Signals · Article · Automotiveシグナル · 記事 · 自動車
In-Vehicle Networks [Explained]
CAN was announced in 1986 and is still installed in billions of nodes a year. Understanding why tells you how the car's nervous system is really being rebuilt.

There is a sentence I have heard at automotive conferences for a long time. CAN is old and slow, and Ethernet is going to replace it. I have nodded along to it more than once. Then you look at the numbers from CAN in Automation, the international group of CAN users and manufacturers, and find that more than two billion CAN nodes are still being installed every year.
Both things are true at once. Cars are moving to Ethernet for the heavy traffic, and the old buses are staying for the jobs they do well. A modern vehicle runs several kinds of network side by side, each chosen for a different job. This article explains how they work, why each one exists, how a single message travels across the car, and why security had to be added to networks that were designed when nobody imagined an attacker plugged into a headlight.
Before networks there was copper
Through the 1970s, if one electronic box needed information from another, engineers ran a dedicated wire between them. As cars added engine management, anti-lock brakes and automatic transmissions, the wiring grew in weight, cost and complexity, and the same sensor reading often had to be wired separately to several controllers.
Bosch started work on a shared serial bus in 1983 and presented it at the SAE congress in Detroit in February 1986 as the Controller Area Network. Intel delivered the first CAN controller chip, the 82526, in 1987, and Philips followed with the 82C200. The Mercedes-Benz W140 S-Class, launched in 1991, was the first production car with a CAN network.
ISO standardized CAN as ISO 11898 in November 1993. The principle that made it succeed was simple. Instead of a wire for every conversation, every controller listens to the same pair of wires and picks out the messages it needs.
How CAN decides who gets to speak
CAN messages are not addressed to a particular receiver. Each one carries an identifier that describes its content, such as engine speed or a door status, and every node on the bus sees every message. Each node decides for itself whether to use it.
The clever part is what happens when two nodes start talking at the same moment. CAN uses two signal states. A dominant bit, representing 0, always overrides a recessive bit, representing 1. All transmitting nodes send their identifiers bit by bit while listening to the bus. A node that sends a recessive bit but hears a dominant one knows another node has a higher-priority message, so it stops and tries again later. The message with the lowest numeric identifier wins, and it is never corrupted by the contest. Identifiers have to be unique on a bus, which is what guarantees a single winner.
Picture a meeting where everyone who wants to speak starts reading out their ticket number at the same time, digit by digit, and anyone who hears a lower digit than their own quietly sits down. By the end, exactly one person is still standing, and they never had to repeat themselves. That is why safety-relevant messages get low identifiers.
Classical CAN uses 11-bit identifiers, with a 29-bit extended format used heavily in trucks. Each frame carries up to 8 bytes of data at up to 1 megabit per second, and at that speed the bus can only stretch a few tens of meters. Those numbers sound tiny next to Ethernet, and they miss the point.
CAN was built for short, frequent control messages that have to arrive in a predictable order, with strong error detection and automatic retransmission. For a high-priority message such as a brake controller announcing wheel slip, eight bytes that reliably win the bus are worth more than raw bandwidth.
Mandated, then everywhere
Regulation pushed CAN into every car. The United States required on-board diagnostics, known as OBD-II, on cars and light trucks from the 1996 model year, allowing several communication protocols. CAN was phased in from 2003 and became the required protocol for that diagnostic connection from the 2008 model year. That rule applies to the diagnostic port, not to how the rest of the vehicle is wired, but it made CAN universal in practice.
Heavy vehicles standardized on SAE J1939, which runs on CAN with 29-bit identifiers that encode the type of data being sent. Tractors, buses, ships and industrial machines adopted related standards, which is part of why CAN in Automation still counts more than two billion CAN nodes installed each year.
