中文 English

Improving Reliability For GaN And SiC

Why these chips are gaining ground, and what still needs to be addressed.

popularity

Suppliers of gallium nitride (GaN) and silicon carbide (SiC) power devices are rolling out the next wave of products with some new and impressive specs. But before these devices are incorporated in systems, they must prove to be reliable.

As with previous products, suppliers are quick to point out that the new devices are reliable, although there are some issues that can occasionally surface with GaN and SiC. Additionally, the reliability requirements for these devices are becoming more challenging for the latest mission-critical applications like automotive. To meet the challenges, the devices may require more and even new reliability testing methods.

Concerns about reliability are not new in the semiconductor industry, and they have been growing recently as the advanced semiconductor content in cars continues to rise and as chips are used for more mission-critical applications such as data centers. At the same time, all circuits wear out over time due to a variety of factors. The key is to be able to predict and prevent failures, and to determine the acceptable range of operating conditions for a particular part.

“Reliability is mean-time-to-failure,” said John Palmour, CTO at Cree/Wolfspeed. “We characterize reliability through a lot of accelerated life testing. You test to fail. You figure out what the accelerating factors are, so that you can project back to normal conditions and predict a lifetime. When I talk about reliability, I’m talking about the fundamental physics of the device. Quality is different. Quality is how many parts per million fails do you have or parts per billion.”

All device types undergo reliability testing, including power semiconductors. Power semis are specialized transistors that boost the efficiencies and minimize the energy losses in high-voltage applications like automotive, power supplies, solar and trains. Power semis operate like a switch in systems, allowing the electricity to flow in the “on” state and stop it in the “off” state.

Power semis are split into two camps—silicon and wideband gap. Silicon-based devices are mature and the reliability issues are understood. In comparison, GaN and SiC power semis are based on wideband-gap technologies, which are more efficient with higher breakdown electric field strengths than silicon. But GaN and SiC are newer technologies with various idiosyncrasies. So customers may want to gain a deeper understanding of the reliability issues for these technologies.

What are power semis?
Power semiconductors are used in the field of power electronics. Using solid-state devices, power electronics control and convert electrical power in systems. These include cars, cellphones, power supplies, solar inverters, trains and wind turbines.

Power semiconductors play a key role in the conversion process. There are different types of power semis, and each one is denoted by a numerical figure with a “V,” or voltage. “The ‘V’ as in VDSS is the maximum allowed operating voltage, or drain-source voltage specification,” explained Alex Lidow, chief executive of Efficient Power Conversion (EPC). “The terminology ‘DSS’ means drain-to-source with the gate shorted.”

The power semiconductor market is dominated by silicon-based devices, but GaN and SiC are making significant inroads. The silicon-based devices include power MOSFETs, super-junction power MOSFETs and insulated-gate bipolar transistors (IGBTs).

Power MOSFETs are used in lower-voltage 10- to 500-volt applications, such as adapters and power supplies. Super-junction power MOSFETs are used in 500- to 900-volt applications. Meanwhile, IGBTs, the leading midrange power semiconductor devices, are used in 1,200-volt to 6.6-kilovolt applications.

IGBTs and MOSFETs are mature and inexpensive, but they are also reaching their limits. That’s where GaN and SiC fit in. These two wideband-gap technologies enable devices with higher efficiencies and smaller form factors. For example, SiC has 10X the breakdown electric field strength and 3X the band gap over silicon. GaN exceeds those capabilities.


Fig. 1: How power switches are categorized. Source: Infineon

SiC is making inroads in several markets, particularly automotive. “The smaller form factor and performance of SiC devices is attractive for electric vehicle and hybrid electric vehicle power train applications,” said David Haynes, managing director for strategic marketing at Lam Research.

GaN is gaining traction in automotive, data centers and other markets. “It is ideally suited for addressing high-volume applications in rapid charging solutions because of its superior performance at high frequencies,” Haynes said.

Reliability is important for all power semis. The goal for any device is to achieve zero failures. In operation, a device may work for a set period or forever. But at times, a product may wear out or fail.

To ensure product reliability, vendors for years have followed the same steps. They are:

  • It’s not good enough to test a device based on the spec sheet, so vendors run various accelerated tests on a device. Generally, the goal is to cause a device to fail.
  • Then, a vendor searches for the failure mechanisms and solves any potential problems.
  • From there, models are developed.

“Once the lifetime under accelerated stress is obtained, the known accelerated model can be used to predict the product lifetime under normal application stress,” according to Alpha and Omega Semiconductor, a supplier of power semis.

It’s not quite that simple, though. Devices are subjected to a battery of accelerated tests. For example, high temperature reverse bias (HTRB) is one common test. HTRB examines the junction degradation of a device under temperature. For this, devices are placed in a specialized HTRB burn-in test system and then subjected to high voltages and temperatures.

That’s just one of many tests. Typically, these tests satisfy the requirements for various reliability standards, such as AEC-Q101 and others. AEC-Q101 defines the minimum stress test for a given component.

SiC and reliability
Each power semi type is different. For example, the power MOSFET is a vertical structure. The source and gate are on the top of the device, while the drain is on the bottom. When a positive gate voltage is applied, a channel is formed between the source and drain.

