pouët.net

Aparently a revolutionary 4th basic electronic component was created by HP.

Can you find a link to the paper? :)
I'd like to take a look :P

Putting that on short words: that reduces a SRAM from 3 or 4 transistors to 1 memristor?

This measn my floppy drive will be outdated soon? :(

the smallest SRAM cells i know use 5 transistors; but memristors actually provide nonvolatile memory, so the best comparison is probably flash memory, which uses one floating gate transistor per bit (for storage at least; not sure how much control logic is involved, but that isn't counted with the memristor either).

from what i gather, memristors will be about as fast to read and write as DRAM (if not faster), without requiring any refresh (since they're nonvolatile). not sure how big one would be or how much control logic is required, but this has the potential to replace everything from cache SRAMs over DRAM memory cells on to memory cards/harddisks/dvds/whatever.

that is, if (and only if) they're reasonably easy to mass-produce, power efficient and can be packed densely enough, and all that before conventional memory technologies are there. no clue about that, but i guess neither do the inventors, yet.

oh, and note that there are other interesting alternative memory technologies that have some or all the advantages of memristors, and are already being produced. google for MRAM, FeRAM, PRAM, SONOS - all in production already, but they're nowhere close hitting a sweet enough spot to dethrone DRAM/Flash yet. another candidate is racetrack memory - there was a recent breakthrough with that too (read: they've actually managed to get it working properly in the lab).

well, we'll find out soon enough who wins :)

i can has memrory?

so this would work the same as the RAM chips on the motherboard (how are they called?)
but they don't need power to keep the data

am i correct?

if so, then there could be something like instant-hibernate, that would be cool

whynot: exactly. In the picture it looked rather simple, so it should probably be easier/cheaper to manufacture than current RAM (even flash is probably more complicated). "Unified" solid-state memory (no more ram/hard drive, no swapping, just loads of fast RAM) would be truly sweet, but I guess we'll need to wait another 10 years for this to hit the mass market...

A step closer to real Computronium.

well, DRAM is 1 transistor+1 capacitor per cell. or you use Z-RAM, which is just 1 transistor and uses the floating body effect for capacitance (about double density compared with current DRAMs; also not in mass production right now, but i think it's already been licensed by some DRAM makers, so i guess it'll appear soon). not sure how big that memristor is, but i do know that those transistors are pretty damn small by now :)

I'm not too sure about the memristor's read speed tho - according to Wikipedia it'll use DC for writing and AC for reading (so that the read access doesn't change the memristor's state). This will need to use some pretty high frequencies to rival current DRAM.

that said, it'll of course still be pretty awesome for embedded devices and instant on/off where performance isn't too critical. But I doubt it'll replace DRAM in PCs and graphics cards too soon.

From TFA :


Plus it seems to be only the "Mark 1" memristor so just wait for the commercial-grade, which will probably include major improvements. :)
Plus it must really cut the power requirements to skip the refresh part of usual RAMs. Think "more lifetime for laptop batteries".

Okay, sorry kb, I just read wikipedia :



...says Wikipedia quoting Nature. You're right about the speed issue. Let's be optimistic though. :)

10x slower than DRAM with low power consumption and nonvolatile is pretty much perfect, actually. if it can be mass-produced in suitable quantities and densities to rival harddisks, that makes a proper memory hierarchy again, without the huge performance gap that's currently between DRAM and harddisks that makes paging such a pain :)

The memristor and SRAM need no refreshes, which is a big advantage compared to DRAM, cause refreshes take time and stall/block access to memory. SRAM uses a lot of components though, the memristor only one.
About size: I'm just saying that it at least look conceptually simpler, which probably makes it easier to shrink and cheaper to produce.

It'll never catch on with a silly name like that.

"Capacitator" isn't much better.

What about "Resemory" or "Resistory" :)

or "bob"

The original article is actually linked from the register:

http://www.nature.com/nature/journal/v453/n7191/abs/nature06932.html;jsessionid=910F4308BEEB721E3DE4181E08898154

They did not invent anything, they merely showed that a concept that was demonstrated in a paper in 1971 can be used to describe several recent devices. This is just a lot of publicity hype.

er, no. the 1971 paper was a completely theoretical argument that there should be something like that. it did not provide any physical model or construction. compare to hypothesizing that it should be possible to split the atom vs. actually demonstrating working nuclear fission.

read abstract

no, you read abstract. "Although he showed that such an element has many interesting and valuable circuit properties, until now no one has presented either a useful physical model or an example of a memristor."

what I understood is: "we might now know how it should work"