Bubble Memory
en.wikipedia.org
en.wikipedia.org
Now, this is even true for flash memory, but this is handled at the controller level and its dynamic during its initial "formatting" at the factory, so it's completely hidden from the host unless "host" is the controller.
OTOH, bubble memory comes with a "this machine has no brain, use your own" sticker on it.
This is all transparent from the host, maybe an occasional "wait a little, I'm kinda busy" signal to the controller. Same for flash, but it's way faster and more transparent, since the real map inside the flash controller is never linear, anyway.
I don't really know why "bad sectors are still bad after 40 years" is so surprising to me, but there you go.
So they are not worn out sectors, they were bad (or not there) from the start. You can't rub them out, because there is not enough magnetic flux or material to store data at that address.
You could buy disk packs with no bad sectors back then if you wanted that ("bit error free"). They cost extra but the premium wasn't all that high. Of course they could still develop bad sectors later.
Controller based management is also true for SD cards of any form, CF cards, USB flash drives or any external drives. This makes only embedded systems and devices like phones with direct NAND/NOR management AFAICS.
https://deramp.com/downloads/mfe_archive/050-Component%20Spe...
I don't remember the exact specs but am 100% sure he mentions that it has bubble memory.
The bubble memory was doomed by the fact that it had to be made from material that was patterned, by using photolithography or similar means.
Because of this, even if its structure was simpler than that of semiconductor memories, its bit density could not be increased much over that of semiconductor memories and its fabrication technology was not significantly cheaper.
Being much slower than random-access memory, while not having much cost advantage, made it non-competitive.
In contrast, the memories that use a homogeneous recording material with bits defined by the movement of a recording head, in the form of disks or tapes, compensate their slowness by having a much lower price per bit.
Thus bubble memories were too expensive in comparison with HDDs and too slow in comparison with DRAM, so no application remained for them in a computer that used DRAM and HDDs.
Flash is also much faster than a bubble memory, because it provides random access to blocks, even if a block is read sequentially. Thus flash was positioned in a good spot in a hierarchy of memories where a compromise is made between access speed and cost per bit.
To achieve a similar speed, a bubble memory would have to store a block, e.g. 4 kB, per bubble ring and have a huge number of bubble rings (i.e. billions, to compete with the current flash memories) to which random access would have to be provided by semiconductor ICs handling the multiplexing and demultiplexing of data.
Due to mixing ferromagnetic and semiconductor technologies, it is very unlikely that such a device could be made at an acceptable price.
There are about a dozen upcoming memory technologies with better theoretical headroom waiting in the wings for DRAM and Flash scaling to slow down. In the early 10's it really looked like Flash was about to be replaced, but then Samsung managed to move Flash to the third dimension. Very soon, either DRAM does the same, or it gets replaced by something.
I suspect "growing up" with a system where the computer itself is nonvolatile (turn it off, move it, turn it on, it's doing exactly what it was doing when it was turned off) changed the way people looked at computers vs modern systems.
It's too bad intel didn't push harder on their optane stuff. I guess it's too bad (for intel) for lots of different abandoned / failed projects.
In a computer that would have only non-volatile memory it would be difficult to ensure that things that must be secret remain secret.
Various events like power failures or software bugs could prevent the erasure of the secrets when they are no longer required.
Total memory encryption, like first introduced in server CPUs by AMD, and then followed by Intel, could solve this problem, but even that depends on the CPU having internal memory that is volatile, so this allows only the reduction of the amount of volatile memory that is needed, not its complete elimination.