Regarding deterministic latency, the Time-Sensitive Networking (TSN) [1] set of IEEE standards address this. The IEEE P802.1DG project [2] in particular defines a TSN profile for automotive.

[1] https://en.wikipedia.org/wiki/Time-Sensitive_Networking [2] https://1.ieee802.org/tsn/802-1dg/

> The great thing about PoE standards is there's so many to choose from.

There is nothing to choose there. They're supersets of each other with increasing power budget, each including the previous revision as lower power classes.

(and PoDL is not PoE; PoDL is for automotive ethernet, which has nothing to do with the RJ45 you see everywhere on consumer/business IT. You never choose between PoDL and PoE, the choice is made when you decided between Xbase-T vs. Xbase-T1.)

> ethernet introduces a degree of non-determinism with respect to time in the link layer

That's what TSN is for, cf. sibling comment.

> plus increased bringup times

That's generally an IP problem, not Ethernet.

> a potentially more costly core switching fabric

I guess it depends on your application; a TSN-capable 8-port automotive ethernet switch is $10: https://www.digikey.com/en/products/detail/microchip-technol... (non-TSN non-automotive is cheaper…)

> heterogeneous infrastructure costs

Automotive ethernet is supposed to replace CAN and the various heterogeneous higher-speed interfaces that were added to deal with higher bandwidth requirements with a homogeneous ethernet world :)

(Source: I work on some software remotely related to some of this shit. Which does mean I'm biased since I don't know whether it will in fact proliferate, I just make it work ;)

Sure. It's also not like there are many ready made devices that are already available for PoE that could be useful in automotive industry.

I guess my main thought was that going to 48V and in the world of low power LED lights and such, combining power and comms into same wires/cables is something that might be appealing.

Automotive ethernet over copper physical layer is completely different than commercial/data center ethernet. It is defined in 802.3ch for multi gigabit (2.5, 5 and 10 Gbps). It uses single shielded twisted pair cables up to 15m long. Cables and connectors are automotive grade.

The power over data line (PoDL, automotive Ethernet equivalent of PoE) is defined by a separate IEEE protocol and its critical specifications like EM emissions tests, ESD tests etc are supplemented by documents created by a consortium of car, electrinics, comnector, cablr and semiconductor producors called OPEN Alliance: https://opensig.org/ . There are parts available for PoDL. The voltage levels from 6V to 60V with 6V increments are supported.

Source: I design both data center and automotive Ethernet chips.

Historically and increasingly, automotive grade electronics are a separate genre to general electronics for reasons of safety.

It's less of a safety but more of a production and environmental constraints issue. Cat5 cables are too heavy for cars. Their connectors aren't made for vibration. They have a lot of emissions which is a problem for the mission critical parts (Ethernet Phy often isn't mission critical). They are expensive, which is actually the biggest reason they aren't good for automotive.

Automotive reliability is only an issue for your ABS sensor, airbag sensor etc. but these are a minority compared to what's in modern cars these days. Real driver is cost, compactness (cost) and harsh environment (temperature and vobrations) and EM emissions. It mainly adds to qualification time, but actual semiconductor design cycle isn't that long. That being said, data center stuff is also notoriously slow to qual.

Ethernet is already used to transmit radio signal over fiber which requires extremely tight variation in latency (probably more strict than automotive).