Infineon key visual in RGB

Safeguarding AI operations against downtime: battery backup unit solutions

We’re bringing you the latest from the world of semiconductors – straight to your ears! From quick takes on trending applications to deep dives on product innovations, our experts give you their take on the tech behind the tech.

Podcast

Feb 28, 2025

Podcast

Today, we will examine the critical role played by battery backup units in keeping AI servers running smoothly, even in the event of power outages. Our guest, Filippo Rosetti, Application Marketing Manager at Infineon, shares his expertise on effective battery backup hardware strategies that ensure uninterrupted power supply to AI systems. From process continuity to seamless power switchover, we'll explore the importance of reliable battery systems in powering AI.

Listen on Spotify Watch on Youtube

Speakers

In this episode of Podcast4Engineers, host Peter Balint speaks with Filippo Rosetti, Application Marketing Manager at Infineon

Peter Balint

Host:

Peter Balint has shaped visual and audio narratives at Infineon since 2021. He’s a video producer with 20 years of experience and has produced podcasts for the past 10 years. Over his career, Peter has interviewed speakers from all over Europe, bringing high-quality media production and engaging conversations to the forefront of his work.

Filippo Rosetti

Guest:

Filippo Rosetti working as Senior Manager Application Marketing at Infineon with a strong technical background: he holds a Master of Science in Electronics Engineering at university of Padova specializing in Power Electronics. He began his career in 2018 as an Application Engineer supporting high side power switches in the automotive market. After 3 years, he decides to change role being Application Marketing Manager for Battery Management Systems in the industrial market, promoting Infineon full system solutions. 

In 2024 his end application focus shifted to AI Datacenter and in 2025 became team leader of a team of engineers developing system solutions in the AI space, shaping the market with new products and innovative ideas.

Additional information

More episodes

Guest: It's really important to have a system that backs you up when you make such long computational efforts. If we talk about ADAS and autonomous driving, you can go into accidents. If we talk about financial platforms, we can talk about loss of money. It's a big deal.

 

Host: Hello and welcome. This is the 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 our journey into the world of powering AI. Today we talk about battery backup units and joining us is Filippo Rossetti. He's an application marketing manager here at Infineon, and I want to start with the question which we talked about a little bit offline. What's the big deal about battery backup, especially in an AI data center? Redundancies have been around forever, and they seem to do well, but I'm hearing that with AI, it's a little bit more critical. Can you fill us in there?

 

Guest: Yes. Thank you very much for the opportunity to talk here. Let's first start with the definition or what is a battery backup unit? Battery backup unit consists of a device, which is usually a battery, as the title is saying, that is providing power during a period of power outage or an AC power loss that can happen in our systems. You have an example here behind our mockup. This is a battery. You can see the lithium-ion cells. and some electronic systems here, which I will go more in detail later on. You said really correctly, redundancy is here for a long time, but why for AI become even more critical to have these systems? Because if you think about the complexity of these algorithms in AI that can take from hours even to days to be processed, if you have a power loss during this high demanding algorithm, you can lose everything and you have to start from scratch. It's really important to have a system that backs you up when you do such long computational efforts. If you also think about real-time processing applications like autonomous driving or like, for instance, financial platforms, they need to process data in real time without latency. What does it mean? It means that if you have a power outage in this little amount of time, you can cause severe damage. For example, if we talk about ADAS and autonomous driving, you can get into accidents. If we talk about financial platforms, we can talk about loss of money. It's a big deal.

 

Host: So literally every second counts.

 

Guest: Yes, every second counts. And let me add this, our server racks are increasing in power and therefore this massive amount of voltage, current, and temperature is hiring the risk of having power outages. We need to address this with specific solutions.

 

Host: Okay. And Infineon as a semiconductor manufacturer, what opportunities lie ahead?

 

Guest: Okay, so let's look at the system here. Here it's a battery. A battery is usually powered by lithium-ion cells, which require battery management systems, which is a PCB that is monitoring your cells of your battery, that is balancing the current among each cell, and it's also protecting your system with dedicated solutions. This is the first socket for Infineon. that we can address. The second big opportunity is the DC-DC converters which are in it. Why? Because as you know, the lithium-ion battery can be charged or discharged, so the voltage is fluctuating. It's not the same. We need this application to have a constant output voltage to address the requirement of the bus voltage. So that's why we need DC-DC converters that keep the bus voltage or the output voltage of the converter stable at the same level as the bus voltage.

 

Host: Okay. And let's take a moment to look at from the customer side. I know Infineon is very active in engaging with customers to find out what works, what doesn't work. But based on this experience, based on talking to the customers, what are some of the pain points and what are some of the challenges being reported back to Infineon?

 

Guest: Okay. The pain points are a lot, but I'll try to summarize the key ones. I would say the first one can be named as the reliability or uptime. We need systems that are reliable in providing power for a certain amount of time. And this is mainly done by the BMS, which can even incorporate functions like prediction of failures of state of charge, state of health, or even the lifetime of your battery to really address safely and reliably the power in that specific amount of time. The second, and in my opinion, the most important one is efficiency and power density. Why? Because we don't need to lose power in this power flow. We need really a high efficiency when we provide power. And we also need to provide it in really a small space. Because our server is full of systems now, the more we go up with the power and the more components we find in the server racks. The dimensions for these solutions are really small, but the power is going up. We need to balance the space and energy density. There are also some others like, for example, scalability and flexibility. Here, you need to have solutions that adapt to the battery voltage or battery change. Why? Because we can address different voltages, different chemistries. We need systems that are scalable, that can, and that the designers do not have to redesign every time if the requirements change. Then, of course, thermal management. If we talk about high voltage and high currents, thermals become a big issue. And we need to have specific, let's say, ways how to address the thermals and also security and safety. We need our batteries to be secure and safe. We know that lithium-ion batteries sometimes may lead into some, let's say, undesired conditions, but we need to have them operating at a safe level.