In the latest devices, “the gate oxide is becoming thinner and thinner; therefore, the electrical field is getting higher and higher,” according to Alpha and Omega Semiconductor.

Over time, the gate oxide may degrade in MOSFETs. Generally, the failure mechanism behind this phenomenon is called time-dependent dielectric breakdown (TDDB). TDDB occurs when the gate oxide breaks down after a period of wear and tear.

To test the reliability of this structure, the device is placed in a specialized TDDB test system and then stressed. TDDB and other failure mechanisms are well understood in MOSFETs.

That’s not always the case for SiC, which is a compound semiconductor material based on silicon and carbon. SiC devices are used in 600-volt to 10-kilovolt applications. Electric vehicles are the biggest market for SiC devices, followed by power supplies and solar inverters.

There are two SiC device types—SiC MOSFETs and diodes. SiC MOSFETs are power switching transistors. A SIC diode passes electricity in one direction and blocks it in the opposite direction.

These devices are produced in 150mm fabs, although 200mm is in R&D. In the production flow, a SiC substrate is developed. An epitaxial layer is grown on the substrate and then processed into a device.

In the flow, the SiC substrates are prone to defects. “Substrate cost, availability and quality are still a challenge,” Lam’s Haynes said. “But defectivity in the wafers and epitaxy are improving.”

Once the wafers are processed in the fab, they are diced and packaged, which is a difficult process. “Silicon carbide is the third hardest compound material on earth,” said Meng Lee, director of product marketing at Veeco. “Due to the high hardness and brittleness of SiC, manufacturers are facing cycle time, cost and dicing performance challenges.”

Despite the challenges, SiC suppliers for years have been shipping products with proven reliability. “A while ago, there was a lot of concern about silicon carbide being fundamentally reliable,” Cree’s Palmour said. “We passed that stage awhile ago. There wouldn’t be any automotive or industrial customers if we couldn’t show fundamental reliability.”

Still, vendors pay close attention to these issues. “Achieving fundamental reliability in silicon carbide is definitely harder because of several things,” Palmour said. “In silicon, it’s well characterized. You design it, put it in a fab, and you do a quality test. You just assume it’s reliable. In silicon carbide, you can’t assume that. You have to understand what the failure mechanisms are.”

Reliability is important in all markets, although some have more stringent specs. For example, the bar is considerably higher where safety is a concern. Acceptable failure rates for automotive are in the parts per billion (ppb) range.

Achieving sufficient reliability starts in the design phase. Then, once devices are developed, they are subjected to various accelerated tests. Devices undergo stress tests for humidity, power cycling, temperature, voltage, and others.

It would take tomes to describe each test. But in general, there are two main reliability issues for SiC devices — gate-oxide and threshold voltage stability.

Like power MOSFETs, SiC devices are vertical devices. SiC uses the same gate oxide material, silicon dioxide, as MOSFETs, but SiC devices operate at higher internal fields. So in operation, the gate-oxide materials may encounter shorter lifetimes.

Generally, though, the gate-oxide issues are understood in SiC. TDDB is the failure mechanism here. “There is also a lifetime limitation in the blocking state, while the device is blocking high voltage. In this condition, the point of failure is generally where the oxide sees the highest electric field from the SiC. Which of these mechanisms dominates the lifetime is dependent on the device design that is being made,” Palmour said.

Nonetheless, the gate-oxide issues have been largely solved. “That was one of the fundamental fears way back when,” he said. “If you do your design right, you can get around that problem.”

Still, it’s important to ensure the gate-oxide is reliable. For this, Cree uses high temperature gate bias (HTGB) and TDDB for on-state testing. HTGB is a burn-in test, which is used to stress the gate oxide. In addition, Cree uses HTRB to determine the blocking lifetimes.

While this issue largely has been addressed, the industry may want to reduce the early TDDB failure probabilities for the gate oxide. “To make SiC MOSFETs as reliable as their (silicon) counterparts, one has to minimize the gate-oxide defect density during processing, and implement clever screening techniques that identify and eliminate potentially weak devices,” said Thomas Aichinger, an engineer from Infineon, in a recent paper.

For this, Infineon recently described a new procedure called a “marathon stress test.” Infineon’s test is capable of stressing 3 x 1,000 SiC MOSFETs in parallel. The devices are packaged and mounted on a board. A stress test is performed in a furnace at elevated temperatures.

“It is important to distinguish between traditional TDDB tests, traditional HTGB test and our new marathon test. All three are stress tests for the gate oxide, but they investigate different failures mechanisms,” Aichinger said. “A marathon test is very similar to a TDDB test; however, there are two important differences. Firstly, the marathon test is performed at a much lower gate bias compared to a TDDB test, because its goal is to detect only early device failures caused by critical extrinsic gate oxide distortions. Most devices, i.e. devices without critical extrinsics, do not fail during the marathon test. Secondly, in a marathon test, a much larger number of devices has to be tested (typically >1000 pcs). This is because devices with extrinsic gate oxide distortions are typically rare and testing large numbers increases the chance of finding some which is necessary to verify a certain extrinsic GOX FiT rate.”

