Software-Defined Vehicles in Automotive

BREAKING: Volvo is replacing the entire “brain” of the EX90. 🤯 Turns out SDVs are way harder than they thought. This is not a faulty sensor recall. It’s a complimentary central-computer replacement for every 2025 EX90, the core compute platform of the car, less than a year after launch. The decision strongly suggests the original hardware didn’t leave enough headroom for the software and features Volvo intended to deliver. Across forums, groups, and early reviews, the pattern was consistent, not one weird issue but a stack of foundational instability: - digital key / key systems failing unpredictably - infotainment freezing, black screens, spontaneous reboots - unstable connectivity - recurring error messages - LiDAR-linked safety features missing at launch - CarPlay unavailable - advanced parking/ADAS features postponed - notable standby energy drain because the computing platform stayed awake Volvo pushed multiple large OTA updates to stabilize the situation and roll out missing features. Even after these waves, owners continued to report significant instability. Here's why a full hardware swap became necessary. Volvo positioned the EX90 as a centralized, SDV-ready flagship built on next-gen compute: - Nvidia for safety/ADAS - Qualcomm for cockpit/UX. On paper, the architecture was clean and modern. In practice, several things became clear: - Integrating multiple compute domains proved far more challenging than expected. - The original Nvidia-based setup appears not to have left enough headroom for the planned feature roadmap. - The EX90 launched on a compute platform that was already behind the newest automotive silicon generation planned for future model years. When an OEM decides to replace the central computer across an entire active model year and retrofit the more powerful dual-Orin platform originally meant for later cars, it implies a straightforward conclusion: - Software updates alone were not enough. - The platform needed more compute, more margin, and a cleaner architectural baseline. So, if a software-defined flagship requires a physical brain transplant within its first year on the road, it’s not a batch issue. It’s a sign that the first generation of “centralized SDV” architectures is already hitting its limits under real workloads. This isn’t about blaming Volvo. They’re just the first to make the fix visible instead of hiding behind “feature delays” or perpetual OTAs. More programs will run into the same wall because the root cause isn’t a bug. It’s the mismatch between SDV ambitions and the actual compute, integration discipline, and lifecycle planning that legacy auto is still catching up on.

The #software problems now engulfing Volvo Cars were entirely predictable, and I wrote about them two years ago in the The Ojo-Yoshida Report. The central issue is that specifying whopping-TOPS Nvidia processors, #lidar and touchscreen cabin controls is the easy part that generates headlines. But making the software all work together at #safety grade, all the time, and under all conditions, is the difficult part. As Volvo is now learning the hardest way possible, this too generates headlines, but