That challenge against flash (not a very popular naming convention these days, it seems) was thrown down by HP Labs Senior Fellow Stan Williams, who posits that the memristor is "an universal memory that over a sufficient amount of time will replace flash, DRAM, magnetic hard disks, and possibly even SRAM." But onto the immediate, albeit aspirational goal (i.e. not a commitment, which he stressed on multiple occasions): Williams hopes to see the transistors in consumer products by this time 2013, for approximately the price of what flash memory will be selling for at the time but with "at least twice the bit capacity." He also claims a much smaller power requirement of "at least a factor of 10" and an even faster operation speed, in addition to previously-discussed advantages like read / write endurance.
With Hynix on board, the goal is to make these "drop-in replacements" for flash memory, whereby the same protocols and even the same connectors will work just fine. For HP, however, Williams says there'll be an initial competitive advantage for the company due to its comfort level with memristors' unique properties, but that other companies will be encouraged to license the technology and experiment with new possibilities in hardware design. Williams wouldn't give any specific product examples where we might initially see the memristor, except to repeat that it'll be anywhere and everywhere flash memory is. Fighting words, indeed. We normally don't get excited about minute hardware components -- not often, at least -- but we gotta say, the seeds of the future look mighty interesting. Can't wait to see what germinates. Highlights from our talk with Williams after the break.
HP Labs' Memristor MilestonesSee all photos
On the timeline
"For the most part within HP we are looking more to differentiate HP products. What we want to do is utilize memristor technology in our products. Within HP, you understand every product we sell has some form of memory in it, whether it's a printer, PC, or enterprise computing system, and of course our new phones. They've all got memory, and so the memristor is a new form of memory which has many of the favorable characteristics of both DRAM and flash, for instance. In fact, we believe the memristor is an universal memory that over a sufficient amount of time will replace flash, DRAM, magnetic hard disk and possibly even SRAM, so we think this is really a big deal.
"Our [aspirational goal] is approximately the price of what flash memory will be selling for in three years' time, but we want to have at least twice the bit capacity of any flash memory. There are a couple things we know in advance that we can do. One is power requirements for this memristor device are going to be much smaller than the power requirements for flash, I'm talking about at least a factor of 10. And so for appliances that require battery, this is gonna greatly improve [them]. Speed of operation is gonna be much better, at least a factor of 10, maybe 100. A lot of issues with respect to tech -- cameras, for instance."
On memristor's advantage over flash
"Based on lab experiments on speed and energy consumptions, these memristors are much faster and much lower energy natively than a flash cell. One of our advantages is that we do understand the theory at a level well beyond what other people... [as] we have been making them now for several years.
"Endurance. The number of times you can write and erase data in a memristor is much larger than flash memory, and that's big problem for flash right now as people scale. Those bits are becoming more and more unreliable. You can only read and erase them a small number of times. It requires a lot of extra circuitry and decision making. We can eliminate a lot of that. Again, we believe we're gonna be able to do a much better price point.
"One approach that we will be taking with Hynix is that this thing will look just like what you're used to. There will be products that will be made that'll just be memory sticks or memory cards and will have all the same protocols, everything else will be the same as you're used to. It's just, they won't require as much energy and you'll be able to drive them faster. For the simplest products, we intend to make these drop-in replacements for flash memory. For HP internal applications, what we'll be doing is essentially using different types of wiring protocols because these devices are faster, because they are have different properties.
"Within our products, we're gonna be taking advantage of some of the special properties by redesigning some of the interfaces inside of our products. that's gonna be one of our advantages, our ability to redesign on an early basis. They have the ability to be addressed randomly like DRAM but they're nonvolatile. You can have a laptop computer or a tablet, whatever you want to think about that you can turn on and off like a lightbulb. It's very frustrating right now, that on and off lag. It sits there and just says 'wait, wait, wait.' Transferring everything from DRAM to the hard disk, it takes a long time to write all that, and when you boot up, you have the exact opposite issue."
Other companies capitalizing on the redesign
"They'll figure it out, yeah. We're not trying to get into the memory business here, and also any tech like this has to be standard, has to be used. Our goal is to eventually have as many manufacturers making this memory as possible so they're all competing on price and performance. We understand it's gonna propagate, our goal is to be step ahead of competition. That is literally what we see as the advantage, being the first."
"We would see that it would be a flash competitor. First products will be competitors to flash [in all the same products].
"What we've done is made simple memory prototypes in our lab. Using technology available to us, we've made prototype chips with limited amounts of memory, but just to prove out the fact that what we have is compatible with CMOS fabs because actually the tech itself is not a silicon tech but it's something integrates with silicon and is the materials and processes are completely compatible with CMOS. That's why we can go to an established foundry like Hynix."
"Our goal with Hynix is, we will demonstrate a technology that everyone else is going to want to have, and once you get to that point, other people are going to want to license and bring it to market themselves. And so we will have IP that will be essentially necessary and licensable, and you know in fact we're gonna want to do that to enable as many other companies as are willing to do this. Our goal is for all memory companies to be making products out of [memristors] at some point. We're gonna be eager to license to other companies this techonlogy in order to create, as I said, an environment where multiple companies are competing with one another... Hynix is essentially the first who wants to get on board. We're working with them first, and their technology is fantastic. Great fabs, great manufacturing capabilities, great engineering."
On other partners beyond Hynix
"Let's just say we're in lots of discussion and when we're ready to announce, we'll announce."