What you're asking for is a fast, cheap, and persistent memory.

It doesn't matter in the slightest whether it's a memristor or some other tech.

And we can approximate this right now with a capacitor to dump out RAM upon power loss. The reason you can't get the experience you want is software, not hardware.

This is theoretically around the corner; Intel and Micron created a memory technology called 3D XPoint that is persistent and is supposed to have latency close to that of DRAM.

Intel just started shipping DIMM modules, though apparently CPU support still hasn't arrived yet: https://www.anandtech.com/show/12828/intel-launches-optane-d... https://www.theregister.co.uk/2018/08/10/optane_dimm_ceremon...

You've got to be able to reboot for when program bugs leave your memory in an invalid state.

Rebooting does not help when your files are in an invalid state. (This may be relevant to software environments which persist the state of the entire "world.")

Any reason this can't be done with flash? I wonder what the speed of a flash-RAM computer would be. I'm sure it'd be slow but it'd be interesting. You'd still have to dump and restore the CPU registers of course.

I think the main issues are that it would be crazy slow, and flash isn't really random access for write— you still have to blank whole pages at a time, so you'd definitely still want a RAM caching layer in front of the flash, even if it was being synced on a second by second basis.

Plus the whole wear levelling thing would have a huge impact on lifespan.

Tiny microcontrollers would be the first ones to benefit from an all-flash architecture, but they all still ship with both flash and RAM onboard.

NAND flash isn't random access even for reading, but generally these limitations are not inherent in the technology, but are motivated by reducing the die area per bit of storage (and thus cost). Purely random access "Flash" is typically called EEPROM and readily available with SRAM-compatible interface (and costs several times more than SRAM).

In the microcontroller space there is for example TI's MSP430FR series which has unified non-volatile storage based on what could be described as core-memory on chip.

Such things were possible with core memory, bubble memory, and FRAM. None are really fast enough by today's standards, though you can get some microcontrollers and battery-backed SRAM replacements made with FRAM.

May end up happening with persistent memory--which may or may not replace current DRAM depending upon who you ask.

We had that: it's called core memory.