The end markets serviced by the semiconductor industry are rapidly adopting power semiconductor devices based on wide-bandgap (WBG) semiconductors, including silicon carbide (SiC) and gallium nitride on silicon (GaN-on-Si). Initially, much of the effort from manufacturers using GaN materials has been in the development of high-electron-mobility transistors (HEMTs) for RF applications such as radar.
More recently GaN-on-Si HEMT technology has been developed for power-conversion applications. This new class of power transistor delivers significant efficiency gains over existing Si power MOSFETs. This has direct benefits in terms of the total cost of ownership for power conversion. GaN devices are already employed in such applications and are being used today in front-end PFC circuits in telecom rectifier systems.
Furthermore, GaN enables higher power density; therefore, conversion solutions can be physically smaller. This is particularly important in certain application areas such as networking and data communications, where users seek to add extra features (e.g., backup battery power) without sacrificing other performance in a fixed server rack space.
In short, HEMTs can be beneficial in any application where power conversion takes place at a relatively high voltage, including ac-ac, ac-dc, and dc-dc. GaN technology is applicable in a diverse range of applications, from data center and telecom, to wireless power, motor control, and audio. The supply chains in these markets have matured to the point that it becomes vital for the semiconductor industry to be able to deliver power transistors based on new technology (such as GaN HEMTs) that are (system-) cost-effective and just as reliable as silicon-based power MOSFETs.
Infineon, a manufacturer of GaN HEMTs called the CoolGaN product family, gives insight into its qualification method that serves as a benchmark and provides the baseline for industry qualification standards.
The Need for a GaN Qualification Plan
The industry has been manufacturing silicon-based devices and perfecting the qualification processes involved for more than 50 years. By comparison, GaN HEMT production is still nascent, with relatively scarce historical test data. While some manufacturers now promote their GaN-on-Si devices as being JEDEC (Joint Electron Device Engineering Council) or AEC Q101 qualified, presently these standards are based on silicon device requirements.
Accordingly, JEDEC has established the JC-70 Wide Bandgap Power Electronic Conversion Semiconductor committee to define how the industry at large should address this gap in common qualification standards. JC-70 contains two sub-committees: JC-70.1 for GaN and JC-70.2 for SiC. JC-70.1 has recently published its first guideline (JEP173 at www.JEDEC.org).
Si vs. GaN
To better understand the possible risks of using qualification standards established for silicon devices, it’s informative to study the differences between Si and GaN device structures and materials and the related qualification tests.
Consider the structure of a SJ MOSFET (Fig. 1a), which shows the source, gate, and drain terminals. During normal operation, electrons flow vertically from source to drain, thanks to an inversion of the channel from p-type to n-type. This is caused by applying a sufficiently high voltage across the gate dielectric, as shown by the red arrow.
1. Illustrated are the cross-section of a silicon MOSFET showing a vertical current path (a) and the cross-section of a GaN HEMT showing lateral current flow (b).
In this diagram, the area indicated by the number 1 shows the p-n body diode, which is stressed during high-temperature reverse-bias (HTRB) testing. The area near the number 2 shows the gate oxide, which is stressed by high-temperature gate-bias (HTGB) testing. The area near the number 3 represents device edge passivation and molding compound, which are stressed during temperature cycling and temperature/humidity/bias testing (THB). The area near the number 4 indicates that top-layer source aluminum and bond wires are stressed during high-temperature operating life.
When we compare this structure to that of the GaN HEMT (Fig. 1b), which has a horizontal structure with lateral current flow, it’s clear that there are significant differences. Perhaps most fundamentally is the formation of the two-dimensional electron gas that forms between the aluminium-gallium-nitride layer (AlGaN) and the GaN layer. The GaN HEMT has no p-n drain to source body diode (number 1 in Fig. 1a). There’s no gate dielectric (number 2 in Fig. 1a). For a typical GaN HEMT, the field terminates at the surface and in multiple locations across the surface, as opposed to the Si SJ FET where the high field terminates at the surface only at the edge of the device (number 3 in Fig. 1a).
Because of these differences, Infineon developed extensions to the standard qualification plan to better stress test its CoolGaN family of GaN-on-Si HEMT devices, which target high-voltage power management and conversion applications. Space prevents a full description (consult www.infineon.com/gan, “whitepaper” for a complete discussion), but what follows is a brief summary of the qualification methodology that the company employs for these devices.
