Accelerated degradation data of SiC MOSFETs for lifetime and Remaining Useful Life assessment
Introduction
Estimating the Time-To-Failure (TTF) or the long-term performance of highly reliable components is particularly challenging. For aeronautic or space applications, the embedded equipment are expected to properly operate for years and even tens of years. Thus, only few devices will fail or degrade during the length of a test carried out in nominal operating conditions.
Usually, the reliability of a component is estimated using data collected during Accelerated Lifetime Tests (ALT). This kind of test consist in applying high level of experimental factors or covariates (temperature, humidity, voltage, current…) in order to obtain failure time data quickly. Then, the TTF and the long-term performance at normal use conditions are extrapolated using the data collected during accelerated tests.
In some reliability studies, it is possible to follow the degradation process over the ageing time. This approach could be especially suitable for mechanism generating “soft” failures which are characterized by the degradation of the device’s performance to an unacceptable level. Many approaches exist to model the degradation of components under severe conditions. In particular Meeker and Escobar [1], Lu and Meeker [2] describe regression-type methods linked to the general degradation path model. Bogdanavicius and Nikulin [3], Lawless and Crowder [4], Park and Padgett [5] or Fouladirad and Grall [6] have presented results obtained using stochastic process (Brownian motion or Gamma process) in order to model degradation phenomenon.
This last approach, which is based on stochastic process, has the double advantage of allowing an estimation of the TTF of a component prior to its integration and to provide a prognostic of the Remaining Useful Life (RUL) during the operation.
This paper is organized as follows Section 2 describes the failure mechanism studied and the associated accelerated tests procedures. Section 3 presents the mathematical aspects of the gamma process and how it can be used for degradation modelling. Then Section 4 focuses on estimating the acceleration factors for both the voltage and the temperature accelerations. In the last section, the estimation of the RUL using gamma process is described.
Section snippets
Observation of threshold voltage instability mechanism
The SiC (Silicon Carbide) MOSFETs (Metal Oxide Semiconductor Field Effect Transistor) was one of the most anticipated wide-band gap power switches. Its physical properties enable the use of such devices in high temperature environment and at higher frequency than equivalent silicon-based devices. Moreover, the isolated gate structure provides an easily controllable device compared to other Wide Band Gap Power electronic structures such as GaN HEMTs or SiC JFETs. Nevertheless, the poor quality
Use of stochastic process for reliability assessment
The use of a gamma process in the field of reliability has been introduced in 1975 by Abdel-Hameed [13] to model gradual damage monotonically accumulating over the time. A gamma process is a stochastic process with independent non-negative increments having a gamma distribution with identical scale parameter. Since then, it has been widely used for maintenance optimisation models based on degradation data [14].
Due to the uncertainty of the deterioration process, it can be regarded as a
Acceleration factors extraction
The first objective of this study is to make it possible to estimate the TTF of a device from the degradation data collected from the accelerated tests described in Section 2. To do so, the complete process carried out, in this aim, is depicted in Fig. 3.
The very first step consists in modelling the degradation data.
With respect to the observed data we propose to use a non-homogeneous gamma process as described in Section 3 and we define the time dependence of this process through the shape
Prediction of Remaining Useful Life
As mentioned previously, the use of gamma process is also relevant for prognostic applications. Indeed using this approach it is relatively straightforward to estimate a Remaining Useful Life and the probability density function associated from degradation data.
The Remaining Useful Life (RUL) of a device or a system is defined as the amount of time from the current instant to the End of Life (EOL) of the system. The RUL estimation is required for the implementation of condition based
Conclusion
For the purpose of this article we have proposed to model the threshold voltage instability mechanism on a SiC MOSFET using a non-homogeneous gamma process. As far as the authors know this study is the very first attempt to model this phenomenon by a stochastic process.
Because of the long term reliability required for aeronautical and space applications along with the short amount time available to age the devices, the assessment of the power SiC MOSFETs reliability have to be performed through
References (16)
- et al.
