Part II: Reliability Prediction Standards

Note: This is the second part of a three-part article. It explores the various reliability prediction standards that are available, including MIL-HDBK-217, Telcordia, NSWC-98/LE1, PRISM, RDF 2000, CNET 93, HRD5, and GJB/z 299B. Part I identifies the major factors in reliability prediction models that contribute to predicting component failure. Part III provides guidelines for making a sound judgment when deciding which standard to apply to a particular analysis.

This article explores the various reliability prediction standards available, including MIL-HDBK-217, Telcordia, NSWC-98/LE1, PRISM, RDF 2000, CNET 93, HRD5, and GJB/z 299B.

Reliability Prediction Standards and Model Examples

MIL-HDBK-217

The earliest standard to appear was MIL-HDBK-217, Reliability Prediction of Electronic Equipment, which was developed by the United States Department of Defense (DOD) in the 1960s. Since then, MIL-HDBK-217 has been updated several times, with the most recent being Revision F Notice 2, released in February 1995. Although the DOD has discontinued updates of MIL-HDBK-217, this standard is still widely used in military and commercial applications throughout the world.

MIL-HDBK-217 Parts Count

MIL-HDBK-217 parts count prediction defines the overall equipment failure rate as:

EQUIP	=	Equipment failure rate
n	=	Number of generic part categories
Ni	=	Quantity of i th generic part
g	=	Failure rate of i th generic part (table lookup)
Q	=	Quality factor of i th generic part

If the equipment consists of parts operating in more than one environment, this equation is applied to each portion of the equipment in a distinct environment. The overall equipment failure rate is then obtained by summing the _EQUIP for each environment.

MIL-HDBK-217 Parts Stress

For a MIL-HDBK-217 parts stress analysis, the models are much more detailed and varied across part types. For example, the model for microcircuits, gate-logic arrays, and microprocessors is:

The model for a low-frequency diode is:

p	=	Part failure rate
C1	=	Die complexity failure rate
C2	=	Package failure rate
T	=	Temperature factor
E	=	Environment factor
Q	=	Quality factor
L	=	Learning factor
b	=	Base failure rate
S	=	Electrical stress factor
C	=	Contact Construction factor

A great deal must be known for each component in a design to determine the above factors. Depending on component type, the model may include other multiplicative factors as well. Assembly and system failure rates are simply the sum of the part-level and assembly-level failure rates, respectively.

Telcordia

The Telcordia standard, Reliability Prediction Procedure for Electronic Equipment, was developed by Telcordia Technologies Inc. It originated from the Bellcore standard developed by AT&T Bell Laboratories. Focusing on equipment for the telecommunications industry, Bell Labs modified the MIL-HDBK-217 equations to better fit their field data. The most recent version is Telcordia I, issued May 2001.

Failure Rates for Devices

The basis of the Telcordia math models for devices is referred to as the Black Box Technique. This parts count method defines a black box steady-state failure rate, _BB , for different device types as:

g	=	Generic steady-state failure rate for the particular device
Q	=	Quality factor
S	=	Electrical stress factor
T	=	Temperature factor

The above inputs, contained in the Telcordia standard, were obtained from statistical data collected over a number of years. For instances where temperature and electrical stress are unknown, Telcordia recommends using a value of 1 for _S and _T , which assumes electrical stress to be 50% of the rated value and temperature to be 40^o C. The steady-state device failure rate, _SS , considers an adjustment to the black box failure rate depending upon the availability of laboratory and field data and device burn-in. For the simplest case where no data is available, _SS = _BB .

Failure Rates for Units

Units are comprised of numerous devices, and in adhering with the prediction paradigm, the failure rate for a unit is generally the sum of the failure rates of all its contained devices. In the Telcordia standard, a parts count steady-state failure rate for units, _PC , is defined by:

SSi	=	The steady-state device failure rate of device i
E	=	The unit's environmental factor
Ni	=	The quantity of device type i
n	=	The number of device types in the unit

As with devices, the unit's steady-state failure rate, _SS , considers an adjustment to this parts count failure rate depending on the availability of laboratory and field data and device burn-in. For the simplest case where no data is available, _SS = _PC .

Failure Rate for the System

The system-level failure rate, _SYS, is simply the sum of all failure rates of the units contained in a system:

First-Year Multipliers

The first-year multiplier, _FY , is defined in Telcordia at the device, unit, and system levels as the ratio of the failure rate in the first year of operation to the steady-state failure rate. Although it is not included in the steady-state failure rate calculations, the first-year multiplier can be used to estimate the failure rate of these items during the first year of operation, or the infant mortality period. For devices, the first-year multiplier calculation depends on burn-in time, device stress, and burn-in temperature at the device, unit, and system levels. First-year multipliers for units and the system are calculated as weighted averages of the first-year multipliers at the device and unit levels, respectively.

PRISM

Developed in the 1990s by the Reliability Analysis Center (RAC) under contract with the U.S. Air Force, the PRISM methodology offers some different approaches to reliability predictions. PRISM considers the inherent failure mechanisms of the physical components comprising a system as well as other factors affecting the reliability at a system or assembly level. These factors include processing variables, empirical data from a predecessor system, and field or test data.

