Defence Technology 16 (2020) 392e400

Contents lists available at ScienceDirect

Defence Technology
journal homepage: www.keaipublishing.com/en/journals/defence-technology

Priori information analysis of optocoupler accelerated degradation

experiment based on failure mechanism verification test
Xuan-gong Zhang a, *, Xi-hui Mu a, Jing Feng b, Hui-zhi Li a
a
Department of Ammunition Engineering, Army Engineering University Shijiazhuang Campus, Shijiazhuang, 050000, China
b
Hunan Ginkgo Data Technology Co., Ltd., Luositang Road, Changsha, Hunan, 410100, China

a r t i c l e i n f o a b s t r a c t

Article history: The optocoupler is a weak link in the inertial navigation platform of a kind of guided munitions. It is
Received 18 April 2019 necessary to use accelerated storage test to verify the storage life of long storage products. Especially for
Received in revised form small sample products, it is very important to obtain prior information for the design and imple-
8 May 2019
mentation of accelerated degradation test. In this paper, the optocoupler failure mechanism verification
Accepted 13 June 2019
Available online 18 June 2019
test is designed and the experimental results are analyzed and the prior information is obtained. The
results show that optocouplers have two failure modes, one is sudden failure and the other is degra-
dation failure; the maximum temperature stress of optocoupler can't exceed 140
C; the increase of
Keywords:
Priori information
leakage current of optocoupler is caused by movable ions contaminating the LED chip. The surface
Failure mechanism leakage current is proportional to the adsorption amount. The increase of leakage current makes p-n
Failure mechanism verification test junction tunneling effect occur which LEDs the failure of the optocoupler. The lifetime distribution model
Accelerated degradation of the optocoupler is determined by the failure physics. The lifetime of the optocoupler is subject to the
lognormal distribution. The degeneracy orbit of the optocoupler leakage current is described by a power
law model. The estimated values of the orbital parameters are initially calculated and the parameters of
its life distribution function are deduced. The above information lays a good foundation for the opti-
mization design and data processing of the accelerated degradation experiment.
© 2020 China Ordnance Society. Production and hosting by Elsevier B.V. on behalf of KeAi
Communications Co. This is an open access article under the CC BY-NC-ND license (http://
creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction degradation test is an important technical approach to solve the

reliability assessment of high-reliability and long-life products.
As a kind of important device of a certain type of guided missile There are a large number of literatures currently studying accel-
inertial navigation platform, a certain type of optocoupler is erated degradation tests (ADT) [1]. At present, the optimization
responsible for isolating various electrical signals for the inertial design of accelerated degradation test is divided into two major
navigation platform and avoiding the interference of the electrical categories: analytical solution method and simulation-based
signals of various components in the inertial navigation platform. method. Tseng and Yu put forward a degradation test termination
Through the electrical parameter detection of the inertial naviga- criterion to make the statistical results of the data most accurate
tion platform of a certain type of guided munition that has reached [2]. Under the constraint of cost, Wu and Chang aimed at the
the service period and expired service, it is found that the opto- minimum mean square error of the p-order fractional life, and they
coupler is the weak link of the weapon system, and its reliability is established a degradation test optimization design method for the
less than traditional weak parts such as accelerometers and gyro- degradation rate obeying the exponential distribution [3]. While Yu
scopes. Since optocouplers are highly reliable and long-life prod- and Tseng did the same thing with the degradation rate obeying the
ucts, traditional life tests not only consume a lot of manpower and lognormal distribution and the reciprocal Weibull distribution
material resources, but also often have long test cycles. Accelerated [4,5]. The above literature all uses analytical methods to optimize
the accelerated degradation test. Because the derivation process of
the analytical method is very complicated, the simulation-based
* Corresponding author.
method is usually used to optimize the design now. Wang and
E-mail address: [email protected] (X.-g. Zhang). Zhang [6] presented the general flow of the optimization design of
Peer review under responsibility of China Ordnance Society accelerating degradation test scheme based on mixed-effects

https://doi.org/10.1016/j.dt.2019.06.011
2214-9147/© 2020 China Ordnance Society. Production and hosting by Elsevier B.V. on behalf of KeAi Communications Co. This is an open access article under the CC BY-NC-
ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
 X.-g. Zhang et al. / Defence Technology 16 (2020) 392e400 393

model under the simulation-based method. Pan studied the

optimal design of step-down accelerated degradation test under
the condition of small sample size using the Monte-Carlo simula-
tion method [1]. Tan studied the optimization scheme of acceler-
ated degradation test based on simulation in the condition of
competitive failure [7]. However, the main problem of the simula-
tion method facing is to specify the form of the degenerate orbit
model and specify that some coefficients follow the inverse Weibull
distribution, which is a big restraint for it. It should be pointed out
that whether it is an analytical method or a simulation-based
method, the optimization design of accelerated degradation test
is inseparable from the priori information. Transcendental infor-
mation includes degenerate orbit models, accelerated degradation
equations and preliminary estimates of correlation coefficients. The
literature of various types of research on accelerated degradation
test optimization methods often uses the prior information given in
other literature, or summed up through historical data [8]. How-
ever, the problem is that due to various reasons, the historical data
of some products are incomplete or cannot be collected. In this case,
a failure mechanism verification test is required to obtain the prior
information. In this paper, the optocoupler used in a certain type of
guided missile was tested through failure mechanism verification
experiment. Based on the determination of its failure mechanism,
the test data was analyzed and deduced, and the prior information
required for the optimal design of the accelerated degradation test
Fig. 1. Schematic diagram of optocoupler structure.
was obtained, which lays a good foundation for optimal design.
This article is divided into five parts, the first part is the intro-
duction, mainly discussing the problem of prior information
acquisition for accelerated degradation test and the significance of
the study in this paper; the second part is the setting of the failure
mechanism verification test program and the description of the test
results; the third part is the failure mechanism analysis of the
optocoupler and the derivation of the lifetime distribution; the
fourth part is the derivation and fitting of the degenerate orbit, as
well as the estimation of the relevant estimates; the fifth part is the
conclusion.

