Purging: A reliability assurance technique for new technology semiconductor devices

Abstract

A new reliability assurance philosophy is described that represents the Bell System approach to solving the difficult problem of providing new device technology for early system deployment with well characterized and desirable reliability characteristics. This philosophy is especially important for high reliability systems such as satellites and submarine cables.It is based on the use of very high stresses, both thermal and nonthermal, to eliminate failure modes which are only weakly temperature activated, and to stabilize degradation rates. The technique is called "purging." HE USE of new technology devices in a system under de-T v e l o p m e n t carries with it the potential for significant improvements in the performance, cost, and market appeal of the new system. However; the new technology devices imply risks for the system designer. Short schedules imply uncertainty in availability and cost, and performance parameters may not be specified precisely. The area of greatest concern and risk frequently centers on device reliability. With the device design not frozen, with manufacturing processes not fully specified, and with manufacturing facilities still under construction, the device developer is required to complete a qualification program that yields responsible estimates of the device reliability and quality.For many systems and related products, the satisfactory reliability and assurance thereof is not available in time for early deployment, The development-manufacturing team recognizes the necessary cost of low assembly yield and field replacement, hopes it won't be excessive and goes on to meet shipment schedules without full reliability assurance. With hard work, a developing data base, and device design and process iteration, the reliability requirements are eventually met. However, for some systems, for example, undersea cables and satellites, such a strategy is not feasible. These systems, because field replacement is not possible or excessively expensive, must have adequate information to make reliability predictions for new technology devices before deployment, despite exceptionally tight schedules.The challenge of implementing a new digital submarine cable system in 1988 has focused attention on reliability issues for new devices, such as optical sources, detectors, etc. Certain of these devices will not have seen field use in other less critical systems before completing design of the submarine cable system. Out of this focus has come a philosophy of reliability qualification that has been exceptionally successful and has implications for all new technology semiconductor devices.

Methodology of Accelerated Aging

“…By this means the original lognormal curve (LN) is mapped into the burn-in curve (By=o. 4 original lasers do not affect the By= 0 A curve, it does not matter that the first 10 percent or so of them fall below the LN curve.) Several conclusions can be drawn from a comparison of the B and LN curves: After about 25-percent failure, the B curve is longer-lived than the original data, e.g., the median is raised from 750 to 900 hours.…”

“…In practice the methodology often reduces to fitting the data as well as possible with a linear degradation model (I changes linearly in time at constant T), i.e., 4 and h as in Fig. 4b.…”

“…Unless the stabilization process happened to actually improve the individual component's life, lifetime experiments on samples from the 40 percent with the longest predicted lives would be expected to yield a distribution which fell somewhere between the Ty= 0 . 4 curve (perfect correlation between the projected and actual lifetime) and the burn-in curve 5 y =o.7i6 (no correlation between the initial degradation rate and the actual life times). (A yield y = 71.6-percent corresponds to a 400-hour burn-in; this curve is not shown in Fig.…”

“…Specific procedures, or combina tions of procedures, are known by such names as inspecting, perform ance evaluating, burning-in, accelerated burning-in, gettering, outgassing, tailored accelerated screening, 2 high-stress screening, 2,3 overstressing, purging, 4 " 6 stabilization, 4,6 etc., and the reader is referred to the literature for their explicit descriptions. In the present section we simply outline a framework for characterizing such reliability proce dures.…”

“…Another truncation strategy, which is incorporated, for example, in the purge [4][5][6] and the tailored-accelerating-screen 2 procedures, is prem ised on the following physical model: All components eventually fail because all possess a common wear-out mechanism. In addition, certain flawed components are also afflicted with a second failure mechanism having a shorter life under normal operating conditions.…”