 

Host: Okay, so this is a whole list of topics to be addressed. I mean, how exactly do we address these things?

 

Guest: Well, I will start with energy density and efficiency. So here we have, or all the semiconductor companies have solutions in most efficient technologies such as gallium nitride and silicon carbide. And these devices can really express the maximum in terms of power conversion and switching losses reduction. These are the solutions that we envision for such kind of converters to address specifically power density and efficiency. If we think about the scalability, also we have ways how to implement these converters and the BMS in a scalable way so that they can fit different battery voltage, battery currents, and also the reliability topic stand in our software. So how we manage these batteries is really important.

 

Host: Okay. And what is done actually in the way of reliability? What kind of developments are there on the forefront that are, we're able to say that it's increasing reliability by X percent?

 

Guest: A key component for reliability is definitely, we call it disconnect switch. These are MOSFETs that disconnect your battery in case of failure conditions. And they can be of any technology, silicon, gallium nitride, or silicon carbide. And they help you to disconnect your battery, so your power, in case of a fault condition to avoid undesired events.

 

Host: So, you mentioned thermal management and let's look at that for a moment. Obviously, this is a big issue in a data center and especially in an AI data center. So, tell me what you're doing to address these thermal issues.

 

Guest: So thermal management is definitely a key topic in data center AI. Why? Because we know that the power levels are increasing, therefore also the current is increasing among our devices in the DC-DC converters. What we do is to address first the scalability and flexibility concept. We can design these converters in a scalable way with cards that address a specific amount of power, and therefore each card needs to dissipate the power in the most efficient way. These DC-DC cards can have a heatsink also that helps our semiconductor solutions to dissipate power and to offer more reliable thermal flow in our battery units.

 

Host: Okay. And you say that the size actually is— the size is an issue. You have no flexibility with making anything bigger, right? You have to stay within parameters, and this is the challenge, right?

 

Guest: Yeah, this is the challenge because we have standards that are telling us that we need to address a specific space. And as the power is increasing, the number of cells inside the battery module is increasing, and therefore the space for the electronics is decreasing. We need to cope with this challenge that for 48 V systems, it's really there and it's important to address.

 

Host: Okay. This is the way it looks today, but let's take a look maybe, I don't know, 10 years down the road. What do we see in terms of specifically battery backup units within a large AI data center?

 

Guest: As we talk about batteries, you know that batteries are continuing evolving across our history. I remember when we started with the lead acid batteries, which are very heavy and not easy to use. Now we have lithium-ion with BMSs, with PCBs that are addressing the monitoring functions. We can even predict the failure of such batteries. They offer reliable systems, how to, of course, provide power in such a short amount of time. In the future, in the battery world, we talk about different technologies that can come. We talk about sodium-ion, we talk about solid-state batteries, we even talk about lithium-air batteries, which can potentially position— be positioned in this application and take the place of the lithium-ion, which we see nowadays in every application.

 

Host: And do you think that's within 10 years?

 

Guest: I think so. Don't underestimate the speed of development in battery technology, even if it's— we think it's quite slow, but the research and development and also the amount of resources behind such technologies is really considerable. And I would really be quite sure that in 10 years we will achieve such advances. And another thing is that these batteries need to have— need to be cyber secured and also will also have a kind of AI algorithms inside. Why cyber secured? Because there will be new standards coming and there are also new standards coming. which are telling us that each module needs to safely communicate with the external world and not be corrupted by the users. And AI is very important because with AI or with predictive maintenance, you can predict the failures of your system. You know when the battery module has to be replaced with such AI algorithms, but which also means you know exactly when to do the maintenance of your overall system. You can reduce the total cost of ownership, which is, by the way, another pain point of our customers. Another trend which we see today in the market is definitely this high voltage architecture. Our customers are talking a lot about that. And it's possible that the BBU will move out from the rack with the power supply units and will move onto a rack called Sidecar, giving space to other components in our racks to, let's say, manage the higher demands of current and voltage. As the voltage of the bus is also going up, the battery voltage will also go up. Because if you keep a 48 V module, your losses will be simply too high to address a 400 or 800 V application. What you need to do is to have a higher voltage battery with its associated BMS and DC-DC converter. And you have to ensure that the new high voltage BMS and DC-DC converters are interfacing well with the new bus requirements.

 

Host: Okay. And this sidecar you mentioned, does this also have specifications in regard to size or do you have maybe a little bit more freedom in that regard?

 

Guest: Specifications are not yet, let's say, written. Our customers and the industry are still evaluating, which is the best technology, which is the best cells configuration, the best dimensions of these modules. But most likely, as you can see here, these modules will increase to address higher voltages because there is simply no space for so many cells in such limited space.

 

Host: It's clear that battery backup has an important part of powering AI, and it’s good to know that Infineon's a very important part of this. And I want to say thank you today, Filippo, for coming in and sharing your knowledge with us.

 

Guest: Thank you.

 

Host: And thank you to our listeners for being here today. If you have any suggestions or if you have ideas for future topics, please feel free to send an email to wepowerai@infineon.com. Thanks, and see you next time.