2. Test scheme design and test data visualization

2.1. Structure introduction and test scheme design

There are two types of optocouplers: one is a nonlinear opto-

Fig. 2. Schematic diagram of optocoupler functional test.
coupler and the other is a linear optocoupler. According to the
classification, the optocoupler studied in this paper is a nonlinear
optocoupler. More specifically, the optocoupler is a switch-type optocouplers, the current transfer ratio (CTR) is a very important
optocoupler, and its specific structure is shown in Fig. 1: parameter, and since the nonlinear optocoupler does not have a
This type of optocoupler is a four-legged photosensitive resis- CTR parameter, CTR is not selected as a parameter to characterize
tance type of optocoupler. The tube foot is plated gold dust to in- the optocoupler performance. Therefore, the ordinary optocoupler
crease the conductivity of the pin, and the head is coated with black and the optocoupler studied in this paper are completely different
paint to further insulate the external light. Its left is LED and its in function. The former is mostly a linear optocoupler, which con-
right is photosensitive resistor. According to the foregoing, this type trols the transmission of current, while the latter is used as a circuit
of optocoupler is used as a solid switch. Its functional test is shown switch. After the previous factory visits and literature research, the
in Fig. 2. test parameters that represent the optocoupler performance are
The test process of Fig. 2 is identical to the operational experi- not only dark voltage and flip voltage, but should be five parame-
ence of the optocoupler in the missile. When the right given ters. These five parameters are: dark voltage, flip voltage, LED for-
to 250 V voltage and left without electricity, the photosensitive ward working pressure drop, reverse breakdown voltage and
resistance (that is, the dark resistance) is not less than 10 MU, reverse leakage current. The first two parameters correspond to the
conversion on the voltmeter readings (dark voltage) is not greater performance of the photoresistor, and the last three parameters
than 215.146 V. When the left side is energized and the current is correspond to the performance of the LED.
12 mA, the right voltage is rapidly reduced, and the voltage repre- For the optocoupler inside the missile control system, most of
sentation number (that is, the conduction voltage) should not be the time is in stock, and the test and transportation time only oc-
less than 1.8 V, also known as the flip voltage. It can be seen from cupies a small amount of time during the storage period. Therefore,
Fig. 2 that the optocoupler studied in this paper is only used as a temperature stress and humidity stress are the main environ-
solid-state switch and belongs to a nonlinear optocoupler. Many mental factors that affect the reliability of optocoupler storage. For
literatures have studied linear optocouplers. For linear
394 X.-g. Zhang et al. / Defence Technology 16 (2020) 392e400