To comprehend the device and structure differences outlined above, the qualification approach developed by Infineon involves a four-stage plan (Fig. 2). This includes an application profile, a quality requirement profile, reliability investigations, and degradation models. These stages are described in brief, below.
2. The GaN Qualification Plan developed by Infineon for its CoolGaN family of HEMT power devices.
First, the application profile defines how the device will operate and be stressed in a given application. This sets the expectations based on the end-market and gives the GaN device manufacturer a clear indication of what needs to be delivered. Through close cooperation with lead customers, the parameters for a specific application or category of applications can be defined.
The second stage determines quality requirements for the application, which will include the maximum cumulative failure rate and parametric drift limits, as well as moisture-sensitivity-level rating. The example on which Infineon based its initial CoolGaN qualification is the external environment telecom rectifier, where a 15-year lifetime is required with a failure rate of 1 per billion device-hours (1 FIT). Other applications may have different target lifetimes and failure rates.
The third element is a comprehensive study of the GaN devices’ reliability failure mechanisms. In particular, failure modes must be characterized into two groups. “Intrinsics” are failures caused by inherent wearout of the device materials and structures, while “extrinsics” fail earlier and arise from defects or processing variation. The investigations are complete when an improvement path is identified to eliminate or reduce extrinsics.
Figure 3 shows a plot that identifies fraction of failure versus time. An example of a Weibull plot that’s commonly used in studying failure behavior for reliability, it’s particularly useful to sort intrinsic from extrinsic failures, as shown.
3. Here’s a Weibull plot showing fraction (%) of failed devices versus time. Extrinsic failures fail earlier than intrinsic failures.
Degradation models—the fourth part of the qualification plan—allow users to predict failure rates under real application stress conditions over the intended lifetime. Such models are generated based on accelerated test to failure data: Devices operate under stress well beyond designed use conditions to force failure to occur at much shorter times (tens or hundreds of hours) as compared to an expected 15 years of target lifetime.
Testing multiple groups of samples at varying stress conditions allows for extraction of a mathematical model that can provide the required predicted failure rate at a much longer duration of operation under less stressful, normal use conditions.
GaN Failure Modes
Several failure modes are documented for GaN HEMTs that either don’t apply to silicon power devices or elicit a more sensitive response from GaN devices than from devices fabricated from silicon. These failure modes may represent a potentially high risk of failure during use if not properly accounted for during device/technology qualification.
Such failure modes include dc bias degradation, repetitive (hard) switching failure (DHTOL), temperature humidity and bias, and dynamic RDS(ON). For the sake of brevity, we will focus on one of the four failure modes as an illustrative example; interested readers can explore details of the other three by consulting the Infineon reliability white paper.
One failure mechanism that applies to GaN HEMTs but not to silicon devices is dc bias degradation. While silicon devices are also prone to dc bias failure, resulting in the need to perform HTRB stress testing, GaN HEMTs exhibit a failure rate that depends strongly on both voltage and temperature when tested at accelerated voltage and temperature conditions.
The Weibull plot shown in Figure 4 demonstrates the accelerated stress time to failure data with dc bias and temperature stress applied to samples of Infineon’s 190-mΩ CoolGaN 600-V e-mode HEMTs at varying voltage and temperature. The failure time responds to both voltage and temperature, with very sensitive voltage dependence. Silicon devices don’t behave in this manner.
Note the data shows linear behavior on a logarithmic plot. We can therefore utilize a model for fractional failure in time, F(t), using an equation of the form shown in the purple box. The equation F(t) has an exponential dependence on both voltage and (1/T); the constants γ and Ea can be extracted from the plot in Figure 4.
4. A Weibull plot and model extraction for dc bias failure mode.
GaN-on-Si wide bandgap devices are now available, and they make it possible to achieve higher levels of efficiency and density in power-conversion applications. However, these new materials and device structures behave differently from their silicon-based counterparts with failure mechanisms that pose user risk if not properly accounted for during technology development and device qualification. The industry needs a robust qualification plan that allows manufacturers to demonstrate that their devices can provide a high level of reliability.
Infineon has qualified its CoolGaN GaN-on-Si devices following a four-stage method that models device failure behavior against stresses during application. Detailed application stress conditions were established, including target lifetime and reliability. Extensive reliability investigations illuminated extrinsic and intrinsic failure mechanisms and blazed a path to improving device robustness. Finally, degradation models enable the projection of failure rates based on accelerated testing.
Tim McDonald is Senior Advisor to CoolGaN Technology Development at Infineon Technologies.