Time-sensitive networking and MACsec allow Auto Ethernet to handle safety-critical functions and enable real-time processing for AI-enabled features and agents.
Key Takeaways:
Next-generation cars are moving from software-defined vehicles (SDVs) to AI-defined vehicles (AIDVs), bringing a slew of reliability and security requirements to ensure sensor data gets where it needs to go, and quickly enough to be useful.
Automotive Ethernet, combined with time-sensitive networking (TSN) and media access control security (MACsec), is needed to enable sensor fusion from multiple modalities, including cameras, lidar, radar, 4D imaging, ultrasonic sensors, and thermal imaging. Data from all of those is then incorporated into an artificial intelligence (AI) model, AI agent, or large language model (LLM), with a voice-based human-machine interface (HMI).
“The AI-defined vehicle is a sophisticated AI vehicle with models in the cloud as well as in the car, and that allows for AI to be pervasive across the vehicle, whether in the autonomous driving system, in the IVI (in-vehicle infotainment) system, or in the controls of the vehicle,” said Dipti Vachani, senior vice president and general manager of automotive at Arm. “This AI-defined vehicle still has all the requirements that go with the software-defined vehicle, but now we’re seeing the increased demand of AI, increased demand of compute, and imagine the software explosion that happens with this AI-defined vehicle.”
Fig. 1: An AI-defined vehicle concept (AIDV). Source: Arm
One advantage of more AI in the vehicle is contextual awareness. “The vehicle not only is reacting to what my camera sees in front of me as a car, but also to the environment I am in,” said Vachani. “Am I driving on a highway? Am I driving in the city? If I’m driving in the city, do I need to change the way I drive? That’s contextual awareness. We’re seeing it in the ADAS systems, as well as in the in-vehicle entertainment. When the vehicle starts to drive itself more, what do you expect in terms of in-cabin and infotainment capabilities? We’re already seeing some of our customers putting LLM models in the car, so you can talk to the car and tell the car, ‘I want to order my Starbucks on the way to work,’ and it’s ordered, and it’s done right.”
Fig. 2: AI-defined vehicle compute platform differentiated for OEM goals. Source: Arm
Not all OEMs will lean into AIDVs right away, but voice is a fast-growing interface, and it’s becoming more sophisticated compared to existing assistants that just turn on music or open maps.
“There is enough compute in some of the more expensive cars out there that this will get better through over-the-air software update capabilities,” said John Weil, vice president and general manager of IoT and edge processors at Synaptics. “In cars that can update, voice already works with services such as Google Voice or Apple CarPlay, but it’s clunky compared to the latest edge AI MCU capabilities for embedded IoT.”
AI and voice capabilities are growing in importance now that mechanical competence is no longer the only differentiator for OEMs. “Differentiation is coming down to things like battery life and in-cabin experience,” said Rob Fisher, senior director of product management at Imagination Technologies. “OEMs will want to own that so they get a connection with their customers, and they can impress their branding on that environment. It’s a complex thing to do, and that’s why Apple CarPlay and Google Android are so popular and well used. But it does grate against the OEMs, because it means they’re losing their customer contact.”
Partnerships and ecosystems around chip companies and automotive OEMs offer a way to compete in today’s fast-moving market. “Creating a seamless in-car experience — as seamless as the way you use your mobile phone — is a massive investment and needs a lot of thought and energy,” Fisher observed.
The role of time-sensitive networking
TSN enables Automotive Ethernet to achieve the low latencies required for traditional automotive real-time subsystems, in addition to enabling the sensor fusion and synchronization with AI processors needed for AIDVs.
“One of the things that really will make a difference in Automotive Ethernet is TSN,” said Imagination’s Fisher. “This came from what used to be called AVB (audio video bridging), which was an audio spec for big installations, where you had to ensure that your audio is matched on different speakers. But it’s now being used in automotive, and it gives you a way to ensure quality of service and windows through which data must arrive at the endpoints. That’s critical in trying to address the things that CAN bus does very well. TSN is one of the key things that will allow Automotive Ethernet to be adopted more widely.”
Standards and adoption of Automotive Ethernet with TSN are increasing. “TSN has already demonstrated its value in enabling low-latency, synchronized communication across automotive subsystems,” said Adiel Bahrouch, director of business development for silicon IP at Rambus. “Standards like IEEE 802.1AS are being deployed in production vehicles, and the industry is now pushing for enhancements that support zonal architectures and centralized compute. Fast startup, dynamic bandwidth allocation, and redundancy are key areas of focus.”
Others agree. “TSN is extremely important for Automotive Ethernet, as it enables deterministic and low-latency communication essential for safety-critical systems such as braking and steering,” said Seung-Taek Chang, SDV solution manager in the Automotive & Energy Group at Keysight EDA. “Traffic shaping and policing help manage bandwidth allocation and prevent congestion, while fair queueing ensures that high-priority traffic, such as ADAS data, is delivered on time without starving lower-priority services. Together, these capabilities are fundamental for allowing Ethernet to reliably replace legacy real-time protocols.”
