Facility level power conversion: SST

This 3000 kilogram single megawatt transformer can be immediately replaced by lightweight power semiconductors surrounded by gate drivers, sensors and microcontrollers. Hello and thanks for joining this episode of Podcast4Engineers. It's the podcast you just have to listen to if you're interested in what's going on in the semiconductor market.

I'm your host Peter Balint and today we continue with our We Power AI series. And today we look at facility level power conversion. And we are specifically talking about the SST or solid state transformer. And today we're joined by Valerio Zorrillo. He's the global application manager for Solid state Transformers and UPS.

Thank you for joining us today. Thank you, Peter, for having me. First question I have is let's look at the way things are done today. We've been using for a long time the low frequency transformer for power conversion. To me it seems like this could maybe be inefficient, definitely bulky, so what now is prompting this change to something like solid state?

That's a good point. So low frequency transformers have been used in decades by many industries, including data centers. They have been very reliable, available, and actually also pretty efficient in the way they connect the end loads to the grid infrastructure on the medium voltage side. However, the recent increase in the power demand driven by the AI workloads has put this technology to its limits, specifically when it comes to space utilization, weight, and flexibility.

Just to give you a reference, a one megawatt low frequency transformer working at 50 or 60 megahertz in terms of operating frequency typically includes a large amount of copper or aluminum and iron. And for a one megawatt system, typically we are considering about 3000kg, of weight for this system. And if you put this into perspective and you multiply the amount of transformers, you may need to power a one gigawatt data center, you may need 1000 of these large systems.

We see we find actually that solid state transformer is is a better way to solve these power connection challenge. And the reason is these 3000kg single megawatt transformer can be immediately replaced by lightweight power semiconductors surrounded by gate drivers, sensors and microcontrollers to make it as reliable as the low frequency transformers, as efficient as them, and more flexible when it comes to space utilization, weight, and scalability.

Okay, so you say they can't keep up with a common needs as of today. Is this recently since this whole wave of of AI usage, and has this been going on for a while. So let's say the the challenges on the traditional technologies like line frequency or low frequency transformers has been in place since a few years now, but it's really the rapid evolution of the AI data center power demands that is pushing to the limits of this application.

So we have been seeing actually, we are seeing today a large increase in the rack power. So this is really the main challenge that the traditional architecture is facing in order to be able to really connect in a more efficient way to the utility medium water utility side. Okay. Now when you think about this, a transformer in a solid state form does it...

are we talking 100% solid state or are there some magnetic components that still transfer over from the older technology? That is absolutely correct. Actually, solid state transformers still use some sort of magnetic component, but the beauty of it is that the transformer is not as bulky as large as the traditional low frequency transformers.

We're talking about high frequency transformers. Just to give you a reference. Still the same one megawatt low frequency to somewhere. That way 3000kg can be 40 times smaller and lighter with high frequency transformers that are typical technology that works in the range of 20 30kHz of frequency and beyond.

So what is needed to be able to use this lighter and more compact technology? In the end, we need power semiconductors that are able to work at high frequency. To have fast switching frequencies and also being able at the same time to work with the voltage levels that are required to interface with the medium voltage grid.

What Infineon provides in this case is actually the silicon carbide technology that is able to both work at high frequency, so providing high efficiency in the power conversion stage, and also having the right voltage classes to be able to interface with the medium voltage levels. And all of these, combined with the proven track record of high voltage silicon carbide technology that is already used in other applications, for example, solar and energy storage systems.

So if you have to summarize what what can SST or solid state transformers offer to this situation of AI centers popping up everywhere and having so much demand, what's the what are the the key points or the key takeaways from using a solid state transformer to ease the issue of too much power needed?

Yeah, to put things into perspective, Peter, I would like probably to to provide a brief overview about how the power has evolved in terms of data center requirements for the rack power specifically. In 2020, the typical rack power was in the range of 10 to 15kW for per rack, then today we are typically talking about 100kW, which is almost ten times or more than that.

And in the near future, there are projections that this rack power will actually reach several hundreds of kilowatts up to one single, up to one megawatt in a single rack. So this evolution of rack power levels is really challenging the traditional way the power is delivered to the Rec. So we, for example, one megawatt rack, there is too much space taken from the power supplies to be actually to deliver the power required for the one one megawatt computing rack.

So SST in this game can play the role of freeing space from the from the white space. So instead of consuming space in the in the rack for the power supplies with the SST we can remove that space, just leave all the space for the computing part, and move the power conversion level, which is still required in the data center up upstream.

So in this case, from the rack level to the facility level, and as close as possible to the medium voltage connection to the grid. Okay. So it sounds like that this is something that's not just nice to have it's almost required right this has to happen. Absolutely. Yes. I mean the traditional architecture is not sustainable for this pace of increase in power demand.

