The dependence of modern life on IC has become deeper and deeper. From consumer mobile phones to electric vehicle industry, IC cannot be separated from IC. In the future, IC products will develop towards smaller size and finer lines. However, as quality requirements become more and more stringent, The standards and requirements for reliability are becoming more and more refined. Because the manufacturing process and steps of IC manufacturing are very complex, and there are many variables that affect the performance of products, reliability testing is very important. As far as product reliability is concerned, the general definition It is how long it takes from normal use to failure.
IC Three stages of product failure
usually IC The product can be divided into three characteristic stages according to the different failure rate, that is, the bathtub curve graph with well-known reliability. ( picture 1) :
figure 1 bathtub curve
- The first stage Infant Mortality : Represents a process in which the failure rate of the product drops rapidly from a high point in the early stage of use. The reason may be due to poor product design or process design. Because it accounts for a small part of the total population, it will be seen steep drop failure curve , as mentioned above, because the process is getting more and more complicated, how to reduce the failure rate of early death to 0 It is unlikely to be achieved, and customers do not want to receive this product, so how to filter out these premature products, usually Using electric field and temperature to do accelerated life experiments, the so-called burn-in If the foundry can use electrical testing in the wafer manufacturing stage, these early-stage defective products can also be screened out.
- second stage Normal life : That is, the stage of actual use of the product. The failure rate in this stage is the lowest, which may be caused by short-term failure caused by electrical interference, or caused by random external defects, or even some incomplete filtering in the early stage of death. product caused
- The third stage is Wear-Out , Under normal use of the product, the actual product will gradually reach the limit of use, and the failure rate will begin to increase. Finally, according to the collected data, define the reliability life and confirm whether it can meet the needs of customers.
IC Common reliability problems of products, the main difference is IC components with IC Packaging is the main IC Common Component Problems such as Oxide Collapse (oxide breakdown) , may lead to leakage current, or even cause short circuit or partial line failure. In addition, hot carrier injection (Hot carrier injection, HCI) ,by MOSFET As a switching element, if the electrons run out of the normal flow path, it will cause the characteristics of the switch to change, resulting in an unstable situation, so it can be analyzed such as critical voltage, conductivity (transconductance) The decay process of other parameters over time; as for the packaging reliability, most of it is caused by the packaging material and process, so it is common such as temperature cycle test, using high and low temperature cycle test to see whether there is degumming or whether there is any change in electrical properties, in addition to high-speed High voltage test is also one of the common tests, and it is often seen in vehicle reliability such as IC Package testing is similar to reliability testing.
Reliability Analysis Using Statistical Methods
The following will mainly introduce JMP Reliability acceleration analysis platform, this platform can get the life of product failure under more severe conditions, and can calculate the life of failure under normal conditions, and then cite an example Statistical Methods for Reliability Data by Dr. Meeker and Dr. Escobar, John Wiley & Sons, 1998 , we want to understand the breakdown of the dielectric and do accelerated experiments at given different voltages (100.3, 122.4, 157.1, 219.0, kV) , want to know about the 50kV What is the life under ( picture 2) , it can be observed that as the voltage is stronger, the life is shorter, and the life dispersion under each voltage is significantly different.
figure 2 Linear coordinates
Then convert the axes to logarithms - Logarithmic presentation ( picture 3) , the dispersion range under each voltage will become close, so when the accelerated experiment is used for analysis, the data needs to be converted to apply the linear model analysis, and then it will be more accurate to estimate the life under normal conditions of use voltage.
image 3 logarithm - logarithmic coordinates
Using the reliability acceleration analysis platform, it can be observed at voltage = 219kV When , a few points are offset from the fitted straight line, but most of them still fall on the line, so the logarithmic normal distribution can get a good fitting effect with a fixed slope ( picture 4) ,
Figure 4 distribution fitting
And according to JMP The selected model is a three-parameter regression model, and the voltage will significantly affect the performance of life, although the models with different position and scale parameters are not significantly different from the regression model. ( picture 5) , you can collect several more sets of data to verify, and finally store the prediction formula ( picture 6) ,
Figure 5 Model selection results
picture 6 prediction formula
If the model is verified correctly, it can be added at voltage = 50kV The distribution curve below ( picture 7) ,learn 10% , 50% , 90% The position where the quantile fails, or enter the corresponding value directly on the table ( picture 8) , or with Profile( picture 9) What is the probability of failure.
Figure 7 Scatter plot
Figure 8 Failure Probability Data Sheet
Figure 9 Profile
When fitting the model, the data transformation can make it easier to use the linear model to estimate, reducing the deviation of the curve model estimation or the occurrence of excessive variation. After adding the data scale transformation, it is easier to observe and confirm the data fit. JMP And can compare the model differences of different effects and select a suitable model, and finally provide the estimated formula and Profile for forecasting.
Reference information:
- semiconductor IC Product Reliability Statistical Physics and Engineering Second Edition
- JMP web video : https://www.jmp.com/en_in/events/ondemand/best-practices-in-reliability-data-analysis/accelerated-life-testing.html
- source : http://reliawiki.com/index.php/Inverse_Power_Law_(IPL)-Lognormal_Model
