Sure refurbish the processor module and replace the eMMC at a proper rework station but replacing the whole media unit esp when the failed part is on a removable board just seems like a huge waste to me esp when the customer is footing the bill.
In my experience, removable SD wouldn’t hold up to an application like this.
If it were held in and held against some robust contact pins it likely would be fine, even under automotive conditions. There are SIM card slots that are used on heavy vehicle computers and controllers (the ones that run the engine and so on) that are robust enough to hold a SIM card in place for years of excessive daily vibration.
I suspect that designing reliable spring contacts isn't as easy as it seems...
The trade-offs for a high vibration environment are very different.
You are not wrong that a soldered eMMC (compared to a eMMC held by a ZIF or a removable SD Card) would be more mechanically stable and potentially faster (A fast uhs-ii sd card compared to a cheap slower eMMC for example), just saying that Tesla already use removable SD cards in production cars today (which doesn't brick your media unit if it dies or dislodges).
Storing the OS on a removable SD Card is (imo) a bad idea, but if they are really interested in these logs they already have a easier to replace removable storage "drive" they could write the logs to instead of burning up the eMMC.
Replace the board, then you can refurbish it in a specialized lab
Same reason why genius bars won't replace components, just boards
On the newer versions of the Media Unit instead of disabling writing the syslog to the eMMC they added more storage to the processor boards which isn't going to solve the problem just push it from appearing to further down the road.
To say that the choice to use eMMC is the fault is to say that ANY use of flash storage is a faulty decision.
If there is sufficient space on the drive to mitigate write amplification, then the media will last a very long time indeed. Longer than the drivetrain, certainly.
The problem is that Tesla doesn't delete logs it has collected from the cars. All Tesla vehicles (as I understand it) upload their logs more or less continually. Currently, the firmware on the cars doesn't delete those logs once they've been uploaded, presumably so that they're available when it is time to service the vehicle.
The disk filling up due to bad log file hygiene, coupled with the fact that the storage is flash-based, together cause the issue.
A simple firmware update to the eMMC chip could make it use a wear levelling algorithm that really does spread writes evenly over the whole chip. Current implementations simply use a sector till it goes bad and then using a spare till that goes bad, etc. After 1% or so go bad, the whole device rejects writes entirely.
With a decent implementation (like ssd's mostly do), you will never be able to kill a card in a lifetime.
eMMC flash is typically a ball-grid array part and it's soldered down hard to the motherboard of the processor. Almost every mobile device has the same setup and it's tricky and expensive to rework this chip.
You need to carefully desolder it off the board, reball the part with the correct masking stencil, then reapply it without torching the rest of the board, missing a solder connection, or zapping any other component with static electricity.
You can see a similar process done on an iPhone here: https://www.youtube.com/watch?v=y4M9uAZlbK4
You can understand why Tesla would rather replace/refurbish the board than try to field-repair it.