HP wants to blow your minds with what's in its development pipeline. The Machine, it says, will introduce a new kind of system architecture that will use memristors and silicon photonics to "replace a data center's worth of equipment with a single refrigerator-sized machine." It will be able to address 160 petabytes of data in 250 nanoseconds. It will pump up to 100 terabytes of storage into a single Android phone.
Whoa! Mind blown!
I mean, you have to admit that the Machine sounds like the best thing to hit technology since John Bardeen, Walter Brattain and William Shockley dreamed up the transistor. It will change everything.
If it ever materializes.
That's a very big, fat "if." For the Machine to deliver, it will need not one technology to make it from the lab to the production line, but two very different technologies to mature into world-beaters. First are those memristors. They are a kind of data- storage technology that, its proponents promise, will have RAM-like access speeds with storage density well beyond today's best NAND flash memory.
Memristors are not merely theoretical. We know they will work. But how well will they work? That's an entirely different question. And if they do work as well as HP hopes, can they be produced in large enough quantities with sufficiently high quality that we'll be able to put them in data centers -- never mind smartphones? I'm asking because HP's plan depends on these things happening.
The company may have its reasons to crank up the hype for the Machine, and I'll get to some of those in a bit. But not everyone at HP has drunk the Kool-Aid. CTO Martin Fink, who's quarterbacking the Machine efforts, admits that memristors may not even arrive in products this decade.
But let's say we do get memristors. That will be worth a hallelujah or two. It's a great thing to be able to put an entire HD video library in your pocket and for data centers to no longer have to deal with a hodgepodge of storage technologies. But getting memristors doesn't get us the Machine, since we will still need silicon photonics -- that is, I/O and networking with light instead of electrons.
Now, silicon photonics are a good deal closer to shipping than memristors. Indeed, Intel promises that it will ship an optical connector that can transmit up to 1.6 terabits of data per second later this year. But the Machine needs much more than silicon photonics to become a reality; it needs silicon photonics to be shrunk down to fit on a motherboard.
Sign up for CIO Asia eNewsletters.