The HP Way

The Americans always do the right thing after they’ve tried everything else first. So too HP.

After years of doing crazy stuff like paying $25 billion for Compaq and $12 billion for Autonomy, HP is taking a new tack – a technology play.

Three quarters of HP R&D staff are said to developing a memristor and photonics based computer which will raise computing speeds and data processing capabilities dramatically.

HP has a prototype computer where the memory is made of memristors with a photonics connection to the processor which transfers data at 6TB/sec.

“This will enable us to deal with massive, massive datasets – ingest them, store them, and manipulate them, and do this at orders of magnitude less energy per bits, per compute,” says HP CTO  Martin Fink.

‘The Machine’, as HP calls it, can scale to 160 peta-bytes with each individual byte capable of being access in 250ns.

Memristors will allow the 100TB memory cellphone, says HP.

Unfortunately all this will be an end of decade, sort of event and, even more unfortunately, HP has been singularly inaccurate with its memristor forecasts. They were supposed to hit markets in 2011.



  1. Thanks Keith, that’s another piece of the puzzle. It’s odd how companies say they’re so close to volume production and after beating the drum about it suddenly go quiet.

  2. That’s right SilverMan, handy for keeping your shopping list.

  3. Having worked on so-called ‘memristors’ using amorphous silicon in the 1980’s when funded by the deep, deep pockets of a certain company who I shall not name (but they had a bit of trouble recently with an oil well…), I can tell you that they work in theory. Which is why so many research labs keep on about them.

    The problem is that they don’t work in practice, as they rely on a ‘forming’ step which can be described a bit like ECT – a zap that might give them memory, but might not.

    After spending millions trying to get them produced reliably, we gave up. I suspect others are having the same problem.

  4. Somebody at HP is trying to pitch a 100TB memory cellphone?

  5. Not all types of MRAM have infinite write endurance but I agree if we can get something like STT-RAM to scale and still work properly then it’s what’s needed. However I think most people (but not HP it seems) see memristers as a Flash replacement where a good controller can work around the wear-out mechanism.

    I believe the reason memristers still haven’t reached the market is they use exotic mixes of elements and so suffer from huge variability effects when shrunk such that the movement of a single Gallium or other low percentage of the composition atom in position can affect the device performance significantly.

  6. Thanks Fabrice, that may be the explanation why memristors are taking so long to get on the market.

  7. Memristors have a limited write cycle endurance. I personnaly don’t believe that an universal memory with a write endurance limitation can succeed.

    DRAMs have unlimited write endurance, so has MRAM. I would better bet on MRAM (wich has actual products being sold).

    Maybe MEMRISTOR memory will displace FLASH, but it wont succed as universal memory.

    (wear leveling management is a mess, main memory cannot be messed up).

    That my two cents

Leave a Reply

Your email address will not be published. Required fields are marked *