Condition-based maintenance for a system subject to a non-homogeneous wear process with a wear rate transition
Reliab Eng Syst Safety
(2011) - et al.
Revisiting MOSFET threshold voltage extraction methods
Microelectron Reliab
(2013) - et al.
Statistical methods for reliability data
(1998) - et al.
Using degradation measures to estimate a time-to-failure distribution
Technometrics
(1993) - et al.
Estimation in degradation models with explanatory variables
Lifetime Data Anal
(2000) - et al.
Covariates and random effects in a gamma process model with application to degradation and failure
Lifetime Data Anal
(2004) - et al.
Stochastic degradation models with several accelerating variables
IEEE Trans Reliab
(2006) - et al.
Effect of interface states on electron transport in 4H-SiC inversion layers
IEEE Trans Electron Devices
(2001)
Cited by (46)
Performance assessment and RUL prediction of power converters under the multiple components degradation
2023, Microelectronics ReliabilityAn improved non-Markovian degradation model with long-term dependency and item-to-item uncertainty
2018, Mechanical Systems and Signal ProcessingCitation Excerpt :Among these categories, data-driven methods only need to establish a stochastic model to fit the observations without any assumption of physical parameters or additional expertise, and thus are more commonly used for the prediction. As far as we know, most of the existing data-driven methods are limited to various types of memoryless Markovian degradation models [14], including Markov chains [15–19], Wiener processes [20–25], and Gamma processes [26–30]. Wang et al. [16] presented a continuous hidden Markov model to describe the degradation states of tool wear, and predicted the RUL of the system based on the milling force signal.
SiC MOSFET threshold-stability issues
2018, Materials Science in Semiconductor ProcessingCitation Excerpt :Instead, it summarizes recent efforts at the U.S. Army Research Laboratory (ARL) in understanding the potential failure mechanisms associated with threshold-voltage (VT) drift [9,10], and our efforts to develop an improved test method for separating out good devices from bad with regard to VT stability [11]. ARL's first paper describing this shift, or drift, in the VT during gate-bias stressing was published over ten years ago [12], with many other research groups around the world subsequently reporting similar behavior [13–28]. Significant improvements in device processing in recent years have led to much, much smaller VT shifts by major commercial SiC MOSFET manufacturers [29,30].
Measurement considerations for evaluating BTI effects in SiC MOSFETs
2018, Microelectronics ReliabilityCitation Excerpt :Threshold voltage (VT) instability was first identified as a potential reliability issue for SiC MOS devices over a decade ago [1]. Since then, it has become a widely acknowledged and studied phenomenon [2–17]. Instabilities of commercial devices have improved over time [18,19], although they are still significantly larger than that of Si MOSFETs [18] and it is imperative to monitor the situation as new suppliers enter the marketplace.
Non-homogenous gamma process: Application to SiC MOSFET threshold voltage instability
2017, Microelectronics ReliabilityCitation Excerpt :Once the lifetime density function is estimated, some position parameters can be extracted such as the Mean Time to Failure, the 100-th percentile or the median time to failure. The knowledge of these parameters makes it possible to compute an acceleration factor in order to extrapolate the lifetime of a component in nominal conditions from the data collected during accelerated tests as presented in our previous study [15]. However, the time needed to gather all the data required could be extremely long.
A prognostic method for predicting failure of dc/dc converter
2017, Microelectronics ReliabilityCitation Excerpt :The prognostic routine includes a methodology for accelerated aging by using purely electrical overstress and corresponding identification of a precursor of failure. Santini et al. proposed to model the threshold voltage instability mechanism on a SiC MOSFET by using a non-homogeneous gamma process to perform the estimation of RUL over the course of the aging process [5]. To create a real-time failure prediction of an electrolytic capacitor applied to a power-electronic system, Abdennadher et al. [6] presented an electrolytic-capacitor failure prediction technique to detect the changes in the equivalent series resistance and capacitance C values in real time.