The video is set to go live on Thursday, January 16th 18.00 CET (17.00 UK, 11 am Eastern)

Interview Partner

In this interview, Joe is talking to Alexander Lange –  Co-Founder and Head of Software Engineering (https://www.linkedin.com/in/alexander-lange-527522119/) at Darmstadt based deep tech startup Xelera Technologies  (https://xelera.io/). He has been with the German Aerospace Center and the European Space Agency.

 

The Startup

Xelera Technologies offers specialized software for FPGA processing, which accelerates cloud computing considerably for certain tasks. Xelera already won two awards from the German federal ministry of economics and energy (including digital startup of the year 2019) https://www.de.digital/DIGITAL/Redaktion/DE/Gruenderwettbewerb/Artikel/preistreager-Digitales-Start-up-des-Jahres-2019.html. The startup has also been named a “Top 50 Startup” by German Blog Für-Gründer (https://www.fuer-gruender.de/beratung/gruenderwettbewerb/studie-2019/top-50/)

 

Recommended Readings (Affiliated Links)

The Fire Starter Sessions by Danielle LaPorte https://amzn.to/38Ug3rU
The 7 Habits of Highly Effective People

by Stephen R. Covey

https://amzn.to/2sGDy78

 

The 8th Habit: From Effectiveness to Greatness

by Stephen R. Covey

https://amzn.to/34DYbOC
The $100 Startup by Chris Guillebeau https://amzn.to/2PEoFet
Outliers by Malcolm Gladwell https://amzn.to/36U41g6
Grit – The Power of Passion and Perseverance by Angela Duckworth https://amzn.to/2SbuEcj

 

 

 

Deep Tech Podcast

If you are interested in German deeptech startups, you can find our audio-only podcast “Deeptech Germany by Startuprad.io” here:

iTunes: https://apple.co/2oWO3Sl

Spotify https://spoti.fi/2qucve8

 

Venture Capital Round

Xelera is currently raising its seed round in Q2 2020. If you are interested in investing, reach out to us, we will get you connected.

 

Further Reading

Formerly with DLR (German Aerospace Center) https://en.wikipedia.org/wiki/German_Aerospace_Center

ESA https://en.wikipedia.org/wiki/Rosetta_(spacecraft)

ESOC https://en.wikipedia.org/wiki/European_Space_Operations_Centre

FPGA https://en.wikipedia.org/wiki/Field-programmable_gate_array

GPU https://en.wikipedia.org/wiki/Graphics_processing_unit

FPGA coding https://www.fpga4student.com/2017/08/what-is-fpga-programming.html

Transcript

Intro: Welcome to Startuprad.io, your podcast and YouTube blog covering the German startup scene with news, interviews and live events.

Joe: Hello and welcome, everybody. This is Joe from Startuprad.io, your startup podcast and YouTube blog from Germany. I have today another deep tech interview for you, because shortly before Christmas, I was invited by Technical University of Darmstadt, one of the premier technical teaching institutes of higher education here in Germany. And at this fair, I met some very interesting startups, of which you will see a few interviews over the next month. And therefore, there is a start up here in the person of Alexander, who will do an interview right now with me talking about their deep tech startup. Alex, and a welcome.

Alex: Hi, Joe. Thanks for having me.

Joe: Completely my pleasure. You are a very technical guy, and let us first go through a little bit what you did before, so people get a kind of feel what type of guy you are, what your company does, then we get a little bit into the market and what you stood up does. I’ve been stalking you on LinkedIn, and everybody would like to reach out to you directly, go down here in the show notes. If you watching this on YouTube, there’ll be a link to our blog and there are all the links. If you are listening to this as an audio podcast, go down, there’s also a link to find our blog without the outgoing links, including everything we’re talking about now. So, you’ve been a technical guy and you’ve been with the German-based aerospace center, right?

Alex: Exactly, German Aerospace Center in the department, Bremen. I also studied aerospace engineering and yeah, I was working so for quite some time, as a student first, and then for some time as a real employee on a satellite project, of a satellite that’s actually flying around earth now.

Joe: Can we talk about what the center actually does and what was your role in it?

Alex: Well, the German Aerospace Center itself is pretty huge. I think it’s one of the biggest, if not the biggest Science Institute in Germany. It has departments all over the country and does lots of things, also not only space related, also planes and energy, transport, a lot of different scientific topics they’re working on. And in Bremen, we were working explicitly on satellites. And we are bringing scientific set of satellite missions into space. And the ones I was… the one I was working on was this rather small one which was responsible for trying out new methods to track ships in the ocean all over the planet.