this type of optocoupler, it is a sealed design. All functional parts are voltage of the photoresistor did not change throughout the test. The
packaged in a metal case. The pins are gold-plated to reduce re- flip voltage as an indicator of the optocoupler function and the
sistivity and resist corrosion. In addition, the optocoupler is placed forward voltage drop characterizing one of the LED performance
in an electronic compartment inside the missile control system. The parameters changed very little. Instead, The two parameters of LED
electronic compartment is integrally sealed with polyurethane reverse leakage current and reverse breakdown voltage varied
foam, which can also effectively resist water vapor erosion. greatly. During the experiment, two optocouplers did not degen-
Therefore, in theory, moisture is only possible to enter during the erate but suddenly failed (145
C and 147
C). After laboratory
optocoupler production process. Therefore, the effect of humidity testing and analysis, it was considered that the internal LED's
on the optocoupler is limited. The main consideration is the influ- connecting line was broken and an open circuit caused their failure.
ence of temperature stress on the storage reliability of the Specifically, the junction of the surface of the LED chip is separated
optocoupler. from the lead connecting it, resulting in an open circuit. This may be
Therefore, the method of stepping temperature stress is used in due to defects in the bonding process, or poor quality of the
the test. Since one of the purposes of this experiment is to deter- transparent adhesive.
mine the ultimate temperature of the optocoupler, the temperature The remaining optocoupler produced significant degradation.
limit is not set. Specific way is as follows: the temperature range is So the optocouplers have two failure modes, one is sudden failure
divided into three parts, the first paragraph is from 30
C to 50
C, and the other is degradation failure. Obviously the two are inde-
10
C for step length; The second paragraph as 50
Ce100
C, 5
C pendent of each other. Fig. 4 and Fig. 5 show the reverse leakage
for step length; The third paragraph from 100
C to test cut-off current and breakdown voltage as a function of temperature and
temperature, with 2
C for step length. At any test temperature time.
point, the test is performed at room temperature. To ensure this, As shown in Fig. 4, the abscissa represents time, the left ordinate
the rules are as follows:in the first stage, the heating time and represents temperature, and the right ordinate represents data
cooling time are one hour; In the second stage, the heating time is measured at room temperature after heating at the corresponding
constant and the cooling time becomes two hours. In the third temperature for 1 h. The blue step line in the figure indicates the
stage, the heating time is constant and the cooling time becomes stepping temperature. Red lines and data points indicate the values
three hours. of the measured parameters. Three hours of heating were per-
The test scheme section is shown in Fig. 3. formed at 50
C, 80
C, and 100
C respectively. After each hour of
Three temperature points are chosen to be tested three times, heating, the temperature was cooled to room temperature and
50
C, 80
C and 100
C. Other temperature points are tested only related parameters were measured again. Therefore, at these three
once. Because the sample is expensive, but considering that the test temperatures, the ladder lines are longer than the rest.
parameters may interact with each other, the failure mechanism is In the figures above, the breakdown voltage and reverse leakage
not necessarily unique. The sample can neither be too much nor too current of the LED inside the optocoupler have very obvious
little, so three samples are used in this test. In addition, the reason degradation phenomenons. It needs to be pointed out that the
for using three samples is that although there is no clear standard reverse leakage current at the initial test stage (less than 80
C) is
and document support, according to engineering experience, the measured in nanoamperes (hundreds of nanoamps), and it is
determination of any one distribution requires at least three points. measured in microamperes (microamps to tens of microamperes)
Therefore, three samples were selected as the number of experi- in the later period. In Fig. 4, the unit is selected as microamperes.
mental samples. When one of the following conditions occurs, the Therefore, the reverse leakage current in the early stage is not
sample is withdrawn from the test: 1. The functional indicator obvious on the graph. So thinking that early data can be ignored is
flipped voltage reaches a given threshold. 2. The data suddenly inappropriate. It can be seen from the figures that the trend of
changes dramatically, the optocoupler will be returned to the last leakage current and breakdown voltage with temperature and time
temperature, and then test again after cooling. If the data can be is generally opposite, and there is an inflection point after the 38th
recovered, it will continue to be heated up, and if not, exit the test. hour. After the 42 nd h heating is finished, the optocoupler cannot
reverse the voltage and it cannot complete the intended function,
2.2. Test data visualization which means complete failure. In comparison, the forward voltage
drop of the LED in the optocoupler does not change significantly,
According to the set test plan, the sample was tested. The dark

Fig. 3. Test scheme section. Fig. 4. Leak current line chart with the change of temperature and time.
 X.-g. Zhang et al. / Defence Technology 16 (2020) 392e400 395

the formation of surface leakage channels in addition [9,10]. The

change in leakage current is caused by the contamination of the
movable ions on the chip, forming a surface leakage path, which
explains why the breakdown voltage exhibits a roughly opposite
change with the leakage current. One of the reasons is that the
existence of the surface leakage channel is equivalent to the parallel
connection of a resistor, which will cause the overall resistance of
the p-n junction to decrease, which in turn causes the reverse
breakdown voltage to decrease. Therefore, the increase of the
leakage current is the cause of the breakdown voltage drop.
Another reason will be explained in the section of determining the
life distribution model. However, the current literature is not clear
for further explanation on how to form a surface leakage channel.
By studying related literature, this problem can be explained from
two perspectives: quantum physics and surface chemistry.
Fig. 5. Breakdown voltage line chart with the change of temperature and time.
3.1.1. Quantum physics aspect
but there is a phenomenon of slow decline. Therefore, it can be It is generally believed that the composition of LED reverse
determined that the failure of the optocoupler is caused by the leakage current consists of the following two currents: the current
leakage current of the LED and the change in the breakdown of the tunnel through the barrier the hot electron emission current
voltage. In fact, after the 36th hour of heating is over, the perfor- over the barrier [11]. In the case of not very high temperature, the
mance of the optocoupler goes into an unstable state. The corre- tunnel current is the main one, and the thermal electron emission
sponding temperature is 143
C. Therefore, considering the failure current is very small and can be ignored.
of three optocouplers the maximum temperature stress of the In the production of LED chips, due to the complete periodicity
optocoupler cannot exceed 140
C. Since the three samples are all of the crystal lattice, the surface state is generated by the sudden
due to failure of the internal LED, the life of the optocoupler is interruption of the crystal surface. The atoms on the semiconductor
equivalent to the lifetime of the internal LED. surface have only one side with adjacent atoms. On the vacuum
side, the valence electron does not have covalent bond with it.
3. Analysis of the failure mechanism of optocoupler and its Therefore, each surface atom has an unpaired electron, which is in a
determination of the life model localized orbit that points away from the surface. This kind of track
is often referred to as a “suspended bond”. These suspended bonds
3.1. Qualitative explanation of failure mechanism adsorb certain positive ions (such as Naþ, Cuþ, etc.) [23], When
these impurity ions “congregate” on the surface of the p-n junction
In general, it is assumed that the degradation is generally and its vicinity, a leakage path traverses across the p-n junction
monotonic, but the leakage current and the breakdown voltage surface [12]. When add reverse voltage, voltage is applied to not
caused an inflection point after heating in the 38th hour. The causes only for the inside of the semiconductor, but also at the same time
of LED reverse leakage are more complicated and varied, specif- will be applied to the surface of the semiconductor, forming a
ically, including production process defects, chip defects, and static surface recombination current or surface leakage current, which
problems. Process defects are mostly caused by excessive silver results in increased reverse leakage current.
glue, chip cracking and solder ball offset or excessive solder balls. On the other hand, according to Roderick's research, on the edge
Chip defects are mostly caused by movable ion contamination and of the contact between metal and semiconductor, the effect of
p-n junction collapse. Static problems are often caused by induction tunnel effect is especially obvious due to the decrease of the width
or friction. In general, the increase in LED leakage current caused by of the barrier and the decrease of the image force because of the
excessive silver glue, solder ball offset or excessive solder ball often concentration of power lines and the increase of electric field in-
occurs during the factory inspection period after the product is tensity [13].The effect is even more pronounced if the surface of a
packaged, and is found at the beginning. The situation of the chip semiconductor near a metal surface accumulates a positive surface
cracking is different, and it is difficult to be found at the beginning. charge that makes the barrier thinner. The effect is even more
After the chip breaks, the silver glue slowly penetrates into the pronounced if the surface of a semiconductor near a metal surface
surface of the chip along the gap capillary, causing the p-n junction accumulates a positive surface charge that makes the barrier
to be short-circuited. However, the cause of this failure cannot thinner. The effect is even more pronounced if the surface of a
explain why the leakage current increases first and then decreases, semiconductor near a metal surface accumulates a positive surface
as does the p-n junction collapse. The same is true for electrostatic charge that makes the barrier thinner. Therefore, due to positive
problems. The accumulation of static electricity generated by fric- ions, the current of the tunnel will become larger, which will in-
tion or induction is often generated during the production phase. crease the overall leakage current.
However, in general, static electricity does not easily accumulate
high voltage. Although the static electricity condition is random 3.1.2. Surface chemistry aspect
and ubiquitous, it is not always able to release enough energy to As is mentioned, the surface of the chip, due to the existence of a
cause damage. Damage caused by static electricity is infrequent and “suspended bond”, will adsorb some positively charged ions.
is a small probability event. So basically this reason can also be Adsorption is the phenomenon that atoms, molecules, or ions of a
ruled out. The above is the reason for determining the leakage substance are attached to the surface of another substance.
current of the sample LED by the exclusion method, that is, the Adsorption is divided into physical adsorption and chemical
mobile ion pollution. adsorption. According to Langmuir adsorption theory, the adsorp-
According to Xu and Xiao, with the increase of temperature, the tion of some positively charged ions on the semiconductor surface
first increase and then decrease of the leakage current are the belongs to chemical adsorption [14]. In the chemical adsorption,
typical active characteristics of mobile ion pollution on the chip and adsorbents and adsorbents combine in the form of chemical bonds
396 X.-g. Zhang et al. / Defence Technology 16 (2020) 392e400