for all the wrong reasons. Worse, the #userexperience is so terrible that customers can, and are, demanding to return the vehicle for a full refund. #Automakers have been ignoring "Steve's Law," which states: "You’ve got to start with the customer experience and work backwards to the technology. You can’t start with the technology and try to figure out where you are going to sell it." Volvo had its product development strategy backwards, and focused first on #technology, not customer experience. It stands today as a case study for the remainder of the automotive industry, and automakers need to learn this lesson: The software has to work entirely when the vehicle is delivered to the customer. The central problem is the concept of a "software-defined vehicle," which can lull a development team into a false sense of security. It is too easy to specify the base technology components and assume the performance and functionality of the software can be upgraded and optimized through over-the-air updates after delivery of the vehicle to the customer. That might have worked for #Tesla, but Tesla was selling to innovators and early-adopters who will accept beta software and bugs. As Volvo has now learnt, you can't do that to the early and late-majority customer bases, which is where the mass-market #OEMs sell vehicles. Bolting #hardware and software together from different vendors and hoping it all works, doesn't work. Qualcomm was first to develop an entire solution under the name #DigitalChassis including automotive-grade #SoCs both for driving (#Snapdragon #Ride) and the #cockpit (Cockpit applications processor), Arriver driving stack, integration of #Android operating system, and pre-integrated functions such as automated parking stack from Valeo, and #driver #monitoring #system from Seeing Machines. BMW Group was first to grasp the benefits of the Digital Chassis and is lead technology partner working with Qualcomm. Volvo will have to rethink its technology roadmap, but the episode serves as a warning to other automakers such as JLR, Mercedes-Benz AG and Hyundai Motor Company (현대자동차). We now know which major OEM has first run into problems with the “software-defined” mindset. But ahead lies another milestone: Which mass-market automaker is first to be bankrupted by software problems resulting from the software-defined vehicle? We likely won’t have to wait long to find out. https://lnkd.in/eU6HUWbz

If you talk to senior executives at legacy automotive companies today, you will often hear a reassuring narrative: "Yes, new EV players are good at software and screens. But they don't understand vehicle dynamics. Our brand is built on 100 years of chassis tuning, suspension geometry, and driving feel. That is our moat." This is The Myth of Hardware Differentiation. And it is a dangerous illusion. In the hardware era, the driving feel was indeed the brand. A BMW felt fundamentally different from a Mercedes, which felt different from an Audi. That mechanical differentiation was worth a massive price premium. But in the Software-Defined Vehicle era, the rules of differentiation have changed. Today, the majority of consumers do not buy a car because