Wafer-Level Burn-In for Improved WBG Device Reliability
Silicon carbide (SiC) and gallium nitride (GaN) power devices are seeing strong adoption across various sectors, including automotive, industrial, and consumer electronics. Each application has a different mission profile and reliability requirements.
Wafer-level burn-in (WLBI) forms a vital, early-detection screening technique to weed out infant mortality failures, detect process shifts, and improve package and module yields and reliability.
Key reliability concerns in SiC and GaN
SiC MOSFETs have seen strong adoption in electric-vehicle traction inverters, industrial motor drives and robotics, renewable energy systems, energy grid and data centers, and more. The use of 1,200-V and higher-rated devices creates efficiency and power density improvements over legacy silicon devices.
GaN HEMTs in the 650-V and lower-voltage class have seen high growth in consumer power adapters, DC/DC converters in data center power supplies and automotive, LiDAR systems, and solar inverters. Low parasitics and switching losses in GaN HEMTs enable high switching frequencies and novel converter topologies. Significant reliability improvements have been made in both SiC and GaN power devices over the last decade, although manufacturing defects still occur.
SiC concerns
The gate oxide in SiC MOSFETs has long been considered the weakest link in terms of failure rates. Process improvements have now made the intrinsic defectivity of this oxide close to that achieved on silicon devices, with time-dependent dielectric breakdown (TDDB) tests showing lifetimes exceeding 100 years at 175°C under maximum-voltage operating conditions.
Extrinsic defects at the gate oxide and its interface are, however, much higher in SiC due to the presence of carbon at the interface. These can form trapping sites that reduce channel mobility in SiC MOSFETs below theoretical estimates.
Other defects can cause infant mortality failures by creating weak spots from localized thinning or trap-assisted tunneling. These can result in high gate leakage and gate oxide rupture. Basal-plane dislocations in the SiC bulk can grow under body-diode forward-conduction mode and degrade the device’s off-state characteristics while increasing its on-state resistance, resulting in application system failure.
Automotive applications demand high quality and reliability, with silicon devices achieving failure rates close to 0 parts per billion over the car’s lifetime. Stricter qualification standards such as AEC-Q101 for devices and AQG-324 for modules, mission profile testing, advanced inspections at the process and packaging stages, and over six decades of process and device experience with predictable failure modes and models have all contributed to this tremendous improvement in silicon device reliability.
SiC is still in a learning phase when it comes to understanding all device failure modes and creating failure models. JEDEC reliability guidelines and standards such as AQG-324 are being continuously updated to add more qualification tests or supplement existing ones.
GaN concerns
The lateral GaN HEMT can have high, non-uniform electric fields under high drain bias that result in charge trapping. The increase in its dynamic on-state resistance is well-understood, and improvements in design, epi growth conditions, dielectric passivation, and metallization have all converted this into a legacy, well-controlled issue.
Specific GaN HEMT failure mechanisms have been covered in standards such as JEP180 and recommended tests such as the dynamic high-temperature operating-lifetime and switching accelerated-lifetime tests.
Increased leakage current that can cause eventual gate shorts from high fields across the Schottky gate/p-GaN layers is another well-understood failure mode, along with stress-induced defects from lattice and thermal expansion coefficient mismatches between GaN and the underlying silicon substrate.
Advantages of WLBI for WBG power modules
Semiconductor devices typically follow a bathtub curve shape of failure rate plotted on a logarithmic time axis (Figure 1). The constant central part of this curve represents an exponential distribution with a constant failure rate λ (expressed in failures per million hours). λ can be determined by reliability and qualification tests on packaged parts that can last over 1,000 hours. This helps predict the failure-in-time rates and lifetime of the part.
The early failure rate is high, with a decreasing failure rate over time that can be generated by defects such as particles, thin-oxide pinholes, and scratches. These follow a log-normal distribution that has a short mean lifetime and a high sigma (variability).
The time-0 early failure rate R can be orders of magnitude higher than λ in the constant failure region. When n number of non-redundant devices are part of a power module, as would be the typical case in an EV traction inverter that can have many (>20) MOSFETs in parallel, the net failure rate would be n × R.