So SST will definitely play a role in supporting the evolution for these increasing rack power levels. Okay. And we've talked about AI data centers. And you've made the case for SSTs in such a scenario. But what else what other situations might benefit from SSTs? Yeah. Data centers and the AI data centers is not the main application or the only application for SST although it's probably the main one where this technology can really help solving the space utilization challenge.

There are, however, other applications as well. For example, I could name the energy storage systems so large utility scale energy storage systems. These systems need reliable, efficient power conversion at medium voltage level. And typically what they do is really transfer energy from into the battery systems, for example, in case of battery energy storage systems.

SSD technology can also be used with the energy storage systems to make sure that this transfer of energy from into the battery to the grid is done in the most efficient and compact way. This is one example. There is also another one which is about megawatt charging systems. So these applications still had a strong demand for medium voltage to low voltage conversion, done in a compact way and in a scalable way, to make sure that the connection to the grid is, is, let's say, sustainable in a way.

So again, SSD and the architecture behind SSD can really support megawatt charging systems of the future as well. Other applications that I found also relevant for SST are for example solar large solar inverters and also marine. So let's get back to data centers for a while. And how is the rollout going as far as SSTs and their use in AI data centers?

I would say the rollout is going pretty fast? So I was traveling throughout regions outside of Europe recently, and I was really fascinated to see the amount of development efforts that is being put into this technology. Demonstrations are up and now today, this year, 2026, and there are field pilots that are expecting 2027.

And all of these is in preparation for wider adoption starting from 2028. So definitely there is a plan and a landscape. And in the entire data center index is actually preparing for that. I also wanted to mention that although the technology related to the SST is already available and scalable today, there are other elements beyond the SST itself that are still needed to make sure that this technology can be scaled up and used widely.

For example, when talking about data centers that are using DC voltages or adopting DC microgrids, there are elements like DC voltage power protection that are required to make sure that the SST technology can be deployed in field. So looking at this landscape of AI data centers and the SST what is it that Infineon can offer or what is it you're working on that could aid in this situation?

So solid state transformers they are power systems that are doing the power conversion from voltage to low voltage. Typically these power systems are using modular architectures with the stacked power cells. Because of the medium voltage connection, it's very important for SST to leverage high voltage semiconductors.

And here Infineon provides their silicon carbide easy pack modules. These modules are really designed for SST and the reason is they have a very wide voltage range from 750V up to 3.3 kilovolts. And just to give you a reference, using a 3.3 kilovolt silicon carbide Easy pack in an SST system can reduce the overall system complexity by 65%.

If this is compared with a traditional 1200 volts silicon carbide device. So that's a lot of reduction in complexity, and high voltage is typically in SST need to come also with high reliability. Infineon is a pioneer in releasing to the market high voltage semiconductor technologies. For example, for silicon carbide, Infineon released the two kilovolts and the 2.3 kilovolts, specifically tailoring the solar and energy storage systems applications for voltage levels in the range of 1500 volt DC, so it's very important to combine the two together still in the direction of reducing complexity.

There are ongoing developments with silicon carbide discrete packages, which is another choice potentially available for SST architectures. And in some topologies and configurations. We also see gallium nitride technologies playing an important role thanks to the high efficiency and high switching frequencies.

I also like to say that there is no power without control. So Infineon solution is really covering everything, not just the power switches, whether they are module or discrete packages, but also on the control side. And I would like to mention, of course, the auxiliary power supplies, the gate drivers, the current sensors and the microcontrollers.

All of them play an important role to make sure the SST can run efficiently, reliably in the high demanding application for AI data centers. Okay, so Infineon offers a nice system that supports each other, and yeah, works for a common solution. Correct. Okay, good. And what if you were to look forward and maybe go down the road a little bit in this area of SSTs and AI data centers...sticking with that theme, what do you see?

I mean, let's say five years down the road do you have a vision of how things will be or maybe ten years even? That's a nice view. So what I personally see is that SST today is already helping transitioning to an architecture that can really reduce the space utilization or the space used by the large power converter systems interfaced with the medium voltage, and also increase the efficiency.

And this is specifically relevant for DC loads. So loads with DC voltages. For example, we can name AI data centers and megawatt charging systems. So this is already something happening today. In the future it could be five years from now for example. SST could also actually help do with the transition from medium voltage AC grids to medium voltage DC grids...and in this case, there is also a further reduction of the number of conversion stages.

So a further improvement of the overall efficiency. And also there is the possibility to integrate seamlessly sources of green energy, which could be for example, hydrogen in the future. Thanks so much for taking some of your time today to explain your part in rolling out SSTs in the AI data center world, where they are probably most needed.

Thank you, thank you Peter, and thank you to our audience for joining today. Make sure you subscribe to the podcast for engineers podcast so you get the latest updates and we'll see you soon.

Facility level power conversion: SST

Kapitelklicken springt direkt hin

Shownotes

Transkriptvom Podcast bereitgestellt · Zeitmarken springen hin