President, Chief Executive Officer & Chairman, POWER INTEGRATIONS INC
PowiGaN - Quality, Robustness and Reliability
Power Integrations has full control of the manufacturing process of its PowiGaN devices, which includes extensive tests ...
About Balu Balakrishnan
Balu Balakrishnan, President, CEO, and Chairman of Power Integrations, has been discussing the company's advancements in gallium nitride (GaN) technology and its role in energy efficiency. At PCIM Europe 2024, he highlighted the introduction of the InnoMux 2 product, which he described as the first to generate multiple outputs with a single magnetic element, eliminating DC-DC converters and increasing efficiency by 10%. He also stated that Power Integrations is the only company commercially offering 1250-volt GaN, and that the company's cascade structure for GaN is more reliable than enhancement-mode alternatives, particularly for data centers and automotive applications. Balakrishnan noted that 70% of carbon reduction will come from energy efficiency and renewables, and that Power Integrations aims to capture a share of the anticipated $1 trillion in carbon reduction spending over the next 20 years. In other appearances, Balakrishnan emphasized the company's strategy of developing proprietary technologies, controlling manufacturing processes at partner foundries, and focusing on breakthrough products. He discussed the quality and reliability of PowiGaN devices, stating that the company runs extensive tests and aims for zero defective parts. He also promoted the InnoSwitch3-TN flyback switcher IC for smart appliances, citing its flat efficiency from full load to light loads and its ability to deliver up to 15 watts without a heatsink. Balakrishnan expressed confidence that GaN will rapidly replace silicon in power switches, and that Power Integrations' integrated solutions make GaN easy to use for customers.
Transcript (14 segments)
B
Balu Balakrishnan0:00
Foreign. [Music] And I'm going to talk to you about GaN quality and reliability. So let's take a look at the bathtub curve. On the x-axis here I have log of time, on the y-axis failure rate, and then the green line here is the rate of failure of a typical product over time. In the early phase of a product's life, you get quality-related, mis-manufactured products. So this is our initial quality part of the bathtub curve. In the far distant time, you have parts that are worn out. They've had some kind of failure mode that has been exercised over time and these parts have worn out. And in the center section, we have devices that were well manufactured, they haven't worn out yet, but they're dead anyway. Well, what happened here? Typically these parts have been overstressed.
So let's talk about the far distant future, the wear-out mechanisms that might affect a GaN device in, well, let's say more than 11 years, which is for context where a hundred thousand hours is, or a century, which would be a million hours. And we can't wait around for a century for our GaN to fail; we want it to fail now. So the best thing to do here is we have to find a way to accelerate time. And the way we accelerate time in the semiconductor industry is we heat things up, we apply voltage to them, and we use humidity.
So to accelerate time at the IC level, we run a test called High Temperature Operating Life. And this test is designed to run a relatively small number of parts, but five lots times 48 parts for 1,000 hours. And that gives us a very good view of whether there is a process dependency and whether any one of these devices has been exercised into failure during the test. And then there's another test we do called HALT, which applies a different, in fact increased, humidity at a slightly lower temperature using the same number of parts for the same amount of time.
But GaN is a new technology, and so we want to be doubly sure that the device inside the IC is robust and long-lived. So we do tests specifically designed for GaN. One of those is HTRB, High Temperature Reverse Bias test, and High Temperature Gate Bias test. And those tests are also a thousand hours at elevated temperature under voltage stress, under a DC voltage stress, to make sure there are no migration issues or long-term degradation mechanisms associated with high voltage at high temperature.
We also do something called Hot Carrier test, and that's intended to find any traps where electrons in the channel can get diverted into oxides and become trapped. We do an electromigration test to make sure that our conductors, all the metal that's on top of the die, are appropriately sized and won't be pushed around by current flows. And we do a Gate Oxide Integrity test to prove the strength of our gate.