of its suspension geometry. They buy it because of its digital experience. — How seamless is the infotainment? — How reliable is the automated driving? — Does the car improve over time via OTA updates? More importantly, hardware is becoming commoditized. Tier 1 suppliers now offer incredibly sophisticated, off-the-shelf skateboard chassis and suspension systems. A new EV startup can buy world-class driving dynamics from a supplier on Day 1. But they cannot buy a world-class software ecosystem off the shelf. Legacy OEMs are fighting a software war using hardware weapons. They are spending millions of R&D hours perfecting NVH while their software teams are drowning in manual traceability spreadsheets just to release a simple bug fix. If the core brand value is still locked in physical metal, while the competitors are building value in continuous software updates, the moat is shrinking every single day. Is the automotive industry overvaluing "driving feel" at the expense of software velocity? #AutomotiveEngineering #SDV #SystemsEngineering #DigitalTransformation #SoftwareEngineering #MappingSpace

𝗬𝗼𝘂𝗿 𝗦𝗽𝗼𝘁𝗶𝗳𝘆 𝗽𝗹𝗮𝘆𝗹𝗶𝘀𝘁 𝗮𝗻𝗱 𝘆𝗼𝘂𝗿 𝗯𝗿𝗮𝗸𝗶𝗻𝗴 𝘀𝘆𝘀𝘁𝗲𝗺 𝗺𝗮𝘆 𝘀𝗼𝗼𝗻 𝗯𝗲 𝘂𝗻𝗰𝗼𝗺𝗳𝗼𝗿𝘁𝗮𝗯𝗹𝗲 𝗿𝗼𝗼𝗺𝗺𝗮𝘁𝗲𝘀. We talk about “Safety” in automotive constantly, but in a Software-Defined Vehicle the hierarchy of risk is colliding. We’re moving from isolated ECUs to centralized compute, forcing “safe” logic to live next to “efficient” apps. 𝗛𝗲𝗿𝗲’𝘀 𝘁𝗵𝗲 𝗿𝗲𝗮𝗹𝗶𝘁𝘆 𝗼𝗳 𝘁𝗵𝗲 𝗦𝗗𝗩 𝘀𝘁𝗮𝗰𝗸: ⚪ 𝗤𝗠 — “𝗗𝗼𝗻’𝘁 𝗮𝗻𝗻𝗼𝘆 𝗺𝗲.” • 𝗘𝘅𝗮𝗺𝗽𝗹𝗲𝘀: Infotainment, cloud-connected AI, ambient lighting. • 𝗦𝗗𝗩 𝗿𝗲𝗮𝗹𝗶𝘁𝘆: The Wild West. The goal is containment: a crashing 3rd-party app must not starve the network or take down the speedometer. 🟢 𝗔𝗦𝗜𝗟 𝗔 — “𝗗𝗼𝗻’𝘁 𝗽𝗶𝗻𝗰𝗵 𝗺𝘆 𝗳𝗶𝗻𝗴𝗲𝗿𝘀.” • 𝗘𝘅𝗮𝗺𝗽𝗹𝗲𝘀: Window controls, parking sensors. • 𝗦𝗗𝗩 𝗿𝗲𝗮𝗹𝗶𝘁𝘆: Keeping these simple services from clogging up the high-speed network. 🟡 𝗔𝗦𝗜𝗟 𝗕 — “𝗗𝗼𝗻’𝘁 𝗹𝗶𝗲 𝘁𝗼 𝗺𝗲.” • 𝗘𝘅𝗮𝗺𝗽𝗹𝗲𝘀: Instrument cluster, warning tell-tales, OTA agents with safety impact. • 𝗦𝗗𝗩 𝗿𝗲𝗮𝗹𝗶𝘁𝘆: The conflict zone. We’re running critical visualization and update logic on complex SoCs with Linux/Android in the mix. High-performance expectations clash with scheduling determinism. 🟠 𝗔𝗦𝗜𝗟 𝗖 — “𝗗𝗼𝗻’𝘁 𝘀𝘁𝗲𝗲𝗿 𝗺𝗲 𝘄𝗿𝗼𝗻𝗴.” • 𝗘𝘅𝗮𝗺𝗽𝗹𝗲𝘀: Adaptive Cruise, Lane Keeping. • 𝗦𝗗𝗩 𝗿𝗲𝗮𝗹𝗶𝘁𝘆: The grey zone. It needs high compute (AI) and strict safety at the same time—often the hardest place to innovate. 🔴 𝗔𝗦𝗜𝗟 𝗗 — “𝗗𝗼𝗻’𝘁 𝗸𝗶𝗹𝗹 𝗺𝗲.” • 𝗘𝘅𝗮𝗺𝗽𝗹𝗲𝘀: Braking, steering, battery management. • 𝗦𝗗𝗩 𝗿𝗲𝗮𝗹𝗶𝘁𝘆: Validation costs prevent weekly OTAs here. We lean on ASIL decomposition and redundancy to split the safety goal into cheaper, testable parts—and we accept much slower update cadences. Once more content runs on central compute, ASIL stops being a quiet four-letter code in a requirements document. It becomes the grammar for how you design, schedule, isolate and monitor software in motion. 𝗧𝗵𝗲 𝗦𝗗𝗩 𝗴𝗮𝗺𝗲 𝗶𝘀 𝗻𝗼𝘄 𝗮𝗯𝗼𝘂𝘁 𝗼𝗿𝗰𝗵𝗲𝘀𝘁𝗿𝗮𝘁𝗶𝗼𝗻: knowing exactly which parts of the stack are allowed to annoy, which are allowed to lie, and which must never be allowed to kill—and then engineering hard boundaries between them. 𝗥𝗲𝗮𝗹 𝗶𝗻𝗻𝗼𝘃𝗮𝘁𝗼𝗿𝘀 𝘄𝗶𝗹𝗹 𝗱𝗲𝘀𝗶𝗴𝗻 𝗰𝗮𝗿𝘀 𝘄𝗵𝗲𝗿𝗲 𝗰𝗵𝗮𝗼𝘀 𝗮𝗻𝗱 𝗰𝗮𝘂𝘁𝗶𝗼𝗻 𝗹𝗶𝘃𝗲 𝘀𝗶𝗱𝗲 𝗯𝘆 𝘀𝗶𝗱𝗲—𝗮𝗻𝗱 𝘀𝗮𝗳𝗲𝘁𝘆 𝘀𝘁𝗶𝗹𝗹 𝘄𝗶𝗻𝘀 𝗲𝘃𝗲𝗿𝘆 𝘀𝗶𝗻𝗴𝗹𝗲 𝗿𝗮𝗰𝗲.