Joe: Also a way to track like little pirate vessels?

Alex: That’s also one of the applications. So, basically, if a ship gets hijacked from pirates that you can track it and follow it and then later get it back, for example.

Joe: Hmm, that is pretty cool. Unfortunately, no little Death Star with laser, aww. What did you do after that?

Alex: Well, after that, I got accepted for a traineeship at the European Space Agency. That’s… they are a bit smaller in total, but they are managing basically the whole space sector all over Europe, giving out contracts to private space companies to build huge satellite projects, like satellite constellations like Galileo and a lot of big scientific missions. One of the most famous ones which was recently successful was the Rosetta mission, where humanity sent the first time a satellite to a comet and landed there and, yeah, all kinds of stuff like that. And there was also involved in a small scientific mission about some satellite which was used to try out many different things to, yeah, see how you can operate future satellites better. And to just launched before Christmas, so just recently and it’s still alive and doing well.

Joe: And surprisingly, you… but first, we should tell the people that the ESOC, The European Space Agency operation center is actually located in Darmstadt. Like, if you don’t have rush hour like 45 minutes away from Frankfurt, if you do have rush hour, it can take you 2 hours ‘til well longer. And this is how you ended up in Darmstadt?

Alex: Exactly. This is how I ended up here. Originally, I’m from Bremen, but that’s how I got here. And that’s also how I met the other people of my founding team, especially Felix, our CEO who initiated this whole project was also working at ESOC, European Space Agency at the time I was there. And so we got connected, and when he came up with the idea, I quickly joined him.

Joe: That is actually pretty cool. So, you’re a let’s say carve out up the European Space Agency. Let’s talk a little bit about how the satellites took you 1 level down to the clouds. What does Xelera Technologies do? And how do you actually improve the world?

Alex: Okay, that’s a little complicated. The thing is, nowadays…

Joe: Well, may I interrupt you here? We don’t do small things here. It’s not a radio.

Alex: It’s true, yeah. So, there’s a lot of talk nowadays that the cloud and the internet is consuming a lot of power, and therefore blowing carbon dioxide into the atmosphere. There are numbers from like all the Bitcoin mining on the planet consumes as much power as all of Ireland and stuff like that. And that’s because we have giant data centers all over the world, which are full of servers, which are just constantly running and doing stuff like streaming our videos, doing our Google searches and things like that. And there are a few alternative technologies to improve the situations, so called hardware accelerators. They are like processes in computers, but kind of different, they have different advantages and disadvantages. And one of those are the FPGAs, so called Field-programmable gate arrays, and they can do certain tasks (especially in the machine learning and AI sector and also video encoding, decoding) a lot faster and more energy efficient than normal servers could never do that. And that’s basically what we are doing. We are replace functionality which is running on conventional servers in the cloud or private data centers, using this FPGA technology to just make it faster and more energy efficient and also more cost effective.

Basically, it boils down to that you have some function, maybe some AI object detection or whatever, which we are running at the moment, or some traditional machine learning. And you can accelerate it by up to 100 times or and reduce the amount of servers you need by putting in these specific carts. And the problem with that is that these cards have to programmed in a different way than normal software. You cannot just run Windows or whatever on these hardware accelerators. You need specific programming languages, specific… specific knowledge to utilize these accelerators. And that’s where we have a lot of action expertise in our companies. Like 2 of our co-founders did their PhD on this topic. And we also got a couple of more employees who have really deep knowledge in that field. And that’s why we can provide the software. So, we are not building the hardware, we are only providing the software for this already existing hardware to make them usable to accelerate all these different tasks and make the world faster and more energy efficient.

Joe: I see I have a lot of questions in my mind. First thing, people who already know me know that I’m not a technical guy. Can I imagine this hardware accelerator like something like a graphics card?

Alex: That’s exactly what it is. So, graphics card is also a hardware accelerator. And they exist for decades already and everyone knows him because you need them in your PC mainly for playing games and stuff. But they are also graphics card inside data centers, thousands of them. And not just for video stuff, they are also really good at doing machine learning tasks and artificial intelligence. They are running all this kind of object detection, language processing, all this stuff, because they are faster than normal CPUs. And an FPGA board looks exactly the same as a graphics card. I can actually show you one. I have one here with me. So, see, just looks exactly the same. You could plug it inside your PC with this PCI interface. But the FPGA has just different architecture it works differently than CPU and GPU and has different advantages and disadvantages. Some machine learning algorithms run better FPGA, some on GPU. The thing with FPGA is that they just came recently into the data centers. They have been around for decades already, but more in the embedded market. And then what started this whole business for us as well was that 2014, Microsoft started using FPGAs for the first time in their data center, for example, to accelerate their search engine, Bing… Bing, and also their office services and Azure cloud. And shortly after that, Itel, like one of the biggest companies in the market, bought 1 of the 2 big FPGA producers, Altera, for more than 16 billion, so it was super huge amount of money invested there. And 1 year later, in 2016, Amazon Web Services, one of the biggest cloud providers, started to put these FPGA into the cloud. And that was kind of the starting signal for us that there is a new upcoming market for this new type of hardware accelerators to make them usable in the cloud and in data centers.

