Why would you need 18 CAN interfaces?

[daqq]

Ti released a chunky processor: https://www.ti.com/lit/ds/symlink/am69.pdf

With a GPU, 8 big ARM cores, 4 smaller ARM cores, 4 DSP cores it seems quite powerful. My question is, in a real world scenario, where do you need 18 CAN-FD interfaces? Isn't the point of a bus that you connect more things to it?

Not complaining, just would like to know the monstrous application that has such demands.

[Jeroen3]

Cars have more than one CAN bus. CAN bus is limited in the number of nodes if there is also a need to a usable amount of communication.

[mikeselectricstuff]

Multiple buses for more bandwidth?

[Whales]

I was going to say that keeping multiple, isolated can busses is good for security & safety (eg don't put the entertainment system on the same bus as the abs brakes or the pedal position pots).  But then I realised that if you connected them all to this unit you would violate that principle anyway.

GPU accelerated ABS.

[Psi]

I can imagine a modern electric car might need that number of CAN buses.

Especially when you start to think about adding self-drive with all the cameras / radar / lidar etc..

You want some separation between different systems.

[mikeselectricstuff]

One of their listed applications is robotics - quite plausible you'd have a seperate CAN bus to each motor/sensor

[JPortici]

A strong point is physical separation of CANBUS networks for security reasons. In the automotive world many, if not most, car brands have the keyfob data running unencrypted in a can line that is easily accessible from the ouside (what could possibly go wrong?) because those ECUs don't have many controllers to begin with, and security is always an aftertought

So you have a beefy multicore CPU and an abundance of CAN controllers, each core gets assigned some and the task it runs will only access those.
ECU for ADAS is an obvious application for this part given the other video and neural network hardware accelerators

Something inustrial for security as well, or a dedicated PLC that does vision and interface with stuff over the network.
Robotics, drones, due to the Cortex R cores. year more or less the list of applications given in the datasheet

Wonder how much it costs, 50? 60 €?
Edit: 80-150€ for 1k units 

[Jeroen3]

There is also the possibility that some of specs are tailored for at customer request.
At this scale of chips using the spare gates to fit some more CAN busses doesn't have that much of an impact.

I think it is specifically targeted as main controller for autonomous vehicles, with the two realtime controllers, camera feeds, audio io, display controller, PCIE lanes (for the neural network chip), ethernet and ecc ddr4. I mean, they sell it as TDA4VH-Q1 as well by the looks of it.

So, this could be a part requested by Tesla. Or when they placed the product they made sure to add additional features to broaden market reach?

[tom66]

A strong point is physical separation of CANBUS networks for security reasons. In the automotive world many, if not most, car brands have the keyfob data running unencrypted in a can line that is easily accessible from the ouside (what could possibly go wrong?) because those ECUs don't have many controllers to begin with, and security is always an aftertought

Yup, serious issue with car thieves now having these one-hit wonders made by people on the dark web which hack CAN bus of various models to steal them.  Interesting write up here.
https://kentindell.github.io/2023/04/03/can-injection/

Solution is to have encryption on the CAN bus for essential components like the ECU/immobiliser.

[JPortici]

A strong point is physical separation of CANBUS networks for security reasons. In the automotive world many, if not most, car brands have the keyfob data running unencrypted in a can line that is easily accessible from the ouside (what could possibly go wrong?) because those ECUs don't have many controllers to begin with, and security is always an aftertought

Yup, serious issue with car thieves now having these one-hit wonders made by people on the dark web which hack CAN bus of various models to steal them.  Interesting write up here.
https://kentindell.github.io/2023/04/03/can-injection/

Solution is to have encryption on the CAN bus for essential components like the ECU/immobiliser.

pro-tip: if you see a suspicious looking guy with a JBL active speaker, well that's not a JBL active speaker. A guy that works with us managed to find the link to buy those, apparently it's googlable if you know the right keywords
this site is probably bait: https://www.lockstart.org/details/details-jbl
as the real ones were with TLD from certain eastern europe countries but they were really advertising as "UNLOCKING + EMERGENCY START" 

[daqq]

Thanks, understood. Makes sense to have the buses separate 

[mac.6]

31x31 0.8mm BGA?
Wonder who is the lead customer (because these beasts are always taylored for a customer).

[Dr. Frank]

Well, I'm working in Automotive R&D, have not yet seen that µC. That's definitely an Automotive component, as it's AEC-Q100 qualified. Its "2 Applications" description is a bit odd, though.

Such components are intended for a central unit, which basically might connect to the about 60 other control units in a passenger car.

Usually you have several separate types of CAN interfaces in a car, like steering-, comfort-, safety- / security-,  information-, entertainment- busses, which can be low-speed or high speed CAN.
 
Nowadays, Ethernet, MOST / optical busses complement CAN for high speed infotainment purposes.

This chip has additional graphic cores inside, so it's probably intended as a central infotainment processor, like inside a Central Display.

Customers are the usual Tier1 manufacturer like Bosch, Denso, Continental, or OEM TESLA.

Frank
 

[Psi]

There's a trend to move away from having 100's of smaller MCU modules spread around a car to having 1 or 2 bigger systems that control everything.

[PwrElectronics]

My current project at work is using a Infineon Aurix with 8 CAN-FD interfaces.  292 pin BGA.  On the board, there are 3 of them used on one version.  One of the CAN's is tied to the main machine controller bus, but the others are intended for sub-bus for things that get their orders from "middle management" as opposed to the main controller.

One thing to keep in mind (not sure if true of this TI part) is many of the pins on the micro can be configured for different things.  So, for that Aurix you could use all 8 CANs but would lose some other I/O in the process.  The other thing to keep in mind is when it comes to the board layout, it may be easier to route depending on which pins you select for your CAN of the available choices.

[chickenHeadKnob]

There is also the possibility that some of specs are tailored for at customer request.
At this scale of chips using the spare gates to fit some more CAN busses doesn't have that much of an impact.

I think it is specifically targeted as main controller for autonomous vehicles, with the two realtime controllers, camera feeds, audio io, display controller, PCIE lanes (for the neural network chip), ethernet and ecc ddr4. I mean, they sell it as TDA4VH-Q1 as well by the looks of it.

So, this could be a part requested by Tesla. Or when they placed the product they made sure to add additional features to broaden market reach?

Exactly this. When ever you see a really weird part it almost certainly was specc'd by a high volume customer, and then the semi-manufacturer decides to make a general release. I remember back in the mid 80-s when anti-lock brakes were coming into vogue then motorola issued that 68000 series part with the very fancy and complicated 4 channel TPU. I was told the whole intel 8096 series and follow ons were prompted  by Robert Bosch contracting a better ECU processor from intel.

New BEV cars have scores of mcu's in the battery management system. a dual CAN bus on the main processor would be through-put saturated.