Академический Документы
Профессиональный Документы
Культура Документы
Reliability has come to rank right along side cost For semiconductor relays, accelerated tests, which
and efficiency when evaluating the design and reduce time and testing costs, are used to predict
application of a system. An electrical system is reliability. Such tests involve applying temperature
composed of a collection of electronic components, cycling and other stress factors, e.g. humidity (85°C
materials and manufacturing processes. None of and 85%RH is often used as a standard
these elements are perfect and each can fail in combination).
some fashion that could influence the system's
functionality. The failure rate, λ, is the fundamental variable used
to define reliability. λ is expressed in terms of
Reliability predicts how often failures will occur. In failures per unit of time. Typically the failure rate of
electronic reliability analysis, it is assumed that a component is between 10-8 and 10-7 failures per
failures are independent and random and are hour and is a function of the applied stress levels.
statistically distributed with a constant failure rate
versus time. Of these factors, operating temperature is
especially important since it has a major influence
Stress prediction methods attempt to predict the on the overall reliability. The relative change in
reliability of components in a detailed manner. failure rate, i.e. the acceleration factor (AF), for
These methods assign a specific baseline failure when the operating temperature of a component is
rate for each component used in electronic design. changed is given by the Arrhenius equation:
This baseline failure rate is then modified by
multiplying certain factors relating to specific ⎡ Ea ⎛ 1 1 ⎞⎤
⎢ ⋅⎜⎜ − ⎟⎟ ⎥
components and stresses applied in the actual ⎣⎢ k ⎝ T1 T2 ⎠ ⎦⎥
application. A commonly used parts stress
AF = e
prediction method has been developed by the US
Military Agencies (MIL HDBK-217F). AF = Acceleration Factor
Ea = Activation Energy in Electron Volts =
However, there is an inherent problem in using part = 0.8V (possible range 0.3V to 1.5V)
stress methods for reliability prediction. Each of the k = Boltzmann's constant = 8.617·10-5 eV/K
sometimes hundreds of components in a design T1 = New Temperature in degrees Kelvin
must be assessed in terms of its specific operating T2 = Test Temperature in degrees Kelvin
conditions. In addition, the reliability prediction
method is rather theoretical and ignores vendor- The activation energy Ea is a factor that models the
specific differences, e.g. in construction or quality. temperature versus failure rate characteristics of a
The only true measure of reliability is derived from component. It will vary somewhat from one
reliability testing. component type to another. Failure rates for "use
conditions" are calculated from the acceleration
Relay manufacturers who offer both factor and the equivalent device hours using stress
electromechanical and semiconductor relays are time and number of samples tested. These
compelled to provide two different kinds of reliability estimates of equivalent use hours are well beyond
data. For electromechanical relays, which suffer the end-of-life hours associated with most
wear during their lifetime, reliability data is best applications when assuming an exponential
obtained from Weibull diagrams. All three parts of a distribution. The relationship between failure rate
bathtub life curve (early failures, random failures (expressed in FIT, 1 FIT = 1 failure per billion
and failures due to wear) can be demonstrated via device hours) and the Chi-squared distribution is
Weibull distribution. By performing suitable tests shown in the following equation.
under appropriate conditions, results can be
entered into a Weibull chart, leading to a failure
probability for a certain number of switching
operations. However, since semiconductor relays
do not suffer wear over the course of their lifetime,
Weibull testing is irrelevant for them.
χ2 χ2 ⎛ RH 1 ⎞
−N ⎡ Ea ⎛ 1 1
⎢ ⋅⎜⎜ −
⎞⎤
⎟⎟ ⎥
λ= =
AF = ⎜⎜ ⎟⎟ ⋅e ⎢⎣ k ⎝ T1 T2 ⎠ ⎥⎦
2 ⋅ EDH 2 ⋅ AF ⋅ CDH
⎝ RH 2 ⎠
χ2 = Chi-squared distribution for a given
confidence level AF = Acceleration Factor
EDH = Equivalent device hours Ea = Activation Energy in Electron Volts =
AF = Acceleration Factor = 0.8V (possible range 0.3V to 1.5V)
CDH = Cumulative device hours k = Boltzmann's constant = 8.617·10-5 eV/K
T1 = New Temperature in degrees Kelvin
T2 = Test Temperature in degrees Kelvin
Chi-squared distribution
RH1 = New relative humidity in percent
Number χ2 at 60% χ2 at 90% RH2 = Test relative humidity in percent
of failures confidence level confidence level N = experimental determined constant =
0 1.833 4.605 = 4.5 (possible range 2 to 12)
1 4.045 7.779 Using the values from our example, this equation
yields a different acceleration factor and therefore a
2 6.211 10.645
different FIT value.