Joe: And that is where your company actually starts. Can we… let’s talk a little bit about how large this market is. You’re talking about like AI machine learning, cloud computing, that’s… that’s a big market, right? That’s a terribly big market.

Alex: I recently saw a McKinsey report, which was saying that the potential total annual value of AI and analytics across all industries would be around 10 to 15 trillion dollars. So, it’s incredibly huge. Of course, the… our share of the market is quite a lot smaller because we are only interested in applications that need a lot of acceleration. Like when you’re only scanning documents to archive them, then it doesn’t really matter for us because that can run overnight or whatever. We’re interested in applications that need like real-time. For example, if you have object detection in a robot to avoid obstacles or something like that, you want to have that happen in as little time as possible, because then your robot can drive faster because it can detect obstacles faster as well.

Joe: Does this also apply at autonomous cars?

Alex: Yes, of course, as well there. They are also using hardware accelerators already for all their AI applications. And that could potentially be a market as well, yeah.

Joe: And I also saw on your website like voice assistance is like the smart speakers.

Alex: Yeah. So, la… speak… speech processing is 1 of our biggest pillars. So, basically, we have kind of 2 application areas where we are really good in, and 1 of those is speech processing. That means that we can like read… so, we interpret what you are saying, put it into text. We also recognize who is talking, that’s also important for many applications to identify as a user also, when you have like a conference or something like that and you want to transcribe who is saying what and things like that. And we are focusing there on applications where you really want to have real-time speed. For example, sometimes you have these smart assistants, but it takes Alexa to answer, like 5 seconds or something like that. But you want to have an answer in like 60 seconds because you want to have some kind of fluent discussion, like you would have with a human. Then you would need some really good technology. And I think we are kind of the only ones who can provide this at the current time.

Joe: Personal experience with Alexa, it’s very helpful. But the only area where it has like a broad understanding of everything that’s going on is when you insulting it. So, all the conversations are very much scripted and you have to stick to the script, but she recognizes a lot of insults. I have… I have a friend who visits me regularly, and he just pokes fun of her all the time, and she recognizes like 90 to 99% of the insults. Not too sure what the coders are thinking there. So, that is like the application area. And what… now we get a little bit more into details. First, we should say I’ve been stalking your website extensively, and I realized you actually working with your brother before we get into the startup itself. How is that?

Alex: It’s kind of interesting. So, he’s a bit older than me, so he’s my older brother, but he’s kind of the best programmer I’ve ever seen. And after we started with the startup, and we were quite successful, I just asked him if he wants to join and he said yes, because I needed a… well, as a startup, you need really, really good people. You cannot afford to have someone slacking around because you’re limited on manpower anyway. And he was the best of the best, so I did this together with him.

Joe: That is undisputedly the most objective statement we’ve ever had here on startuprad.io, “My best… my bigger brother is the best coder in the world,” totally, you nailed it, awesome. Now… now, let’s talk about what those 2 brothers and all the other people at Xelera Technologies are actually doing, because you trying to explain it very dumb down, but let’s try to get to what you guys are actually doing, what your magic is, right?

Alex: Okay, so that’s quite interesting. So, as I mentioned before, we are using this FPGA technology, but it only reached the data centers quite recently. So, there are not a whole bunch of people out there who can use it. I mean, you have millions of programmers who can program CPUs, but only a couple of thousands who… on the planet who can do FPGAs in a… yeah, efficient way. And we have a whole bunch of this expertise in our company who are just coding these machine learning and AI algorithms on the specific hardware or for the specific hardware. So, they’re really doing low level stuff optimization on a really hardware near layer to get like the maximum speed out of these hardware accelerators, the best efficiency we can get. But that alone is not sufficient for the customers because, for them, it’s really hard to use these low level software for these platforms. So, we also, the other half of the team (and that’s where my brother and me are actually working in) is doing high level software to connect these low level FPGA parts to the rest of the world, basically. So, we provide the interfaces or integrations into standard frameworks. For example, they are data sign frameworks like Apache Spark that can use our FPGAs to accelerate machine learning. And the user doesn’t need to know anything about this complicated hardware below through our integrations. He can just keep writing his same data science scripts just like before. He just needs to install our software and of course, 1 of the hardware ports, and it suddenly runs 100 times faster than before.

