Best of LinkedIn: Software-Defined & AI-Defined Vehicles CW 35/36

00:00:00: Provided by Thomas Allgaier and Franis, based on the most relevant LinkedIn posts about software-defined AI defined vehicles in calendar weeks thirty five and thirty six.

00:00:10: Franus supports tier one automotive suppliers with early stage market validation for their R&D efforts by combining secondary research direct OEM expert interviews and facilitated customer meetings.

00:00:22: you can find more info in description

00:00:24: Right, so in this deep dive we are really setting expectations right up front.

00:00:27: We're going to unpack the most critical trends around software-defined vehicles or SDVs and you know AI defined vehicles that are circulating right now.

00:00:35: Yeah definitely because the mission of this deepdive is to explore how the industry has just completely overhauling vehicle architecture where bridging that massive gap between functional safety and cybersecurity leaping toward physical AI And honestly, grappling with the harsh business realities of selling these digital features.

00:00:53: It's a lot to cover.

00:00:54: so let's start with the fundamental digital plumbing.

00:00:57: The plumbing yeah

00:00:58: because before a vehicle can become this rolling AI brain?

00:01:01: The architecture has to be completely rebuilt from the ground up.

00:01:05: we're moving away from having dozens of scattered ECU.

00:01:08: those

00:01:08: are electronic control units right

00:01:10: exactly little isolated chips that used to control just the windows or just breaks.

00:01:15: We Are Moving From That Towards Centralized Computing.

00:01:18: But Joachim Langenwalter points out a pretty staggering bottleneck here.

00:01:23: The industry is currently facing A one point seven five billion euro middleware problem or one

00:01:29: point seventy-five billion just for Middleware.

00:01:31: Yeah, I mean the Middleware's essentially Just the translation layer between the operating system and the apps.

00:01:36: yeah How do you rack up a billion dollar bill just building the translation layer?

00:01:39: Well,

00:01:39: it shires driven almost entirely by duplicated siloed effort.

00:01:43: Currently there are seven major initiatives across the industry and they're all chasing the exact same.

00:01:49: two to five central compute nodes Oh wow!

00:01:51: Yeah And building one complete platform One that actually meets strict safety compliance over ten year lifespan costs roughly two hundred fifty million euros.

00:02:01: Okay so multiply that by seven independent initiatives.

00:02:04: Exactly, manufacturers are burning hundreds of millions to build the exact same invisible foundation instead investing in features a customer actually sees.

00:02:13: Let me push back on that inefficiency bit.

00:02:16: is this just like every smartphone maker trying to invent their own version of Bluetooth?

00:02:20: Instead of competing on apps?

00:02:22: because now here Patel advises using a lift and shift approach framework such as LINC to just reuse existing apps simply.

00:02:31: Right, porting it over?

00:02:33: Yeah like rather than overcomplicating the transition if you have a perfectly good breaking app why spend a quarter of billion euros rewriting

00:02:40: that?

00:02:40: I mean the lift and shift approach sounds incredibly practical but The engineering reality on the silicon itself is much harsher.

00:02:48: Legacy ECU code was written for dedicated microcontrollers so it expects guaranteed Dedicated processing time.

00:02:55: So when you lift and shift that code onto a shared system on a chip, it suddenly has to compete with a dozen other apps.

00:03:01: You can containerize the codes sure but you often introduce latency That simply wasn't there on the old hardware

00:03:06: Which means eventually developers have to figure out how this centralized computing actually functions on the dashboard And Daniel George recently highlighted This by if this is kind of amazing running the video game doom on the Android SDV stack.

00:03:22: That is wild, I mean it sounds like a tech flex right?

00:03:25: It

00:03:25: totally does but actually shows how display controllers manage hardware isolation.

00:03:31: in modern cockpit infotainment system might take you know twenty seconds to do cold boot But regulations say safety-critical stuff, like the reverse camera has to be visible within two seconds.

00:03:42: Right you can't sit there waiting for an Android logo to load before you back out of your driveway?

00:03:46: Exactly so.