Fig. 3: Automotive Ethernet with other in-vehicle networks. Source: Keysight EDA
TSN’s importance as a traffic cop is increasingly important as more data is moved around a vehicle. “It’s all about how queues of traffic in the chip are managed,” said Jon Ames, principal product manager for the Synopsys Ethernet IP portfolio. “If you want to transmit a packet, but the lines are really busy, you have to wait until that finishes. TSN looks after all of that.”
TSN is not exclusive to Ethernet, and some of these functions have been part of Ethernet for many years, but it’s always improving. “For a designer building silicon for cars, it is important to understand all the real-time pieces,” Ames said. “You might want to connect your door opening or rear-facing camera to something at the same time. You might want to connect your lidar system to something. All of those have to work all of the time without delay, whether it’s buttons to open doors, or detecting a person in the dark.”
Without TSN, Automotive Ethernet runs the risk of dropping packets. “You’re running wires to your switches, and when you press a switch, you want it to act straight away,” said Ames. “This ensures that your message doesn’t get lost or delayed, which is really important. If you’ve got several things that are transmitting packets down the wire, then to a switch, and then you link to the ECU or to the computer, there’s a chance with Ethernet that you might drop your packet. That’s not cool, because then it takes a software layer to realize and retransmit.”
In vehicles, TSN is used for a wide range of functions, including its original use case of audio. “If you want to send audio over your wires, you’ve got to make sure that your audio is transmitted regularly, so that you don’t get any dropouts in the sound, and the same goes for video,” Ames explained. “It’s all about traffic shaping, making sure that everything gets fair access to the wire. There’s a policing element, which says, ‘I’m receiving stuff, but maybe I shouldn’t take one of these packets.’”
Determinism is more important at lower speed Automotive Ethernet compared to the higher speeds, which are coming. “10BASE-T1s is important right down at the physical level,” he said. “It’s lighter due to using a single unshielded twisted pair (UTP) of wires, and it’s also logically lightweight. TSN is important to enable reliable connectivity when there might be constrained bandwidth. If you’ve got plenty of bandwidth, you don’t necessarily need so much TSN functionality because when we set the packet, it just goes. But if you’re running 10 megabits per second, you want to make sure your packets really get there when they need to.”
David Fritz, vice president of hybrid-physical and virtual systems for automotive and mil-aero at Siemens EDA, agreed that TSN has different value for different Ethernet speeds due to the bandwidth utilization rate. “If you have something that runs around 10 gigabits per second — which we do have in Automotive Ethernet SerDes running 10 Gbps, and we’ve had 5 Gbps for five years, so it’s pretty mature — then the amount of traffic that you transmit across that is a small percentage of that total bandwidth capability. So arbitration is never really a problem. Determinism doesn’t hit until you get up around, say, 80% utilization of that bandwidth. That means if I had terabit Ethernet, then I would never have to worry about determinism because everything’s going to get there faster than I need it.”
While these developments mean a car is truly starting to look like a data center, and the sectors do borrow technology from each other, important differences remain.
“In data centers, SerDes enables the highest Ethernet speeds, including 100G, 200G, 400G, and moving toward 800G and 1.6T, as specified by standards like IEEE 802.3ck, to maximize performance and power efficiency,” said William Chen, product marketing group director for design IP at Cadence. “However, automotive-rated SerDes are generally not adopted in mainstream data center switches or NICs. Their design constraints, such as lightweight single-pair twisted cable, asymmetric links, and performance envelopes, do not align with data center requirements, which often favor fiber or established copper for reach, density, and performance.”
Fault tolerance and redundancy
Fault tolerance is an important part of the Automotive Ethernet reliability equation.
“With a lot of OEMs, what we’re observing is that they’re doing either zonal or ring architecture, which are used interchangeably,” said Benjamin Tan, senior applications engineer at Infineon Technologies. “You might have two switches in the front of your car, and two in the back of your car, with another switch in the center, which will probably be connected to your seats for up, down, left, right, because it has 10BASE-T1S. If one of these paths is cut off, the data path can be sent in the other direction, as well. The fault-tolerant aspect is that, if a cable is removed, the data path for the camera and all the sensors will still go through. If another path is cut off, it would flow another way.”
Similar to many automotive standards, fault tolerance is about redundancy. “They want to make it reliable in case something happens during startup of the car, or maybe one of the wires comes loose, at least the data path can still go to the central processing unit, which is possibly located in the front of the car,” Tan explained. “Fault tolerance with double redundancy is based on IEEE 802.1CB, frame replication and elimination for reliability (FRER).”
Reliability expectations differ between automotive and the data center. “Data centers demand extremely low bit error rates, high uptime, and redundant systems, while automotive components must operate reliably in harsh environments, with different failure modes and qualification standards such as ASIL and ISO,” Cadence’s Chen noted.
Automotive safety testing standards, such as ASIL A and B, must be considered from an IP and silicon point of view. “It’s about rigorously going through all of the cases of the IP in different modes — switching, etc. — to make sure there are no cases where it might fail,” said Synopsys’ Ames. “Then it’s about additional IP to improve the functional safety of the IP itself. For example, you might add something to protect registers so you could be very sure that when the register’s written, it doesn’t get corrupted, and then you have duplicate registers.”