🚗 What is an SDV? An SDV (Software-Defined Vehicle) decouples hardware and software, enabling: ✔️ Over-the-Air (OTA) Updates: Add features/fixes without dealership visits (e.g., Tesla’s Autopilot upgrades). ✔️ Hardware Abstraction: Software runs agnostic to underlying ECUs/sensors. ✔️ Data-Driven Services: Monetize vehicle data (e.g., predictive maintenance, insurance telematics). Unlike traditional cars, SDVs evolve post-purchase like smartphones. ⚙️ Key Enabling Technologies ✅ Automotive Ethernet: High-speed backbone (10Gbps+) for data flow ✅ Zonal Architecture: Replaces 100+ ECUs with few domains' controllers ✅ Cloud/Edge Compute: Offloads AI tasks (e.g., sensor fusion, navigation) ✅ Hypervisors: Run multiple OSes (Linux/QNX/Android) on one chip ✅ Service-Oriented Arch. (SOA)Apps access vehicle functions via APIs (e.g., SOME/IP) 💡Core Benefits 1️⃣ New Revenue Streams Subscription features (heated seats, autonomy tiers, performance boosts). 2️⃣ Reduced Costs Fewer hardware variants → simplified manufacturing. Remote diagnostics (DoIP) → fewer recalls. 3️⃣ Enhanced Safety Real-time threat detection + security patches (e.g., blocking CAN bus attacks). 4️⃣ Personalization Driver profiles, infotainment, and driving modes adapt via ML. 🏁 Industry Adoption ▪️ Tesla: Pioneer (8+ million OTA updates since 2012). ▪️ BYD: "Smart Car" platform with 5G connectivity. ▪️ GM Ultifi, Ford BlueCruise, Volkswagen VW.OS. ▪️ Tech Giants: Qualcomm Snapdragon Ride, Nvidia DRIVE, Huawei HarmonyOS. ⚠️ Challenges ▪️ Security: 100M+ lines of code → hacker targets (ISO 21434 compliance critical). ▪️ Safety Certification: Mixing ASIL-D systems with consumer apps. ▪️ Legacy Integration: Bridging CAN/LIN networks with IP-based SDV stacks. ▪️ Data Privacy: GDPR/CCPA compliance for user data. 🔮 Future Trends ▪️ AI Co-Pilots: LLMs for natural voice control + proactive assistance. ▪️ Vehicle-to-Everything (V2X): Real-time traffic/road hazard sharing. ▪️ Federated Learning: Cars improve AI models without sharing raw data. ▪️ Blockchain: Secure OTA updates + usage-based insurance. 💡 By 2030: 95% of new cars will be SDVs (McKinsey). 🔧 Development Tools ▪️ Simulators: CARLA, NVIDIA DRIVE Sim. ▪️ Frameworks: Eclipse Velocitas, COVESA Vehicle Signal Specification (VSS). ▪️ Security: Argus Cyber Security, GuardKnox. #SDV #SoftwareDefinedVehicle #Automotive #ConnectedCar #OTA #AutonomousDriving #V2X #FutureMobility #Tesla

Software-Defined Vehicle (SDV) – The Industry Got It Backwards. Everyone talks about #SDV as if it’s defined by: 🔹 Centralized ECUs → Helps with updatability, but not a requirement. 🔹 Over-the-air updates → Just a deployment method, not the core. 🔹 Automated driving features → Possible use case, not the foundation. 🔹Cloud connectivity → Enables data-driven development but isn’t SDV itself. 🔹High-performance computing (HPC) → Supports SDV, but software must be independent of compute platforms. 🔹 Service-oriented architectures (SOA) → Useful for modularity, but SDV is more than just API design. All of these are tools. ❌ Not the core. At its core, SDV means one thing: Software must run across multiple hardware generations. And that changes everything. Today, vehicle software is tightly coupled to hardware architectures. That won’t work in an SDV world. The shift required: ✅ Functional software must derive from the system, not from the hardware. ✅ Hardware (at least from the 2nd generation onwards) must derive from the system and functional software. And even that won’t be enough. SDV demands a new abstraction layer - a ‚Vehicle API‘ - to isolate functional software from hardware specifics. ➕A low-level software layer that allows new hardware generations to integrate seamlessly without breaking existing software. This isn’t just a new E/E architecture. It’s a new way of developing cars.