PRISM RACRates Models

In PRISM, the math models describing component failure rates are known as RACRates models. These models are the foundation of the PRISM analysis. PRISM provides RACRates models for capacitors, diodes, integrated circuits, resistors, thyristors, transistors, and software. The model for each of these component types is different. For example, part failure rate, _P , for capacitors in failures per million calendar hours is:

OB	=	Operating base failure rate
EB	=	Non-operating (or environmental) base failure rate
TCB	=	Temperature cycling base failure rate
SJ	=	Failure rate due to solder joints
EOS	=	Failure rate due to electrical overstress
G	=	Reliability growth factor
C	=	Capacitance value factor
DCO	=	Operating duty cycle factor
TO	=	Operating temperature acceleration factor
S	=	Operating electrical stress factor
SR	=	Operating series resistance factor
DCN	=	Non-operating duty cycle factor
TE	=	Non-operating temperature environmental factor
CR	=	Temperature cycling rate acceleration factor
DT	=	Temperature cycling ΔT acceleration factor

For components not having RACRates models, PRISM provides Non-electronic Parts Reliability and Electronic Parts Reliability Databooks (NPRD-95/EPRD-97). A multitude of part types can be found in these databooks with failure rates for various environments. However, the additional features explained below are not applicable to NPRD/EPRD parts.

PRISM Process Grades

The general PRISM failure rate of a system, _PSYS , is:

PG	=	Process grade
P	=	RACRates failure rate of the i th component
SW	=	RACRates failure rate of software

The process grade, PG, is a multiplicative scoring factor generated from a series of questions that specifically apply to the system under analysis. The general principle behind process grades is to adjust the predicted failure rate by accounting for other factors which can contribute to reliability such as the manufacturing process and design engineer experience, which can be difficult to quantify. Two process grade models are available: the Inherent model and the Logistics model. The Inherent model considers only those processes that relate to system design and build. The Logistics model additionally takes into account external conditions affecting reliability in the field. Generally, process grades may consider part quality, infant mortality, environmental conditions, design techniques, reliability growth, manufacturing practices, system management, induced failure from external stresses, defects that cannot be duplicated, and wearout.

Predecessor Analysis

Predecessor analysis is used to capture the evolutionary nature of design. It allows you to use empirical data collected for a predecessor product that is similar to an assembly or the entire system under scrutiny. Predecessor analysis produces a combined failure rate based on the predicted failure rate of the current and predecessor products, and the field failure rate of the predecessor product.

Bayesian Analysis

Bayesian analysis is used in PRISM as a method to integrate test or field data into the assembly-level or system-level calculations. The data is used to optimize the reliability prediction accounting for variables from a real system that cannot be included in models. The empirical data is weighted depending on the size of the dataset; thus, the larger the dataset, the more it contributes to the failure rate.

RDF 2000

RDF 2000 is a French Telecom standard that was developed by the Union Technique de l'Ectricite (UTE). The previous version of RDF 2000 is referred to as CNET 93. The most recent release was in July 2000. RDF 2000 provides a unique approach to failure rate predictions in that it does not provide a parts count prediction. Rather component failure is defined in terms of an empirical expression containing a base failure rate multiplied by factors influenced by mission profiles. These mission profiles contain information about operational cycling and thermal variations during various "working" phases. For example, component failure rate, , for Class I ceramic capacitors with y operational cycles and j thermal variations is:

(t )i	=	i th temperature factor related to the i th junction temperature of the capacitor mission profile. This factor is determined by an Arrhenius equation with a specified activation energy. It is a function of ambient temperature during the i th working phase.
i	=	i th working time ratio of the capacitor for the i th junction temperature of the mission profile. This is the ratio of time per year that the capacitor is 'on' at a particular junction temperature.
on	=	Total working time ratio of the capacitor
off	=	Time ratio for the capacitor being in storage (dormant) mode
(n )i	=	i th influence factor related to the annual cycle's number of thermal variations seen by the package, with the amplitude ΔTi . This factor is an empirical expression based on the number of cycles per year, ni , with thermal variation having the given thermal amplitude.
ΔTi	=	i th thermal amplitude variation of the mission profile.

Similar models exist for many component types in the RDF 2000 standard. The failure rate of the system is determined by summing all of the component failure rates.

Other Standards

NSWC-98/LE1
The Handbook of Reliability Prediction Procedures for Mechanical Equipment, developed by the Naval Surface Warfare Center, contains reliability models for mechanical devices such as springs, bearings, seals, motors, brakes, etc. The most recent version of this standard was released in September 1998.

HRD5
The Handbook of Reliability Data, published by British Telecom of the United Kingdom, is based on the former RDF 2000 standard, CNET 93. The most recent version was released in 1994.

GJB/z 299B
The Reliability Calculation Model for Electronic Equipment is a Chinese standard translated into English by Beijing Yuntong Forever Sci.-Tech. Co. Ltd in May 2001. This standard was developed for the Chinese military, and it is very similar to MIL-HDBK-217. It includes both a parts count and parts stress prediction analysis methodology.

Relex Reliability Prediction

Relex Reliability Prediction can use any of the standards described in this document to assess the reliability of your system design. Relex's superior integration enables you to mix calculation models within a single project, allowing you to select the model most appropriate to each part or component in your design. A comprehensive matrix of the model coverage for the four most widely used standards can be found at www.relex.com/resources/art/art_predmodels.asp. With Relex Reliability Prediction, the additional analysis capabilities available in one model can typically be applied to any model you use. For example, Telcordia calculation methods for taking field and test data into account in the prediction can also be used with MIL-HDBK-217. For additional product information, refer to www.relex.com/products/relpredsoft.asp.