00:03:47: to fix this George's experiment shows how the chip utilizes two completely isolated virtual machines.

00:03:53: They blend the frames from the safety system and the infotainment System independently right at the hardware level without sharing a compositor.

00:04:01: Oh So if the android side crashes The Safety Side just bypasses that bottleneck entirely.

00:04:06: Yeah,

00:04:06: it just keeps rendering this binometer directly to the display.

00:04:09: But getting that hardware magic requires serious compute power and strategies are diverging wildly on this.

00:04:15: Mate Al's Bizarra shared some Omnia survey data showing in Japan, seventy-nine percent of premium models mandate over seventy percent compute headroom at production.

00:04:25: Seventy percent?

00:04:26: So they're leaving processor mostly empty for future updates...

00:04:29: Exactly!

00:04:30: For their whole life cycle.

00:04:31: But then fifty four percent entry level models cap their head room Under thirty percent.

00:04:36: Wow!

00:04:37: Yeah, so it's a huge divide.

00:04:39: but honestly cramming all that compute power and extra headroom onto fewer chips creates a terrifying new vulnerability.

00:04:45: What happens when the entertainment software crashes?

00:04:47: the critical safety systems?

00:04:49: Right which brings us to theme two.

00:04:51: Safety in security just cannot exist in silos anymore.

00:04:54: J-Want D. Mahajan introduces this concept of freedom from interference.

00:04:58: Freedom From Interference?

00:04:59: Yeah, so you have an ASILB rated instrument cluster that has a strict safety certification sharing the exact same silicon as uncertified QM rated Android infotainment system.

00:05:10: Oh man!

00:05:11: So if android hangs... Hardware isolation is literally the only thing preventing it from blanking out a safety warning or delaying the reversing camera.

00:05:19: Exactly, The engineering discipline shifts entirely to building an impenetrable wall inside that chip.

00:05:24: But okay let me ask you this Even with hardware isolation What if the software update itself?

00:05:31: is hijacked.

00:05:32: That's the exact question Venetius Tadiozine brings up, he points out that safety without security is simply not safe and attacker doesn't respect your engineering boundaries.

00:05:42: so if the software update chain is compromised all of yesterdays' safety assumptions are just completely void!

00:06:00: instantly trace direct and transitive dependencies all the way from the silicon vendor straight to the vehicle.

00:06:06: Yeah,

00:06:06: it's like a genetic trace for code And because of this standards are finally evolving to fix the silo problem.

00:06:12: Jamie Williams talked about the ASQMS initiative.

00:06:15: Oh right!

00:06:15: The one out in China?

00:06:16: Yeah

00:06:17: It ensures that a cybersecurity threat analysis, ATARA directly triggers a functional safety hazard analysis or a HERARA.

00:06:25: So they don't just log into security vacuum anymore.

00:06:27: Exactly...it forces teams to talk.

00:06:30: A compromised Bluetooth stack instantly becomes a functional safety review.

00:06:35: That is a huge organizational shift, and hey real quick before we move into how this secure foundation allows the vehicle to actually start thinking I want to casually remind you to subscribe.

00:06:45: You definitely don't wanna miss our future deep dives in it's fast-movie mobility space

00:06:50: For sure because once that centralized Secure Foundation has built The vehicle stops just running software begins to think We're jumping from SDV to ADV?

00:06:58: The AI-defined vehicle.

00:07:00: Exactly, Amit Guhti and Prashant Jadhav outlined this trajectory at the Santa Clara SDV Summit.

00:07:06: It's an evolution of software defined to AI-Defined And ultimately physical AI.

00:07:12: And Alexei Nebelokhin noted that... ...AI computers rapidly shifting from data center right into the vehicle edge.

00:07:18: I mean we are basically turning cars in mass of AI computers on wheels which is sparking a whole new semiconductor race.

00:07:24: it

00:07:24: is.

00:07:24: but i have ask Are we just slapping the term AI on basic lane assist and calling it a day?

00:07:30: See, I have what i thought too but Harisha Karatekaramashapa had great insight.

00:07:35: It's hard no.