The supporting cast
Not every job needs CAN. LIN, developed by a consortium of carmakers and chip suppliers formed in 1998, which released LIN 1.0 in 1999, is a cheap single-wire bus that runs at up to about 20 kilobits per second. One commander node polls a handful of simple devices such as window switches, mirror motors and seat controls. It is standardized as ISO 17987 and is still common wherever cost matters more than speed.
FlexRay went the other way. Founded in 2000 by BMW, DaimlerChrysler, Motorola and Philips, it runs at up to 10 megabits per second on two channels and divides time into fixed slots so every critical message has a guaranteed place in the schedule. Its first production use was BMW’s adaptive chassis system on the X5 in late 2006. The consortium disbanded at the end of 2009, and the standard lives on as ISO 17458, but most new high-speed designs now look to CAN FD and Ethernet.
Infotainment once had its own network too. MOST, backed by a cooperation formed in 1998, sent audio and video around a ring of plastic optical fiber. Ethernet has since taken over much of that work.
Cameras and displays usually sit outside all of these. Their raw video needs far more bandwidth than a shared bus can provide, so they use dedicated high-speed point-to-point serial links, some proprietary and some newer open standards.
CAN keeps growing up
CAN did not stand still. CAN FD, introduced by Bosch in 2012, keeps the same arbitration but switches to a faster bit rate once a node has won the bus, and raises the payload from 8 to 64 bytes. CAN in Automation says typical networks run the faster data phase at about 2 megabits per second, point-to-point links at 5, and up to 8 with newer transceivers. It was standardized in the 2015 edition of ISO 11898-1, and the original Bosch version and the ISO version do not interoperate, a detail that still trips people up.
CAN XL goes much further. Work began in late 2018, and it is now part of the 2024 edition of ISO 11898-1. It carries up to 2,048 bytes per frame, reaches data rates of 10 to 20 megabits per second with special transceivers, and can carry Ethernet frames inside its own. That last feature tells you how its designers see the future. CAN XL is meant to sit alongside Ethernet in a zonal car, not to fight it.
Adoption takes time. CAN in Automation expects CAN XL to appear in passenger cars toward the end of this decade.
Ethernet learns to drive
Standard office Ethernet was never going to survive in a car. It used too many wires, emitted too much electrical noise and cost too much. Broadcom’s BroadR-Reach changed that by sending 100 megabits per second full duplex over a single unshielded twisted pair. In 2011 BMW, Broadcom, NXP, Freescale, Hyundai and others formed the OPEN Alliance to turn it into an open standard.
BMW had used conventional Ethernet for diagnostics and software flashing since 2008. It put single-pair Ethernet into production first, in the X5 launched in 2013, where single-pair Ethernet linked the surround-view parking cameras to their controller. Broadcom claimed the cabling cost about 80 percent less and weighed about 30 percent less than the previous solution. IEEE then standardized the approach as 100BASE-T1 in October 2015, followed by 1000BASE-T1 at one gigabit in 2016, 2.5, 5 and 10 gigabit versions in June 2020, and 25 gigabits in 2023.
The most interesting addition goes after CAN’s own territory. 10BASE-T1S, completed by IEEE in 2019, runs at 10 megabits per second and lets several nodes share one wire pair. An optional scheme called physical layer collision avoidance gives each node a turn in strict rotation, which keeps timing predictable. It means simple sensors and actuators at the edge of a zonal car could speak Ethernet directly, without a translation step in the zone controller.
Ethernet does not guarantee timing by itself. Automotive designs add Time-Sensitive Networking features from the IEEE 802.1 family, such as a shared clock across the network and scheduled transmission windows, so control messages arrive when they must.
Above the cables sit software protocols. SOME/IP, which began at BMW and is now part of AUTOSAR, and DDS let applications offer and request services across the network, which is what makes a service-oriented software-defined vehicle possible. Diagnostics moved to Ethernet too, through Diagnostics over IP, which lets a workshop or the factory reflash large software images far faster than over CAN.
