HP and Hynix to produce the memristor goods by 2013
18 months to go, baby
An HP/Hynix memristor product should be here in 18 months – and PCM, MRAM and RRAM are all memristor-type technologies...
As reported by EETimes, this came out at the International Electronics Forum in Seville, in a presentation by Stan Williams, a senior HP Labs Fellow.
The report quotes Williams saying: "We have a lot of big plans for it and we're working with Hynix Semiconductor to launch a replacement for flash in the summer of 2013 and also to address the solid-state drive market."
HP will target multi-layer memristor technology at the DRAM market in 2014/2015, with the SRAM market in its sights after that.
Ongoing manufacturing trials involve running hundreds of wafers through a full-size Hynix fab. Williams said that memristor cost/bit could be an order of magnitude less than NAND, after initial costs are out of the way. We don't know if he was talking about single level cell or 2-bit or 3-bit multi-level cell (MLC) NAND though.
Apparently, as well as price, energy usage, read and write times, retention and endurance are all so much better than NAND that HP thinks the idea of memristors replacing NAND is not far-fetched at all.
It takes less than 10 nanoseconds to read a bit; endurance cycles go beyond 1012 and retention time is counted in years.
Williams held up the prospect of layering memristor memory on top of processor cores, saying: "We could offer 2GB of memory per core on the processor chip. Putting non-volatile memory on top of the logic chip will buy us twenty years of Moore's Law." To which we say: "Wow!"
They're all memristors
Williams said that phase-change memory (PCM) and metal-oxide resistive RAM are both types of memristor, meaning magneto-resistive RAM which IBM and Crocus Technology are working on, is a memristor variant.
That implies that Unity Semiconductor's CMOXe technology is also part of this wider overall memristor technology push. So too is Samsung's STT-RAM. Williams said Samsung has a larger lab working on its resistive RAM technology than HP does.
Gone in a flash
If non-volatile resistive RAM, in the shape of HP/Hynix memristor devices, is going to hit the flash market by the mid-point of 2013, then what does that say about the future of NAND-based solid state drives (SSDs) and PCIe flash cards?
At present we are undergoing a transition to 2X nanometre NAND process chips and 2-bit MLC is becoming mainstream.
If memristor technology is as good as HP is saying, then perhaps we won't see sub-20nm NAND or 3-bit MLC NAND at all. Particularly as both NAND speed and endurance worsen as we move to sub-20nm process dies and to 3-bit MLC.
Suddenly flash looks vulnerable, but that will only be the case if HP is right. Is it? Ladies and gentlemen, place your bets please! ®
" Putting non-volatile memory on top of the logic chip will buy us twenty years of Moore's Law."
Me spies a fundamental problem...
Just how are they proposing to get the heat from the processor die to the heatsink then, with another chunk of chip in the way?
Nothing new here. HP Labs is (and has always been) where HP is actually good. It's just not that sexy to Wall Street. You have all those non-productive capital and R&D costs.
Services OTOH is very sexy mainly because it has a low capital cost. It's just that HP has not figured out that it's not something you can buy into, but have to build over a long long time. Consequently they keep throwing away (downsizing) the parts of their company that actually do services well, paying over the odds for an external services outfit and then failing to understand why customers don't like either their old or their new services offereings.
Start by having a quality product, and build your services around that. Throw away your product and customers will follow and get their servies from whereever they now get their (replacement) product.
On the one hand, HP bails out of tablets and looks like they're getting out of PCs too - in order to concentrate on software and services (e.g. by spunking way over the odds for Autonomy).
and OTOH, HP cuts some serious semiconductor mustard.
Joined up? right hand meet left hand ...
This is just one of the projects that was saved from that idiot Prith's savaging of HP Labs. Stan and the lads probably through a combination of (a) being very good and (b) being close enough to the seats of power to ignore Banerjee survuved and look as though they are going to come very good.....
Non-volatile storage directly on the core?
Hmm, not sure about that one. We don't put much RAM on the core - most RAM in a system lives elsewhere, and its data gets pulled across to the cache as needed. The point of memristors is that they get you fast rewriteable non-volatile storage. But the point of a cache is that it's *NOT* a permanent store.
You can't use it instead of RAM, that's for sure. 10^12 write cycles sounds amazing - but run the numbers and you find that even if you only wrote to it at 100kHz rate, that would only give you 115 days of continuous use. So we still need the old architecture of volatile data living in RAM for processing and input/output data being written elsewhere at infrequent intervals for long-term storage.
IOW, it's a really good, fast, reliable non-volatile storage mechanism. Nothing more, nothing less. Currently Flash SSDs are a bit niche, but it seems likely that memristor NV will basically blow away Flash and hard drives for everything except really large storage application, following the same principle as tape drives being used for storage/backup when hard drives were the fastest available technology for small/medium data.
Sure, the idea of "full shutdown" will basically go away. With near-instant storage of system state for hibernation and recovery, that's all anyone will do. But this is already the case - most people will already hibernate their laptops instead of doing a full shutdown.
So it really ain't a game-changer. It might change the names on the storage offerings, but it won't change what we do with PCs, how we do it, or how the system architecture is set up.