Accelerated stress testing
Accelerated stress testing in automated test equipment at wafer probe is one way to weed out these early failures. Due to the requirement for multiple touchdowns to cover the entire wafer, overall test time concerns typically restrict these to the order of a few seconds. Acceleration models need to be well-understood to apply the stress biases, which may not be the case for many failure modes in WBG devices due to the maturity level.
The risk of overstressing results in walking-wounded die that are damaged from the stress test itself. Because the early-life period of the bathtub curve covers several hours, a burn-in test on this time scale that balances device stress voltage and temperature conditions to avoid inducing failures can be more effective and safer for screening out early-failing dies.
Given the relatively high die cost of WBG devices, production WLBI systems provide a test solution that can perform infant mortality screening much more efficiently and cost-effectively than module-level burn-ins, as illustrated in Figure 2.
Aehr Test Systems FOX-XP WLBI platform
Aehr Test Systems is a global leader in semiconductor test and burn-in solutions. Its FOX-XP system, shown in Figure 3, is a multi-wafer test and burn-in system.
Tester ‘blades’
The FOX-XP accommodates up to 18 tester “blades,” allowing parallel testing and burn-in of up to 18 wafers at a time. A modular blade architecture is used whereby each blade is configurable with up to eight modules or instruments per blade.
In a SiC or GaN configuration, this would typically include a universal channel module for measuring results, a bipolar-voltage channel module with positive and negative capability for driving the gate, and a very high-voltage channel module providing up to 2,000 V on the drain for reverse-bias testing.
Each blade is independent and can run a different wafer or recipe than other blades, with up to 3.5 kW of power delivered to and dissipated from each wafer.
Wafer probe interface
The FOX WaferPak is a wafer probe interface that includes a thermally controlled wafer chuck and full wafer contactor housing the wafer. The WaferPak can accommodate wafer diameters from 4 to 12 inches with temperature capability up to 150°C.
Highly durable and inexpensive micro pogo pins are typically used to make contact with the wafer bond pads. Up to 50,000 pins can be placed on a single WaferPak, ensuring a single touchdown can contact the entire die on wafers up to 300 mm in diameter. During testing, a vacuum-sealed cavity is created at the wafer, and during high-voltage stress testing, a nitrogen inert gas flow is introduced to prevent arcing.
Aligner
A FOX WaferPak Aligner can be integrated with the system to automatically do pin-to-pad alignment of the wafers and pack into the WaferPak. The WaferPaks are then automatically inserted into each blade in the FOX-XP tester for burn-in, then unloaded and the cycle repeated, all under SECS-GEM control, allowing unattended, lights-out operation.
Example of FOX-XP test capability
The FOX-XP can be used to perform a range of burn-in and characterization/process monitoring tests, such as:
- High-temperature gate bias with VGS ranging from 40 V to –30 V, monitoring IGSS gate leakage
- High-temperature drain bias with VDS up to 2,000 V, monitoring IDSS drain leakage
- Body-diode forward-bias stress, with a total source current of >1,000 A/wafer
- Threshold voltage measurements
- TDDB measurements for gate oxide reliability monitoring
For TDDB testing, up to 128 VGS levels can be distributed across all dies on the wafer, enabling extensive TDDB gate characterization in a single burn-in cycle.
Figure 4 illustrates results obtained from this test. The left part of the figure shows a binned wafer map. Here, the bins correspond to the VGS of the failing die. The right plot shows the time plot of each failing die.
The time at failure is tracked for each die, which can then be used to construct the typical Weibull plots to determine TDDB lifetimes. The test software provides 100% traceability for wafer and die location across all tests.
WLBI and device characterization
Aehr Test Systems’ FOX-XP is a modular test platform that performs production WLBI and device characterization with a high degree of parallelism. WLBI can effectively screen out infant mortality failures on WBG devices, making it an essential screen not just in automotive applications such as EVs but also in other industrial applications such as data center power converters, central solar inverters, telecom/5G, and robotics.
The company also manufactures an engineering FOX-NP test system for reliability and characterization, along with pilot-line development and monitoring. This one- to two-wafer system is fully hardware- and software-compatible with the production FOX-XP system and supports manual loading of WaferPak, as well as singulated die/module testing with DiePak carriers.
Cover image: Adobe Stock
The post Wafer-Level Burn-In for Improved WBG Device Reliability appeared first on Power Electronics News.