Follow one message across the car
Take wheel speed. A sensor at each wheel reports to the brake and stability controller, usually through a dedicated sensor interface rather than CAN. The brake controller turns those readings into a CAN or CAN FD frame on the chassis network, with an identifier whose priority decides when it wins the bus.
A gateway, or in a zonal car the nearest zone controller, receives that frame. It checks whether the message is allowed to leave its domain, then forwards it to another CAN bus or wraps the data in an Ethernet message and sends it across the backbone to the instrument cluster and the central computer. The speedometer you read and the driver assistance software deciding whether to brake may be looking at the same original reading, delivered by different networks with different timing guarantees.
A network with no locks on the doors
Classical CAN has no built-in authentication or encryption. Any node can send any message, and every other node will believe it. That was reasonable in 1986, when the only way onto the bus was through the car’s own wiring. It stopped being reasonable once cars had cellular modems and Bluetooth.
In 2015 security researchers Charlie Miller and Chris Valasek showed they could reach a Jeep Cherokee remotely through its Uconnect head unit and send commands that affected driving functions. Fiat Chrysler recalled about 1.4 million vehicles. In April 2023 automotive security engineer Ken Tindell and researcher Ian Tabor published how thieves had stolen Tabor’s Toyota RAV4 the year before. They pulled back the bumper to reach the headlight connector, plugged in a device that impersonated the smart key controller, and injected messages telling the car a valid key was present.
Defenses now come in layers. Gateways and zone controllers separate networks and block messages that should not cross between them. AUTOSAR’s Secure Onboard Communication adds a truncated authentication code and freshness value to sensitive messages, including on CAN, so receivers can reject forgeries and replays. Ethernet links can use MACsec, defined in IEEE 802.1AE, and CAN in Automation has specified CANsec, a security layer for CAN XL. UN Regulation 155 now requires manufacturers selling in Europe and other adopting markets to manage these risks across the life of the vehicle.
None of it is free. Authentication adds bytes to messages that only had eight to start with, and keys have to be managed across dozens of controllers for fifteen years or more. Security is now a network design constraint, not an afterthought.
What the network really means
The car is not moving from CAN to Ethernet. It is moving to layers. LIN and CAN stay at the edge for simple, cheap and predictable control. CAN FD, CAN XL and 10BASE-T1S handle the middle. Multi-gigabit Ethernet carries cameras, sensor data and software traffic across the backbone to central computers.
Each layer answers a different question. How much data needs to move? How certain must its timing be? How much can each connection cost? And who is allowed to send it? The engineering skill is no longer picking one network. It is deciding which message travels on which network, with which guarantees, and how each one is protected from the next person with a laptop and a pair of wire strippers.
Words you will hear
CAN. Controller Area Network. A two-wire shared bus where messages carry priority identifiers and the lowest identifier wins access.
CAN FD and CAN XL. Newer CAN generations with larger payloads, up to 64 and 2,048 bytes, and faster data phases.
LIN. Local Interconnect Network. A low-cost single-wire bus for simple devices such as switches and small motors.
FlexRay. A time-triggered 10 megabit per second bus with fixed transmission slots, used in some chassis systems.
Automotive Ethernet. Ethernet adapted to run over a single twisted pair, from 10 megabits to 25 gigabits per second.
Gateway. A controller that connects networks, filters what may cross between them and translates between protocols.
SecOC. AUTOSAR Secure Onboard Communication. Adds authentication codes to messages so receivers can reject forged ones.
This article is part of the AutomotiveCloudWatch Explained series, which also covers software-defined vehicles, zonal architecture, over-the-air updates, drive-by-wire and autonomous driving. For the operating system that runs on many of the controllers sending these messages, read How QNX Works and Why It Is Difficult to Replicate.
How QNX Works and Why It Is Difficult to Replicate
If this helped, subscribe free at automotivecloudwatch.com to get each new explainer when it publishes.
All opinions are my own and do not reflect those of my employer.
Get the next Executive Briefing by email
Each new briefing, in your inbox.
Subscribe