We’re entering the era of software-driven mobility. But is everybody ready? Over the summer, I had the pleasure of interviewing experts and analyzing the thoughts and opinions of ~600 automotive execs from ~200 companies to understand the state of software transformation in the industry. The result of this work is our latest report:  The software-driven mobility era: Beyond vehicles I believe it is one the most-comprehensive reports of its kind and it highlights the critical juncture at which the majority of the automotive industry finds itself today. And while I encourage you to read the report in its full, glorious detail, here’s the story in as close to a nutshell as the character limit here allows: 1️⃣ Almost every automotive company recognizes the value and importance of software to its business. Cost savings, improved efficiency, reduced environmental impact, elevated ability to create and launch new services, and new ways to enhance customer experiences were the top benefits cited. 2️⃣ Most automotive companies will become software organizations. 3️⃣ Software-defined products and services will account for MORE THAN HALF of OEM revenues by 2035. And most agree software will emerge as the single biggest source of competitive advantage.  On the right track? Maybe, but although there were positive signs at IAA Mobility, the road ahead figures to be anything but smooth. 4️⃣ Despite acknowledging the importance of software, progress towards achieving software transformation and delivering software-defined vehicles, a key enabler, has been slower than expected. Only around 1 in 5 OEMs has fully scaled their software-driven mobility use cases. And only 1 in 10 organizations has decoupled hardware and software architectures and development cycles. So, what’s holding companies back? Here’s an abbreviated list: - Lack of skills, and difficulty in reshaping organizations and driving a disruptive cultural shift; - Inability to reform long-established practices to define, engineer, and maintain products and services; - Achieving compliance, maintaining high safety guarantees, and ensuring appropriate cybersecurity. And as if that were not enough, the situation is compounded by the fact that: 5️⃣ Digital-native new entrants are not standing still – they’re redefining customer experiences at lightning speed and ramping up competitive pressure. New players (mainly Chinese) are standardizing and collaborating on common functions, preserving innovation capacity to develop differentiating features and services (e.g. autonomous mobility) Importantly, they’re applying a ‘user-centric’ approach to deliver integrated and extensive user experiences that go far beyond the one-time purchase of vehicles, continuously delivering value – inside and outside the vehicle, across the mobility experience.  So, what’s to be done? To find out, you’ll just have to read the report 😉 https://bit.ly/4g9faii 

𝐇𝐲𝐩𝐞𝐫𝐯𝐢𝐬𝐨𝐫𝐬 𝐰𝐢𝐥𝐥 𝐧𝐨𝐭 𝐭𝐚𝐤𝐞 𝐲𝐨𝐮 𝐭𝐨 𝐋𝐞𝐯𝐞𝐥 𝟒 𝐝𝐫𝐢𝐯𝐢𝐧𝐠. 𝐌𝐂𝐎𝐬 𝐦𝐢𝐠𝐡𝐭. 🚦 Most people think hypervisors are enough to manage safety in consolidated vehicle compute. They are not. A hypervisor is like building walls between applications. It locks resources for each partition, keeping safety-critical and non-safety tasks separate. That works for today’s ECUs and some early SDVs. But walls are static. They cannot adapt when your AI workload suddenly demands 10x more GPU. They cannot balance dozens of mixed-criticality tasks running side by side on a single SoC. They cannot extend resource guarantees to the edge or cloud. This is where the 𝐌𝐢𝐱𝐞𝐝-𝐂𝐫𝐢𝐭𝐢𝐜𝐚𝐥𝐢𝐭𝐲 𝐎𝐫𝐜𝐡𝐞𝐬𝐭𝐫𝐚𝐭𝐨𝐫 (𝐌𝐂𝐎) enters the picture. 🔹 Think of an MCO as the 𝐭𝐫𝐚𝐟𝐟𝐢𝐜 𝐜𝐨𝐧𝐭𝐫𝐨𝐥𝐥𝐞𝐫. Instead of static walls, it actively decides who gets CPU cycles, memory, and timing windows at any given moment. 🔹 It ensures safety-critical functions always behave as if they were alone on the chip. 🔹 It adjusts resources dynamically depending on the driving context. Highway cruising vs. urban navigation? Different needs, same safety guarantees. 🔹 It extends orchestration beyond the car into edge nodes and cloud environments. Without something like an MCO, Level 4 autonomy either becomes too rigid (hardware-locked safety partitions) or too fragile (static hypervisors under dynamic AI load). 👉 With an MCO, safety stays uncompromised while flexibility finally scales. That is the missing link to making SDVs manageable and L4 driving realistic. Now the open question: Would you trust a software-defined orchestrator to guarantee safety as much as silicon itself? I’d love to hear your perspective. Drop your view in the comments. 👇 #SDV #Automotive #SoftwareDefinedVehicle #AutonomousDriving #Cloud #EdgeComputing #AI #Mobility #FutureOfCars