Joe: How much hardware are we actually talking? Because you said 1 hardware board but actually, how much would you need like to seriously upgrade a data center?

Alex: Yeah, that’s a good question. Kind of depends on the applications, because for some applications, they’re far better, and for some, just a little bit better. But in general, we have seen applications like where your recommendation engines are stuff like that, which are done on clusters of 100 servers. And if you just outfit each server with 1 of these cards, you suddenly only need 5 of them instead of 100, because with these 5 cards, 5 servers can do the same in the same amount of time than 100 servers could do without the cards. So, they are data centers with thousands of these cards, but depending on the application, you usually reduce the amount of service drastically you need.

Joe: Hmm, I see, see, see, see. And basically, how do like companies or actually startups use your tools? So, basically, you are a startup, you do machine learning, you want to seriously accelerate that stuff and then you buy one of those or many of those accelerated cards. And then how do you guys, actually… how did those guys actually use your software, your tools?

Alex: Okay. Well, it depends a little bit on the platform where they want to go. For example, in the cloud, most big clouds like Microsoft Azure and Amazon Web Services, they already have these FPGA instances. So, you can just go to AWS, launch an instance, and it has an FPGA there. But you on your own won’t be able to do anything with it. So, you also download our software and install it on that instance and, yeah, then you can basically use it out of the box. We offer, as I said, several integrations, which enabled you to use standard frameworks to utilize our software. But of course, we are also doing a lot of projects with customers to integrate our things directly into their proprietary software. So, then they would just buy a bunch of these boards for their own data center, and we would implement the interface to connect basically our IP for the boards to have their own software.

Joe: Hmm, I see, see, see, see. And everybody who would… who’s now interested in the software, for you guys, you can go down here in the show notes and have a look at… or have a look at www.startuprad.io/blog. There, you can find the interview and all the show notes of it. But your software is just not any software. You guys been recognized with some awards already, right?

Alex: Exactly. We won quite some competitions just last year as the we were got the main prize for the digital startup of the year 2019. And the year before, we also want the founder’s competition of digital innovation 2018. We got both prizes in Berlin, and yeah, it’s nice to get recognition out there.

Joe: And we should say it’s… those been awarded by the Federal Ministry of Economics, and let me check… Economics and Energy, right? And you also got 1 of the top 50 startups for the widely read blog for (unclear) [23:39] for founders, right?

Alex: Mm-hmm, exactly, yeah.

Joe: Sounds already pretty good. Then there are the usual startup radio question, how you guys are financed right now?

Alex: Okay. Right now, we got the so called exist funding. It’s meant for bringing research from universities to… yeah, to the business world. So, basically, we work together with together with TU Darmstadt, the Technical University here, and did a startup founded out of there. And we got funding from the government, yeah, for the time period of 1 and a half years where they supported… supported us. The… it will run out in March, so we are actually in the process of looking for investors now to push us a little bit more. We kind of can sustain us right now. We have lots of projects going on, and we also grew quite a bit. We are, at the moment, 8 fixed employees and a couple of students working for us. But it’s… we kind of need some investment to grow a bit faster because this window for this opportunity just opened because of all this FPGA hype that’s currently happening. So, we kind of want to go as fast as we can before other people are overtaking us. Because all the competition we have at the moment, they are all startups, there are no big companies trying to do the same as we do. But a lot of them are in the Silicon Valley, and they usually get quite a lot more funding than you can get over here. So, yeah, that’s what we’re trying to get right now as well.

Joe: You guys are looking for something like a Series A?

Alex: Yeah. Well, at the moment, it’s kind of more like seed funding, but yeah.

Joe: Everybody who’d like to reach out to you guys, we surely connect you, just reach out to us. You see, my… well, sorry. We’re working with 2 cameras here and there’s my microphone in the way. But you see here my email address, reach out and we connect you guys. Only thing left for me to say is thank you very much, greatly appreciated you taking your time, and very interesting startup. Wish you all the best.

Alex: Thanks a lot, yeah. And it was really nice to do this interview with you, and thanks a lot for the invitation.

Joe: Completely my pleasure. Thank you very much. Bye-bye.

Alex: Bye.

Outro: That’s all, folks. Find more news, streams, events and interviews at www.startuprad.io. Remember, sharing is caring.