to approximate chemical reactions, and thus require certain acti-

vation energy [15]. Therefore, chemical adsorption usually takes rL
R¼ (3)
place at a certain temperature, and in a certain range, the adsorp- Dl
tion speed is accelerated with the increase of temperature. Unlike According to Ohm's law, the surface composite current I is:
physical adsorption, chemical adsorption is basically carried out on
monolayer on the surface of adsorbent and has saturability. At the U UDl
I¼ ¼ (4)
same time, due to the influence of thermal motion, the desorption R rL
reaction is also carried out, namely, ions leave the semiconductor
Surface composite area A satisfies:
surface. So both reactions go together, also called the “confronta-
tion reaction” [16]. Since the adsorption reaction is the exothermic A¼l  L (5)
reaction, when the temperature is further increased, it will not be
conducive to the adsorption. When the reaction rate of the And then,
desorption is greater than the adsorption rate, the total adsorption
capacity of the moving ions on the semiconductor surface will A
l¼ (6)
decrease, and the leakage current will be reduced by combining the L
previous quantum physical interpretation. Then the expression formula of I becomes:
Based on the above analysis, the increase of the leakage current
of the optocoupler is due to the formation of surface leakage UDA
I¼ (7)
channels which is due to the contaminated movable ions on the r  L2
chip. As mentioned above, the leakage current and breakdown
In the above equation, L is a fixed value. Assuming only one kind
voltage of the sample are generally opposite, and the inflection
of ion adsorption, then r is a fixed value. As mentioned above,
point is also the same. After calculating the correlation coefficient of
chemical adsorption is basically carried out on monolayers of the
the two, it is 0.9867. It can be seen from the correlation coefficient
surface of the adsorbent. In other words, the thickness of the
that the correlation between the leakage current of the optocoupler
adsorbed ion layer is determined by the ions themselves. Assuming
and the breakdown voltage is very close to 1, which can correspond
that only one kind of ion is adsorbed, D is also a fixed value. In
to the previous physical interpretation, thus forming a surface
addition, since the reverse voltage U applied each time is fixed, and
electric field or surface composite current.
the surface resistance caused by adsorption is in parallel with the
impedance inside the semiconductor, and does not involve a
3.2. Mathematical relations of surface current and surface complex voltage division problem, U in the above equation is also constant.
quantity Assuming the tunnel current is negligible (several nanoamps to
several tens of nanoamps), I is proportional to adsorption area A,
The foregoing qualitatively explained that as the adsorption area thus:
increases or decreases, the surface current increases or decreases.
IfA
The relationship between surface current and surface adsorption
complex amount is deduced by the surface resistance formula A is the surface adsorption area. According to the surface chemical
below. interpretation above, it is actually the amount of adsorption.
As shown in Fig. 6, suppose that the surface recombination area Therefore, the surface leakage current is in direct proportion to the
is A, the length parallel to the current direction is L, the thickness of amount of adsorption, and the reverse leakage current curve can
the ion layer is D, and the width of the surface compound amount is fully represent the curve of adsorption amount.
l, then the ion source bottom area S is:

S¼D  l (1) 3.3. Determination of optocoupler life distribution model based on

According to the surface resistance formula [17], there are: failure physics

rL In Sections A and B, the causes of changes in leakage current and

R¼ (2)
S breakdown voltage are given by analysis. The breakdown voltage is
reduced due to the increase of leakage current, and the increase of
Then:
leakage current is due to the formation of surface leakage channels
due to movable ion contamination, and the physical and chemical
explanations of the increase of leakage current and the quantitative
relationship between the surface and the composite amount are
given. As mentioned above, one reason for the drop in breakdown
voltage is that the presence of a surface leakage path is equivalent
to a resistor connected in parallel, which causes a decrease in the
overall resistance of the p-n junction, which in turn causes a
decrease in the reverse breakdown voltage. However, the parallel
resistance formed is not the essential cause of the deterioration of
the p-n junction performance of the LED, but the performance of
the p-n junction is degraded due to the presence of the surface
leakage channel changing the surface state of the p-n junction
[18].This is another reason for the breakdown voltage drop.
According to the semiconductor p-n junction tunnel breakdown
theory, the tunneling probability of a microparticle with a mass of m
Fig. 6. Ion adsorption schematic diagram. and with energy E of crossing the barrier u(x) is:
 X.-g. Zhang et al. / Defence Technology 16 (2020) 392e400 397

chemical mechanisms that cause product failures, and the internal

xð2 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
relations between product failures and the conditions of use
2
P ¼ exp  2m½uðxÞ  Edx (8) (environmental stresses). Typical failure physics models include
h
x1 cumulative damage models, chemical reaction theory models, and
models based on degradation rates. The model describing the in-
If the tunnel length is short, or the barrier height u(x) - E is low,
ternal damage accumulation process of materials is called cumu-
the tunneling probability will increase, which causes the carrier
lative damage model. Common cumulative damage models include
concentration in the cut-off layer to increase to form a large
Paris model, power law model and Birnbaum-Saunders model
tunneling current under the applied electric field. As mentioned
(abbreviated as BS model) [24]. The basic idea of the reaction theory
above, if a positive surface charge accumulates on the surface of the
model is that the damage and degradation of materials and com-
semiconductor near the metal, the barrier at the edge is made
ponents are generally caused by physical-chemical processes such
thinner, and the tunneling effect is more pronounced. In addition,
as corrosion, wear, and diffusion. When these processes continue to
Xue [19] pointed out that when the “dangling bond” on the surface
a certain degree, the failure will occur. The model derived from the
of the p-n junction adsorbs ions, defects are formed. These defects
modeling of degraded orbits by means of physical-chemical pro-
will form local quantum states around them, and the corresponding
cesses is called a reaction theory model. In essence, what it shows is
energy levels will be within the forbidden band, causing the deep-
the continuous accumulation of damage [25]. The model based on
level defect density to become larger [20]. The region where the
the degradation rate differs from the previous two models. Both of
deep-level defect density is larger has a larger tunneling probability
the above models directly obtain the degradation track of product
of the electron crossing barrier u(x), so that the reverse breakdown
performance by analyzing the product failure mechanism. The
voltage is low. The reverse breakdown voltage will affect the carrier
model first establishes the degradation rate model, and then uses
transport process when the LED is working in the forward direction
the integral method to obtain degenerate orbits.
[21], which will further darken or flash the LED when it is working,
In summary, it can be seen that because the failure mechanism
which makes the optocoupler voltage flipping difficult. Therefore,
of the product is already known, it is more appropriate to use the
the essential cause of LED failure inside the optocoupler is that
cumulative damage model or the reaction model than to use the
tunnel breakdown occurs, and the tunnel breakdown occurs due to
model based on the degradation rate.
the leakage path formed on the surface of the p-n junction.
From an intuitive perspective, according to the modeling theory
Gao [22] obtained the log-normal distribution of the p-n junc-
of the reaction theory model, the degeneracy model of the opto-
tion under the premise that the p-n junction tunnel breakdown
coupler leakage current is the most suitable for the reaction theory
caused the failure, so no further derivation is made here. Therefore,
model. The reaction theory model is given by foreign scholars when
based on the above analysis, the life of the LED can be determined,
studying the failure of circuit boards due to the growth of fine fibers
that is, the life of the optocoupler is also subject to a lognormal
between insulating materials [26]. The basic assumption is: Let
distribution from the perspective of p-n junction tunnel break-
material A1 undergo a chemical reaction to A2 under certain envi-
down. Therefore, based on the above analysis, the life of the LED can
ronmental conditions (such as temperature, humidity, electricity,
be determined, that is, the life of the optocoupler is also subject to a
etc.) as follows:
lognormal distribution from the perspective of p-n junction tunnel
breakdown. Since two samples in this test produced sudden fail-
ures and the sample size was too small, it was impossible to A1 ! A2 (9)
determine the distribution pattern of the optocoupler life in the
case of sudden failure. From the perspective of accelerated test Where k is the reaction rate constant and the reaction is expressed
design, it is only possible to consider the degradation process first, as:
and then consider the case of sudden failure. Therefore, the life
distribution is initially determined to be a lognormal distribution, dA1 dA2
and the regression model is fitted and the relevant estimates are
¼  kA1 and ¼ kA2 ; k > 0
dt dt
obtained. This is always better than blindly performing a formal
Solving the above two differential equations, we can get the
accelerated degradation test on the optocoupler.
relationship between A2 and time t (including A1 (0) and A2 (0)).
However, the above differential equations are derived from the
4. Determination of degradation models and preliminary reaction (I). According to the theory of reaction kinetics, the reac-
estimation of related parameters tion is a first-order reaction, and the differential equations of
second-order reaction and the above become very complicated. The
4.1. Determination of degradation models first-order reaction requires that only one kind of material reacts, or
when the two materials react, the measurement coefficient and the
When the deterioration of a product's performance degradation initial concentration are equal, or when more than one kind of
characteristics or product health status can be characterized by a material reacts, all the substances other than A are in excess (the
key performance parameter, failure performance modeling, amount exceeding A is at least 10 times) [27]. The fine-conducting
degradation distribution modeling, generalized degenerate fibers studied by foreign scholars are composed of cupric chloride.
modeling, stochastic process degradation modeling, Gauss-Poisson Compared to chlorine, the amount of copper is almost infinite, and
model, and degradation modeling with change point can be it is far more than the amount of chlorine. Therefore, a degenerate
considered to establish a performance degradation model, and then model can be constructed through a reaction theory model. For the
product reliability analysis can be implemented [23]. Since the optocoupler leakage current parameters, due to the opposing re-
failure mechanism of the degradation of the optocoupler has been action is already very obvious in the late period, obviously neither
analyzed in detail through the analysis of the verification test, the the semiconductor surface nor the ionic impurities can produce an
degradation physics model of the leakage current is used to overwhelming advantage to the other party in terms of quantity. On
establish the degradation model. The so-called failure physics the other hand, it is assumed that a “chemical reaction” occurs
model refers to the corresponding degenerative orbital model between the surface of the semiconductor and the ionic impurities,
established by analyzing the changing laws of the physical or but the quantitation coefficient and the starting concentration
398 X.-g. Zhang et al. / Defence Technology 16 (2020) 392e400