I spent three decades watching industries convince themselves their complexity was unique: healthcare, banking, telecom, enterprise software. Different industries, but the same thing kept happening inside major transformation programs. People lost sight of what the system was for and who it was supposed to serve. Then automotive did it with the software-defined vehicle (SDV). I was not watching from the outside. I believed in the destination early, worked inside the thesis, and I still think the destination is right. Vehicles should improve over time. Capability should be updatable. Customers should not need to buy a new vehicle just to get meaningful improvements. But the execution philosophy broke the thesis. I wrote the full diagnosis here: https://lnkd.in/edBGCWgT The piece covers the numbers, the standards spiral, the compute trap, the certification wall, the AIDV rebranding, Tesla and BYD, and the question I still do not think the industry has answered honestly: who was this actually being built for? Because too much of the SDV movement looked less like product transformation and more like enterprise architecture theater. The consultants got paid, the platform vendors got their contracts, the Tier 1s built proprietary layers on top of every standard and charged integration fees to make them interoperate, and the OEM absorbed the schedule risk, the program cost, and the warranty exposure while the customer was left hoping the vehicle would simply work on a Tuesday morning. The piece gets specific. The Volvo EX90 case is in there and it holds. This is not an argument against software-defined vehicles. It is an argument against an execution model that turned the SDV from a customer-centered architecture into a capital-market narrative, a standards exercise, and a compute procurement race. Tesla and BYD do not disprove the argument. They expose how hard this transformation is inside an inherited operating model, which is exactly the condition every legacy OEM is operating under. The vision was right. The execution was a controlled demolition of the thesis that funded it. Read the article first. Then come back here or discuss directly on Substack. I expect disagreement and I am genuinely looking forward to it. If you work in automotive, Tier 1, SDV platforms, ADAS, OTA, enterprise architecture, or vehicle software, tell me where I am wrong. Not the complexity defense. The specific argument. #automotive #sdv #enterprisearchitecture #adas #ota #vehiclesoftware #digitaltransformation #operationalassurance #autonomousvehicles #softwareengineering

How do you run an infotainment app and a steering-angle calculation on the same chip without risking lives? The consolidation of automotive E/E architecture is pushing us toward massive, centralized High-Performance Computing (HPC) nodes. We are successfully moving away from dozens of isolated, single-function ECUs. But this transition has introduced a massive, low-level engineering headache: In a true Software-Defined Vehicle, your central computer might simultaneously run: 1. ASIL-D functions (e.g., torque vectoring, advanced driver assistance) 2. ASIL-B functions (e.g., digital instrument cluster graphics) 3. QM (Quality Management) functions (e.g., cloud telemetry, over-the-air updates, infotainment) Historically, these lived on physical hardware islands. Today, we are forcing them to share the exact same silicon, memory busses, and cache. How do we prevent a memory leak in a streaming app from starving the CPU cycles needed for predictive braking? It comes down to three pillars of modern automotive middleware and virtualization: 🔹 Freedom from Interference (FFI) via Type-1 Hypervisors: Utilizing hardware-enforced virtualization (like ARM microarchitectures supporting EL2 execution levels) to strictly partition memory and peripherals. 🔹 Time and Space Partitioning: Moving beyond simple round-robin scheduling. We need deterministic, time-triggered scheduling (like ARINC 653 concepts adapted for automotive) to guarantee that ASIL-D tasks get their dedicated execution windows, no matter what. 🔹 Advanced QoS in Automotive Middleware: Ensuring that data distribution frameworks (like DDS or localized SOME/IP implementations) can prioritize safety-critical signals over background diagnostic data. The transition to Zonal and Centralized computing isn't just a hardware consolidation play; it’s a foundational rewriting of how we handle deterministic computing. What is the biggest bottleneck you’re hitting when balancing hypervisor overhead with strict real-time determinism? Thanks, Ani #AutomotiveEngineering #SoftwareDefinedVehicles #EmbeddedSystems #SDV #AUTOSAR #Hypervisors #VehicleArchitecture