All of the rigorous testing puts up development costs for an IP or for silicon. “Having redundant functionality adds to the silicon cost,” Ames observed. “It’s an interesting thing, because we’re looking at very cost-sensitive applications, and yet we are adding cost to the IP to enable that. However, the volume will be there. The fact that it’s Ethernet means you don’t have to develop things in the first place, and it doesn’t matter what the technology is — CAN or Ethernet — you need to have that same kind of reliability or functional security in the IP.”
Others agree. “If you look at automotive trends at the moment, you have vehicles with lots of different types of sensors,” said Matthew Bubis, director of product management at Imagination Technologies. “You have different cameras, microphones, lidar, etc. There are a huge number of different inputs that are coming in, and there are a bunch of compute and potentially AI, as well, and often both of them combined. That has to provide an output in a complex system of steering, speed, and brake control, and all of these control devices. It has to do it extremely reliably while meeting all of these safety standards, because if it gets it wrong, there are some significant safety consequences for the driver or for the public. Robotics is in a very similar space.”
MacSec to protect connectivity and revenue streams
The incremental functionality coming to the Automotive Ethernet space is mostly around reliability, but security is also a large part.
“Designers must meet stringent security requirements for safety-critical applications, and these are often in flux as threats evolve,” said Rambus’ Bahrouch. “Integrity, authenticity, confidentiality, and intrusion detection all add processing overhead, which can increase latency and degrade the very speed gains that ultra-high-speed Ethernet promises. Balancing security with deterministic performance is one of the hardest considerations.”
Interoperability and API standardization become essential at the ecosystem level. “This ensures that security mechanisms are consistently applied across vendors without compromising scalability, cost, or latency,” Bahrouch explained. “Standards like IEEE 802.3ch and MACsec (IEEE 802.1AE) are evolving to meet these demands, although interoperability, silicon integration, and cost remain significant hurdles.
Fig. 4: MACsec vs. IPsec vs. TLS (transport layer security). Source: Rambus
One of the main attack vectors for vehicles is connectivity.
“The car is now a rolling network, and it has a gateway to the internet or the cellular network,” said Dana Neustadter, senior director of product management at Synopsys. “Then, on top of the cellular network, we’re now operating an IP network. The bandwidth of that today is somewhat limited, but it’s still a pretty quick link, and if you’re in an urban area, that’s a decent bandwidth link. The attack vector is to come in through the gateway and get into the vehicle system, and then start trying to break that down. As we move toward more Ethernet in vehicles, people are adding MacSec to the Ethernet, which means that the only things that are allowed to get onto that network are the ones that are admitted because they’ve proved that they’re supposed to be there.”
MACsec reduces risk for certain kinds of physical attacks. “The most vulnerable components are weak devices in the infotainment system, or elsewhere in the vehicle system, that may allow an adversary to break in and get access, get some software running on the network, and then start exploring what’s in the network there,” said Neustadter. “That goes back to the original Jeep attack. Security has gotten much better. Vehicle manufacturers, OEMs, and Tier 1s in the motor vehicle industry who control access to the platform level of the system now recognize they have a responsibility and an actual, genuine need to provide a secure system, because they’re also providing a safe system. There’s no safety without security.”
Fig. 5: The case for layer 2 security. Source: Rambus
Another use case for MACsec is to help ensure secure software updates. “If you do an update to the car, you’re going to make very sure it’s reliable,” said Synopsys’ Ames. “MACsec is used on that connectivity, which could be at the vehicle workshop or another location.”
It can also protect the vehicle from unauthorized widgets. “OEMs recently talked about MACsec from a different perspective,” Ames added. “They want to be able to control the third-party aftermarket. They want to ensure that you can’t just plug other widgets into your car, and MACsec can help with that.” Off-brand aftermarket widgets can be a security risk, but the main concern is that they reduce OEM revenue streams, he explained.
Conclusion
Automotive Ethernet and standards such as TSN and MACsec are enabling vehicle network designers to use the latest high-speed protocols for safety-critical elements in the vehicle, which were previously reserved for CAN bus and other older protocols and standards.
Whether Automotive Ethernet will entirely replace other networking protocols, it is the undisputed backbone of next-gen software-defined and AI-defined vehicles.
“We’re building off of what we’ve already created with software-defined vehicles into the next generation of the AI-defined vehicle where more AI workloads are going into the car,” said Arm’s Vachani. “The entire industry is thinking about how to deploy more AI workloads in the car. No one is debating whether more AI is going into the car, because that’s going to happen. That’s not an ‘if,’ it’s a ‘when.’ It’s more of a conversation of, ‘What does it look like, what does my brand stand for, and how do I want my consumers to experience AI in my car?’”
Related Articles
Auto Ethernet 10BASE-T1S Steps Up, With Tbps On The Horizon
High-speed data movement will be required in future vehicles, probably including optical, but challenges persist.
How Long Will CAN Stick Around As Rival Networks Speed Up?
New in-vehicle networking technology will likely take over as more AI is added, but in the near term, designers face challenges integrating new with old.
 |
 |
 |
 |
 |
 |
 |
 |
Leave a Reply