cannot be guaranteed to be equal. Therefore, in theory, the reaction rate factor and is related to environmental stress.
theory model is not suitable for the degradation modeling of the For degraded products, according to the definition of degrada-
product. tion failure, when the performance reaches the failure threshold
Based on the experimental data, a reaction model was fitted with time, the product is considered to be ineffective, and the
preliminarily. The leakage current values at 50
C, 80
C, and 100
C corresponding time is the life of the product [30]. For the opto-
are used here. There are three values at each temperature. Through coupler, if the leakage current at time t is y(t), the failure threshold
matlab fitting, it was found that its decision coefficient R-square is is D. When the leakage current reaches the failure threshold with
less than 0.1. This shows that the reaction theory model is not time, the optocoupler fails, and the lifetime of the optocoupler is
supported from the perspective of data. the first time that the leakage current reaches the failure threshold:
From the previous analysis, we can see from the perspective of
physical modeling of failure, the reaction model is not suitable. Te ¼ infft : yðtÞ  D; t  0g
Consider the cumulative damage model. As mentioned earlier, it
Taking into account the storage conditions, the leakage current
usually has three forms: Paris model, power law model and BS
is monotonically increasing with time, and the lifetime of the
model. The Paris model is often used for mechanical failure caused
optocoupler can be obtained from the following equation:
by the expansion of metal fatigue cracks. The BS model is often used
for failure due to cyclic loading. The power law degradation model  
D  C 1=b2
is often used for modeling of degradation such as resistance yðTe Þ ¼ D0Te ¼ (12)
degradation, cable aging, and concrete corrosion [28]. Theoretically b1
speaking, the former two types are not suitable for the modeling of The lifetime of the optocoupler follows a lognormal distribution,
leakage current degradation of optocouplers. Therefore, the use of a thus lnTe  Fðme ; s2e Þ.
power law degradation model to construct a degradation model of
  
leakage current is considered. The power law model describes the D  C 1=b2
relationship between the degradation amount and the generalized ln  F me ; s2e
b1
time [29]. Assume that the performance parameters of a product
change monotonically with the extension of working hours, and the Further,
performance parameters at time t satisfy:
1 1 
lnðD  CÞ  lnb1  F me ; s2e
yðtÞ ¼ b1 t b2 ; b1 > 0; yðtÞ > 0 (10) b2 b2
In the above formula, (D-C) and b2 are constants, therefore, b1 is
Where y(t) is the degenerate quantity; b1 and b2 are the degenerate
obeying a lognormal distribution, and is denoted as b1  LNðmb ; s2b Þ.
model parameters, b1 is the degeneration rate factor, which is
mb and sb represent the logarithmic means and logarithmic stan-
related to the working environment stress, b2 is the shape
dard deviation of the degradation rate distribution respectively.
parameter of the degeneracy curve, and only related to the
In summary, the degeneration trace of the optocoupler leakage
manufacturing material of the product, t is the generalized time.
current can be described by the hybrid effect model as follows:
Since the leakage current of the optocoupler is already present
at the factory, the power law model is modified to add an initial
yðtÞ ¼ b1 t b2 þ C (13)
value C. Due to the uncertainty of production assembly, the initial
value of leakage current C can be regarded as normal distribution, 
thus C  Nðmc ; s2c Þ. For the sake of simplicity, C is taken as a constant b1  LN mb ; s2b (14)
here. Therefore, the power law model can be rewritten as:

