicting the magnitude of the EBIC signal. The procedure is based on published depth-dose curves and an analysis of the value for the fraction of energy backscattered which was shown to be essentially constant over the energy range of interest. Despite the fact that several simplifying assumptions were made, reasonable agreement was achieved with an experimental result.

3.1. Standard Reference Materials

The multipass interlaboratory test (NBS Spec. Publ. 400-17, pp. 8-9) of SRM 1520 [1] was completed. The purpose of the test was to establish both the stability of this SRM over a reasonable period of time and the level of agreement to be expected by any two laboratories using the SRM for referee pur

poses.

Each of the six participating laboratories, including NBS, was required to measure the room temperature resistivity of both wafers of its resident SRM set once a month for a twelve-month period. The measurements were rade according to the standard four-probe method [2], except that only two repetitive measurements on each wafer were required instead of the normal ten. Raw data were submitted directly to NBS for calculation and reduction. This part of the experiment served both as a data base from which to estimate the time stability of the SRM sets, and also as a screen against sudden changes in any laboratory's measurement response which would influence the second part of the study.

In the second part of the test, two complete SRM sets, two silicon wafers at each of the two resistivity levels, were circulated to each of the participating laboratories with a frequency of one laboratory per month. Two complete cycles of the laboratories were sade during the study. The estimate of precision to be expected in referee use of the SRM's was generated from these data. The resistivity was again measured according to the standard four-probe method [2] but only five repetitions were required per wafer. The use of two complete sets in this part served both to enlarge the data base and to provide allowance for possible specimen breakage during the study.

The data reported by one participant showed a statistically significant high-side bias for all measurements. Since it was determined that this participant failed to follow the measurement procedure with respect to probe specifications and correction for temperature, these data were excluded from the final analysis of the results of the experi

Rent.

The stability was evaluated by analyzing the drift of the resistivity values obtained on the ten resident specimens during the course

of the experiment. The results are summarized in table 1. The data from each resident specimen were fitted to individual linear regressions [3] as a function of month number. The slopes of the individual regressions, if statistically significant above measurement scatter, indicate the extent of drift for that specimen-laboratory combination over the course of twelve months.

Individual estimates of slope and intercept with their estimated standard deviation are shown in the table. While seven of the ten estimated slopes were negative, indicating slight decrease of resistivity with time on balance, only one of the calculated slopes was larger than its estimated standard error. In addition, the small values of correlation coefficient suggest that the resistivity changes are not linear with time.

With separate analyses for high and low resistivity levels, pooled estimates indicate that only slopes greater than 0.018 and 0.00018 ohm centimeters per month, respectively, would have been statistically significant at the 95 percent confidence level. The point of interest is the typical drift, if any, of all the specimens at either the low resistivity or the high resistivity level, taken as a whole for each level. Since there was only one slope in either category which exceeded its respective significance limit, this is well short of the 95 percent yield expected from the analyses performed if the hypothesis of detectable drift for material of either resistivity level were true. Hence, as a whole, this hypothesis must be rejected.

A statement of the precision to be expected in various referee measurement uses was derived from analysis of the measurements on the circulating sets of wafers using the calculated values of within laboratory and between laboratory variation for each specimen used [4]. Table 2 summarizes the results of the measurements and gives estimates from these results of the repeatability for replicate measurements within a lab and the reproducibility of measurements between two labs which could be expected with 95 percent confidence if the same specimens were used for referee measurement. The estimates of repeatability and reproducibility [4] are broken into three situations, namely those where the two laboratories involved would measure the specimen one, five, or ten