00:07:36: ADS and Autonomous Vehicles are actually first consumer scale implementations of physical AI.

00:07:43: Physical AI meaning operating in real world.

00:07:45: Yeah exactly systems continuously operating open dynamic environments.

00:07:49: this tech is directly converging with humanoid robotics.

00:07:52: now

00:07:53: Okay, but to make physical AI actually work without hallucinating the data architecture has to change right?

00:07:58: Oh completely Dennis Newman posted about always sets unified world model which addresses exactly this.

00:08:04: instead of fragmented data where like The camera perception owns an object list and the radar planning owns a different hypothesis You have a single versioned reality.

00:08:13: So everyone is reading from the same script.

00:08:16: Exactly.

00:08:16: If a sensor suddenly reclassifies the truck as a bus, that change propagates instantly across the entire safety and planning system?

00:08:24: No delays!

00:08:25: That is brilliant on technical level but you know technology's advancing brilliantly in theory.

00:08:29: The real question of how companies are organizing to build it And how customers react to actually paying for

00:08:35: it The business realities yeah It's rough out there.

00:08:39: Take Chinese market for example.

00:08:41: Rene Schindler pointed out that AI-enabled autonomy is actively outweighing traditional brand heritage and prestige for premium car buyers right now.

00:08:50: The software literally IS the luxury, no?

00:08:52: It really is...

00:08:53: And the sheer velocity of this shift is just staggering!

00:08:56: Mona Elfickey shared this insane statistic.

00:08:59: Five hundred and forty-two new vehicle models were launched in China In just five months.

00:09:03: Wait, five months?

00:09:04: How is that even possible?

00:09:05: I know right.

00:09:06: it raises serious questions about how safety engineering and validation can possibly scale at That speed.

00:09:11: you can't fast track physical crash testing.

00:09:15: You really can't And you could see legacy automakers structurally reacting to this pressure.

00:09:21: Karthik Chandrasekaran highlighted how Volvo actually closed its Stockholm site.

00:09:25: Just a move for fifty software and AI roles directly to Gothenburg.

00:09:30: Consolidating everyone!

00:09:31: Exactly, proving that the new competitive advantage is having tightly centralized highly integrated engineering teams rather than these massive distributed hubs all over the world.

00:09:42: But here's harsh reality of monetization right?

00:09:45: Think about buying a smartphone but cameras disabled until you pay monthly fee.

00:09:51: Ihan, AK and Michael Welsh have been warning everyone about this.

00:09:54: whole functions on demand trend.

00:09:56: Oh it's the

00:09:56: subscription models!

00:09:57: Yeah

00:09:58: while software as a product is obviously growing locking pre-installed hardware behind a subscription creates a massive user experience problem.

00:10:06: It

00:10:06: really does like.

00:10:07: you've already paid for the seed heaters to be physically put in the car And now they want ten bucks a month to turn them On

00:10:11: exactly if feels way less Like personalization more like a permanent advertisement sitting inside of car You already own.

00:10:18: yeah people hate that.

00:10:19: It's such a tricky balance.

00:10:21: And, you know looking at all this from the subscriptions to chips and safety I want leave with final provocative thought that we haven't touched on yet.

00:10:30: Oh about scale in sensors.

00:10:33: The rise of software defined connected vehicles from strategic competitors isn't just an automotive issue or even basic data privacy issue.

00:10:43: unlike compromise phone Millions of connected vehicles act as massive moving sensors, continuously mapping physical infrastructure.

00:10:51: Ports, airports power grids

00:10:54: all of it!

00:10:54: Exactly in high definition.

00:10:57: The ultimate question of the STV era might not just be about software subscriptions at all but a challenge on an unprecedented national security scale.

00:11:05: That is definitely something to think about.

00:11:07: Well if you enjoyed this episode new episodes drop every two weeks.

00:11:11: Also, check out our other editions on charging and battery tech commercial fleets.

00:11:14: And autonomous mobility markets.

00:11:16: absolutely thank you so much for hanging out with us On this deep dive.

00:11:19: don't forget to subscribe.