yðtÞ ¼ b1 t b2 þ C; b1 > 0; yðtÞ > 0; C > 0 (11) (i) In the above equation, b1 and C are random, which respec-
tively reflect the differences of the initial values and the
Using matlab to fit data of 50
C, 80
C,
and 100
C.
There are
degradation rates of different optocouplers. For simplicity,
three values at each temperature. The fitting results are shown in
consider C as a constant. b2 is a fixed constant for the
the following table:
adsorption and leakage mechanisms.
As can be seen from Table 1, when using the modified power law
(ii) Leakage current degradation rat b1 meets the Arrhenius
model to fit the data, from the R-square, the fitting effect is very
equation. Its logarithmic standard deviation sb has nothing
good. The reason why the value of C changes greatly is because the
to do with the level of accelerated stress, that is, sb does not
data used is the amount of degradation of the leakage current of the
change with temperature. The relationship between the
same optocoupler at different temperatures. The data represented
logarithm mean mb of b1 and temperature T (
C) satisfies:
by the C value means the amount of deterioration of the leakage
current before 50
C, before 80
C, and before 100
C. According to a2
the comparison with the actual value, the error of the fitted C value mb ¼ a1 þ a2 x ¼ a1 þ (15)
T þ 273:5
does not exceed 0.8%. b2 as the shape parameter of the degeneracy
curve should itself be fixed. It can be seen from Table 1 that b2 floats Where a1 and a2 are model parameters of acceleration equation.
up and down around 2. Its mean value is 2.141. b1 is a degradation Through the b1 value obtained from the previous fitting, the values
of a1 and a2 can be estimated by the least squares method, and the
linear unbiased estimation of sb can be performed. The specific
Table 1
results are shown in Table 2.
Regression model parameters fitting results.

b1 b2 C R-square


50 C 2.641 2.369 235.4 0.8732 4.2. Optocoupler lifetime distribution model parameters
80
C 6.917 1.790 423.1 0.8864
100
C 131.4 2.264 1269 0.8519
The optocoupler degradation failure model is:
 X.-g. Zhang et al. / Defence Technology 16 (2020) 392e400 399

Table 2
Relevant estimates table. yðtÞ ¼ b1 t b2 þ C
a1 a2 R-square sb

27.82 8803 0.8199 2.0355 b1  LN mb ; s2b

Fe ¼ PfTe  tg In the above equation, b1 is random, b2 and C are a fixed con-

As mentioned earlier, for the sake of simplicity, assuming that C stant for the adsorption and leakage mechanisms. Leakage current
is a constant, the degenerate failure model of the optocoupler can degradation rate b1 meets the Arrhenius equation. Its logarithmic
be deduced from the definition of degradation failure: standard deviation sb does not change with temperature. The
relationship between the logarithm mean mb of b1 and temperature
n o
Fe ðtÞ ¼ PfTe  tg ¼ P ½ðD  CÞ=b1 1=b2  t T (
C) satisfies:

a2
mb ¼ a1 þ a2 x ¼ a1 þ
T þ 273:5
¼ Pfb1  ðD  CÞ=t b2 g ¼ 1  Pflnb1  lnðD  CÞ  b2 ln tg


lnðD  CÞ  b2 ln t  ða1 þ a2 xÞ
¼1F
sb (4) The storage life of the optocoupler follows a lognormal dis-
tribution. Its lifetime distribution model is:


ln t  ½lnðD  CÞ  ða1 þ a2 xÞ=b2

¼F (16) ln t  me
sb =b2 Fe ðtÞ ¼ F
se
In the above formula, F{∙} is a standard normal distribution
function. Let:

me ¼ ½lnðD  CÞ  ða1 þ a2 xÞ=b2 Where me ¼ ½lnðD  CÞ  ða1 þ a2 xÞ=b2 and se ¼ sb =b2 .

(5) The preliminary estimate of each parameter is: b

b 2 ¼ 2:141,
se ¼ sb =b2
a
b 1 ¼ 27:82, a s b ¼ 2:0355.
b 1 ¼  8803, b
Then the above equation can be simplified as:
The determination of the failure mechanism helps clarify the

 ðt "  2 # upper temperature limit and temperature selection during the
lnt  me 1 1 1 lny  me
Fe ðtÞ ¼ F ¼ pffiffiffiffiffiffi exp  dy formal test, avoiding over-temperature and failure mechanism
se 2pse y 2 se changes. The determination of degenerate trajectories and life
0
distribution models and the estimation of various parameters are
(17)
the basis for selecting the optimal design for accelerating degra-
In this way, the lifetime distribution model of optocoupler under dation tests. It is also possible to avoid the misadjustment of the life
degradation is deduced. From the above equation, the lifetime is model in the later experimental data processing. The above priori
obeying the lognormal distribution, and the logarithmic mean and information lays a good foundation for both the accelerated
logarithmic standard deviation are respectively me and se. degradation test design of the next step and the formal acceleration
degradation test.
This work is supported by the National Natural Science Foun-
5. Conclusion dation of China of China (No. 61471385).