So these are the tests we run for industrial and commercial customers. But what about automotive? Automotive folks like GaN too. So we run more tests. And what we run primarily for automotive is H3TRB. And this is a test that further increases the humidity and temperature over the standard HTRB. And then we run a series of tests that are designed to switch the part on and off multiple times to prove that it's immune to thermal shock from being switched on. And that's Power Temperature Cycling and Intermittent Operating Life testing. There's actually a military spec test. And those are the automotive tests we run, but we run them on a larger sample size. So all of these tests are run on a sample size of 77 devices.
So now we've taken care of the long-term wear-out mechanisms. Let's take a look at the short term, the manufacturing quality problems, the parts that fail in early life. And that's this area of the curve. So at PI, how do we gauge how many of these parts are we allowed to let through into our product stream? And the answer to that is zero. We're not allowed any of these parts. And so our curve actually looks like this, which would be a very uncomfortable bath. But how do we do that? How do we make an essentially perfect product stream? Well, the answer is prior to final test, we run a lot of additional in-process monitors. And what we're trying to do with those in-process monitors is we're improving the quality, we're reducing the number of poor parts that make it into final test. And that's critical for improving product quality.
So how do we find these parts and eliminate them before they even get to final test? Well, the way we do that is firstly our in-process monitor of our epi. Epi is absolutely critical when you build GaN. It's the most important part of generating a GaN transistor. And because we build our own epi, we don't buy it from a fab. No, we don't buy finished wafers either. We build everything. It means we have complete control over the process and we can check it at various points. And that means that only the best wafers make it through to patterning. After a wafer has been patterned into a transistor, we then run some stress tests. And the stress tests are there to find and either kill or ink out the parts that are unacceptable or have a risk of a quality problem before the parts even make it to final test.
So how do we validate that our in-process monitors and our yield improvements and our final test regime are going to yield a perfect product stream out of the other side? And the way we do that is with a test called ELFR, Early Life Failure Rate. And that's highly analogous to High Temperature Operating Life, but instead of doing a small number of parts for a long time, we do a large number of parts for a relatively short time. And in this case, it's 800 parts for 48 hours, multiple lots to ensure that we don't have any process slot dependencies.
So we've spoken about reliability and wear-out mechanisms and how we guarantee initial quality. But we've got a middle section here. These are parts that are perfectly good, they've been manufactured perfectly, they haven't worn out yet, but maybe they're dead anyway. What happened? Well, the answer is overstress.
So how do we assure that our devices are robust under temperature, voltage, and humidity, that they're going to find out there in the wild? Well, we run more tests. These tests are MSL, Moisture Sensitivity Level; uHAST, which is also a moisture test under pressure; TMCL, which is a temperature cycling test which tests for differential heating and cooling; and High Temperature Storage Life, which is the absolute high temperature test.
So one of the most important parts of a power semiconductor is its ability to withstand voltage. So let's talk about robustness under high voltage conditions. We specify two types of GaN: 750 volt BV GaN and 900 volt PV GaN. But our BVs are not specified in the same way as a silicon device where you'd be talking about a breakdown voltage. So even though we talk about BV, that's not how it's set. There's a property of GaN called Dynamic RDS(on). Dynamic RDS(on) turns out that when you apply high voltage to GaN, it actually increases the RDS(on) of the device. It recovers later, but under the initial voltage and for a short while afterwards, it is a higher RDS(on). And it turns out somewhat arbitrarily that we've chosen five percent as being the limit that we can accept an increase in RDS(on), and that is how we choose the BV of our GaN devices.
So if our BV is being set by Dynamic RDS(on), where does it break down? There's got to be a number. And it turns out that for both of these GaN families, the actual physical breakdown is somewhere around 1400 volts. So there's masses of margin with respect to the datasheet BV limit.
So with this battery of tests for wear-out mechanisms to ensure reliability, for initial quality, process monitors, voltage margin, and tests related to temperature and moisture sensitivity, whether you're making an adapter or an automobile, you can have confidence in Power Integrations. For more information and to find out how we can help you with your GaN design, go to power.com.