Besides the gate-oxide, the other big issue with SiC is threshold voltage instability. “The threshold voltage of the MOSFET can move depending on bias,” Cree’s Palmour said. “That’s a well-known phenomenon and everybody looks at that.”

Threshold voltage instability is caused by a failure mechanism called bias temperature instability (BTI). BTI is a degradation phenomenon in transistors.

HTGB is one way to test the problem, although this is still a work in progress. The SiC industry is discussing these and other issues in the JEDEC JC-70 group. The goal of this group is to devise a standard testing technique and unify the various specs in the arena.

This is particularly important in automotive, where OEMs are demanding zero defects in devices. “The fundamental reliability of silicon carbide is proven and it’s good. Now, it’s about how do you prevent escapes, early fails and things like that,” Palmour said. “The challenges that we have now going into automotive is to address the very high-quality expectations you have for automotive devices. They want to know what the mean-time-to-failure is. They are also concerned about ppm or ppb fail rates.”

This in turn requires a series of defect inspection and screening steps. Like reliability testing, this is also a challenging process.

GaN issues
Meanwhile, GaN, a binary III-V material, is used for LEDs, power semis and RF devices. GaN-based power semis are used in automotive, data centers, military-aerospace and other apps. GaN power semis range from 15 to 900 volts.

GaN devices are made in 150mm fabs. In EPC’s GaN flow, a thin layer of aluminum nitride (AlN) is deposited on a substrate, followed by a GaN layer. A source, drain and gate are formed on the structure, forming a lateral GaN device.

GaN, which is less mature than SiC, has some issues. “We hear of defects causing reliability issues later in the line,” said Kevin Crofton, senior vice president at KLA. “MOCVD processes and subsequent steps can create particles, which current inspection tooling is not always capable of finding. We hear device makers are asking for better materials for compound semiconductors. It’s all about learning, but our sense is that material supply and quality is improving.”

For that reason, customers tend to ask more questions about the reliability of GaN. “GaN has been phenomenally reliable,” EPC’s Lidow said. “We have only three failures in 123 billion hours. That’s about two orders of magnitude better than MOSFETs. But a lot of questions are asked because it’s relatively new.”

Still, like all power semis, GaN devices are subjected to no less than nine reliability stress conditions. Each condition has a test. In just one example, the gate electrode in a device undergoes a stress test. For this, vendors use an HTGB test. Typically, TDDB is the failure mechanism here. More specifically, the silicon nitride layers may fail, not the GaN layers.

That issue is well understood. For GaN devices, though, dynamic on-resistance is the biggest issue. “They used to call it ‘current collapse’ because the device on-resistance would go to infinity. Once GaN devices became commercialized, the magnitude of the shift in on-resistance was greatly reduced,” Lidow said.

Dynamic on-resistance is problematic. For example, a GaN part might have a 1-milliohm spec. But after 100 hours in the field, the same part can turn into a 10-milliohm device.

For some time, the industry theorized that dynamic on-resistance is caused by one failure mechanism called hot carrier injection (HCI). “What happens is that if you send an electron across a very high electric field, it will gain a lot of energy. It’s like being shot out of a gun. And it becomes a very energetic electron. It has enough energy to penetrate certain layers in your device and get trapped,” Lidow said.

The traditional way to test this condition is HTRB. In many respects, though, HTRB falls short. In response, EPC developed a proprietary microscopy technique. It provides a view of the photons that are emitted when energetic electrons get trapped. Using the technology, EPC has found a way to solve the HCI problem.

Other problems also can surface. For example, EPC’s latest GaN devices have new and breakthrough power densities. “We are going to find new failure mechanisms as we increase the power density by 300X. The first problem will be electromigration limits in the conduction layers. Our new designs are EM limited, so we will have to innovate our way out of that physics. One way is to integrate multiple power devices on the same chip. In so doing, you can actually reduce the number of power-in and power-out terminals. Those are the ones that are EM limited,” Lidow said.

“A second challenge is to increase the internal electric fields inside the device,” he said. “The size of the device scales inversely with this peak electric field. If we could tolerate peak electric fields equal to the critical breakdown field of GaN, we could make our devices about 100X smaller. The problem is that dynamic on-resistance is eliminated by keeping peak fields below a certain limit. We need to figure out how to increase peak fields without re-introducing dynamic on-resistance. The solution involves improving material quality, improving dielectric quality, improving device design, and probably several things that we have yet to discover.”

Conclusion
Clearly, GaN and SiC devices are intriguing technologies. The industry is just beginning to understand their potential and properties.

That’s why this technology is making inroads in the market, and it’s happening mostly at the expense of more traditional technologies.

Related Stories

The Ins And Outs Of Silicon Carbide

SiC Foundry Business Emerges

Power Semi Wars Begin



2 comments

Kevin says:

Excellent article, fascinating also. many thanks.

David Leary says:

Understandable that transistor instability in these materials is driven by same mechanisms (BTI + HCI) as silicon transistors. This would suggest that circuits using these materials would benefit from aging-aware design and production test guardband methods that the silicon IC industry has developed.

Leave a Reply


(Note: This name will be displayed publicly)