In this paper, through the data analysis of the optocoupler fail- References
ure mechanism verification test, the prior information of the
optocoupler life assessment is obtained. The conclusions are as [1] Pan Gang, Yu Liang, Meng Ying, ect Jun. Optimization design of step-down-
follows: stress accelerated degradation test on condition of small sample. Instrum
Techn Sens 2013;5:91e3.
[2] Tseng ST, Yu HF. A termination rule for degradation experiments. IEEE Trans
(1) The optocouplers have two independent failure modes, one Reliab 1997;46:130e3.
is sudden failure and the other is degradation failure. [3] Wu SJ, ChangC T. Optimal design of degradation tests in presence of cost
constraint. Reliab Eng Syst Saf 2002;342:109e15.
(2) The maximum temperature stress of the optocoupler cannot [4] Yu HF, Tseng ST. Designing a degradation experiment. Nav Res Logist 1999;46:
exceed 140
C; the increase of the leakage current of the 689e706.
optocoupler is due to the formation of surface leakage [5] Yu HF, Tseng ST. Designing a degradation experiment with a reciprocal
Weibull degradation rate. Qual Technol Quant Manag 2004;1:47e63.
channels which is due to the contaminated movable ions on
[6] Wang Ya-shun, Zhang Chun-Hua, Chen Xun. Study of simulation based
the chip, and the surface leakage current is in direct pro- optimal design for degradation test. J Astronaut 2008;29:380e4.
portion to the amount of adsorption; the increase of leakage [7] Tan Yuan-yuan, Zhang Chun-hua, Wang Ya-shun. Study on simulation based
optimal design method for accelerated testing with competing failure modes.
current causes the tunneling effect of the internal p-n junc-
J Natl Univ Def Technol 2011;33:130e5.
tion of the LED, which in turn affects the performance of the [8] Wang Chang-an, Yue-ting Niu, Xi-hui Mu. Prediction of storage life and
LED and causes it to fail. accelerated-life-testing time for gyroscope. J Chin Inert Technol 2016;24:
683e7.
[9] Xu Ai-bin, Shao-ping Li, Ou Ye-fang, Zheng Ting-gui. Main default modes and
(3) The degeneration trace of the optocoupler leakage current failure mechanisms of domestic optoelectronic coupler. Electron Prod Reliab
can be described by the hybrid effect model as follows: Environ Test 2000;6:7e10.
400 X.-g. Zhang et al. / Defence Technology 16 (2020) 392e400

[10] Shi-man Xiao. Package and corresponding failure mechanism of optical cou- international semiconductor lighting forum, pp.270-273.
plers. Semicond Technol 2011;36:328e30. [21] Li Jun, Fan Guang-han, Yang Hao. Impaction of the interfacial layer on leakage
[11] Pernot C, Hirano A, Amano H, et al. Investigation of leakage current in GaN p-n current of LED. Chin J Quant Electron 2015;25:615e8.
junction. Jpn J Appl Phys Part 2 1998;37:1. 202-1 204. [22] Gao Cheng, Xiong Yuanyuan, Yang Daming. Study on the evaluation method
[12] Min-Woo Ha, Choi Young-Hwan, Lim Jiyong, et al. SiO2 Passivation effects on of plastic package photocoupler's storage life base on POF. Electron Test
the leakage current in AlGaN/GaN high-electron-mobility transistors 2018;2:45e6.
employing additional Schottky gate. Jpn J Appl Phys 2007;46:2291. [23] Sun Quan, Jing Feng, Zheng-Qiang Pan. Long-life product life prediction
[13] Rhoderick EH. Metal-semiconductor contacts. Monograph: Clarendon Press; technology based on performance degradation. 2015. Monograph.
1978. [24] Meeker WQ, Escobar LA. Statistical methods for reliability data. New York:
[14] Ding Ying-ru, Qin Guan-lin. Solid surface chemistry. Monograph: Shanghai John Wiley&Sons; 1998.
Science and Technology Press; 1988. [25] Ramirez JG, Gore WL. New methods for modeling reliability using degradation
[15] Sun Ying, Zhang Chang-qiao. Semiconductor chemistry. Monograph: Shan- data. Stat Data Anal Data Min 2001;8:223e6.
dong Science and Technology Press; 2014. [26] Meeker WQ, LuValle MJ. An accelerated life test model based on reliability
[16] Liu Bin, Lu Rong. Physical chemistry. Monograph: Huazhong University of kinetics. Technometrics 1995;37:133e46.
Science and Technology Press; 2000. [27] Pirin PJ. Reaction kinetics. Monograph: Clarendon Press; 1980.
[17] Qin Zeng-Huang. Electrical engineering. Monograph: Higher Education Press; [28] Chan CK, Boulanger M, Tortorella M. Analysis of parameter-degradation data
2011. using life-data analysis programs. In: Annual reliability and maintainability
[18] Li Xiao-bai, Ma Nong-nong, Hou Xiao-yuan. Current leakage about GaAsFET symposium; 1994. p. 288e91.
surface. J Funct Mater Devices 2013;1:47e52. [29] Liu Juan. Research on step stress accelerated degradation test of electrical
[19] Song Xue, HAN Yan-jun, LUO Yi. Improvement on GaN-based LED reversed connector. PhD. Thesis. Zhejiang University; 2013.
leakage current characteristic by passivation process. Semicond Optoelectron [30] Pan Zheng-Qiang. Accelerated degradation modeling and experimental design
2016;27:164e6. method based on typical stochastic processes. Monograph: National Univer-
[20] Lin Fu-he, Song Qin-qin, “Analysis of the influence of reverse breakdown sity of Defense Technology Press; 2015.
voltage on the reliability of Ga-N based LED tubes,” The 8th China