Well, hello. This is Nitan Dahad with E Times. I'm here at the NXP Semiconductors booth at CES 2026 and with Lars Reger. Good afternoon, my favorite interviewee. So Lars, you made an announcement here at the show around the Edge IQ, EIQ stuff. Let's talk a little bit about that and then I want to talk more broadly about the whole edge AI movement as it were now.
Yeah, perfect. The big problem statement that we are all seeing is how do you work with your models or with tool chains, with things that you have developed, and how do you bring them to edge devices? So how do you inference models and then compress them in a way that they are not glossy but land on the favorite edge electronics? For that, you need to have tool chains, and ideally, you are not only dealing with the AI model but also having tool chains where you can safely constrain what the edge device should do. So this is not only AI, it is building the bridge from agentic AI into building agents, and that is a very different thing because you need to add safety and security around it as well.
Now you said agents. Now we see agents as in co-pilot and all that kind of stuff, but...
What I mean is robots. I mean hardware agents. So intelligent systems at the edge that take over responsibility from the Neatins and glasses of this world, and you trust them and let them do the shopping, your fridge, your thermostat, heat the house, or...
Do the rolling robot. Yeah.
So one of the things you've been saying for a few years is context-aware device engagement and AI, and that's what you've got here. So we've been talking about it for years, but what's different now and why is everybody suddenly talking about edge AI, and how easy or difficult is it?
Yeah, very honestly, the big driver is efficient AI. Sending everything to the cloud is definitely not getting you to the humanoid robot. But as a human, the brain consumes 20 watts of energy. Normally, if you send everything to the cloud, you're in the thousands of watts for smart questions. Now, my fridge doesn't need to paint a Picasso or compose a song, but it needs to understand how many bottles of milk I have to buy for the next weekend. That's a very limited task and not always needed to have a complete AI system running. So these power stages being efficient, being specific on the task that the device is made for, that is the big magic now. And this is why everyone is so upbeat on moving to the edge, because your fridge has a certain mission profile, your car has one, your window blinds, your thermostat, but all on the energy efficiency side and less on the ultimate compute performance side. Now, how do you build tool chains that can suddenly cater for that? And how do I help the intelligent system maker of the future to make the microwave oven, the toaster, and all that stuff intelligent enough by driving maximum efficiency?
And okay, so that also goes to the big topic here, physical AI and robotics, yes, and that plays into this very well. And you've been talking about barrier-free living, a barrier-free life, and part of that is the robotics, part of that is enabling things to happen in terms of being context aware.
So devices that take over responsibility, or where you as an end user are willing to hand responsibility over to devices. Whether this is a tiny little hardware form factor or whether this is a bot on the internet for travel booking or whatever, I don't care. The only thing is I have to be able to trust these devices because otherwise I'm not willing to hand over my responsibility to a device. Now if the devices take the responsibility and I can trust them, then my life gets barrier-free because I need no brain power to say where are my keys, where is my wallet, and so on. This is all seamlessly done.
And it's the world that anticipates and automates, like our marketing folks would say.
Yeah, and what have you been doing over the last year or two with the acquisition of Kinara and everything else, all contributing to that I guess?
So TTTech was exactly bringing this safe and secure middleware for these type of devices because you need to be safe and secure. This is the nerd language for trust your device. And the other part is Kinara is of course on the efficient AI level. So we are combining a Kinara AR2 which can carry 9 billion parameters and an i.MX microprocessor, and the combination is below 8 watts and is writing you already a complete doctoral report for your X-ray machine. So these type of things, only of course when the X-ray machine is active and running. So therefore we are going back into the power concept of the devices that we have around us at the moment, and we do not need orders of magnitude more of energy, and that's the magic that is the stuff the barrier-free world that I'm driving.
