HP Chip Breakthrough: CPUs Could Store And Process Data - HotHardware
HP Chip Breakthrough: CPUs Could Store And Process Data

HP Chip Breakthrough: CPUs Could Store And Process Data

Currently, transistors can be produced as small as 22nm. Most are produced at 45nm or larger. But even the most optimistic engineers admit that this squishing process that has enabled the CPU world to follow Moore's Law (double in power every 2 years) won't work forever. Eventually, a new type of processor building must occur if we expect our chips to continue to get faster, and HP is working hard to discover that new route before we hit a wall using the current technique.

According to new research published by the company, new memristors could be that missing link. These tiny devices were actually thought and proposed around 40 years ago, but they just started fabrication in 2008. Now, HP has found that they can be used to not only crunch data like a typical CPU, but store information like RAM. Basically, CPUs build with these new memristors could be much more efficient than using the CPU + memory tandem that we rely on today.



HP says this discovery represents the fourth basic circuit element in electrical engineering, and it can perform logic, enabling computation to one day be performed in chips where data is stored, rather than on a specialized central processing unit. There's still no definite time frame for which HP intends to ship these, and by our estimation, there's still a lot of traditional CPUs to be built before we hop on an entirely different bandwagon. But just look at the GPU market. Ten years ago, the GPU only handled a very small amount of tasks, and today, these cards handle very advanced computations on their own with no real help from the CPU. Could RAM one day become unimportant as it's integrated into the CPU? We, and our benchmarks, are eager to find out.


Highlights

  • The latest findings about the memristor are detailed in a paper published this week in the journal “Nature” by six researchers at HP’s Information and Quantum Systems Lab, led by R. Stanley Williams. These developments follow the HP Labs team’s first demonstration of the existence of the memristor in 2008.
  • HP has created development-ready architectures for memory chips using memristors and believes it is possible that devices incorporating the element could come to market within the next few years.
  • HP researchers also have designed a new architecture within which multiple layers of memristor memory can be stacked on top of each other in a single chip. In five years, such chips could be used to create handheld devices that offer ten times greater embedded memory than exists today or to power supercomputers that allow work like movie rendering and genomic research to be done dramatically faster than Moore’s Law suggests is possible.
  • Eventually, memristor-based processors might replace the silicon in the smart display screens found in e-readers and could one day even become the successors to silicon on a larger scale.

Memristor chips and energy use

  • Memristors require less energy to operate and are faster than present solid-state storage technologies such as flash memory, and they can store at least twice as much data in the same area.
  • Memristors are virtually immune from radiation, which can disrupt transistor-based technologies – making them an attractive way to enable ever smaller but ever more powerful devices.
  • Because they do not “forget,” memristors can enable computers that turn on and off like a light switch.


0
+ -

So how am I going to upgrade my RAM??? Unless these things are so great that I don't need to. Or maybe they can still keep traditional RAM that we are using today but the RAM that is built into the cpu is more like a replacement or addition to the cache.

0
+ -

This is interesting but am I going to be concerned on how much memory it has when I'm looking for one?

Maybe this will surpass Moore's Law but I don't know how many people are ready to switch to HP's new chip.

0
+ -

"enabling computation to one day be performed in chips where data is stored, rather than on a specialized central processing unit."

That's so cool, it's mind-boggling. Your memory becomes a separate CPU...or rather the memory could one day make the CPU redundant. Any reason for Intel to worry?

0
+ -

I read this and get a vision of on dai Ram, CPU and GPU. Shrink down a power supply and use virtual storage and today's netbook could be tomorrow's wrist watch.

0
+ -

These tiny devices were actually thought and proposed around 40 years ago, but they just started fabrication in 2008.

This statement disturbs me. You would have think engineers would have looked back at this idea earlier to make the CPU more efficient rather than just focusing on clock speed and a number of cores....

0
+ -

The reason they haven't looked into this, and a "feature" not cited in the article, is that reading/writing to memristors is slower than hell.

Don't expect any revolutions.

0
+ -

3vi1:
a "feature" not cited in the article, is that reading/writing to memristors is slower than hell

Normal 0 false false false MicrosoftInternetExplorer4

 

Mentioned in the article: "Memristors require less energy to operate and are faster than present solid-state storage technologies such as flash memory, and they can store at least twice as much data in the same area."

More info here,.....

I think that it's something that they'll have to play around with before they realize the full potential of the technology. It could be the next great thing, but maybe not. Time will tell.

Pretty Picture of one here,....It looks a little like sliced cheese.

0
+ -

I'm curious how you delineate a memresistor between memory and computation, or can it do both?

I read (at dailytech) that these could be manufactured in the same bulk processes as silicon is made on. If that actually pans out this'll be an interesting change of hands.

0
+ -

There were many things that were developed over the past forty years that could have been made in to reality by now.

Everything just like computers, they wont give you the full version. not when they can break it down and sell you bits at a time every two years. Just to keep from having to come up with something new. That's where the term, never put all your eggs in one basket comes from. If you knew how many they had, they couldn't charge you as much for exclusivity. If they had alien technology which is the final version, and they gave it to everyone at a reasonable price. Then where would the jobs and profit be in that.

looks like someday we will have just a big block of silicon to handle everything. Although last time I heard it was going to be a crystal block with the lasers as interface devices.

Remember Moore's Law is a business model, and a term used to show production cost to profit over a period of years. it does not reflect the actual science. We were supposed to be on Mars by the 1980's, yet thanks to economics they didn't see the profit.

They started the scam back in the 70's when the 5" Floppy had a huge capacity of 360KB. Then in the 80's we get the large capacity of the 3.5" with 1.44MB! Wow, can you imagine what you can do with that today :P

If they could produce these Memristors, then it would totally go against Moore's Law. SO, we need to tell the FTD to give us the final version :)

0
+ -

I wondered that myself as Intel is on a Tick Tock road map and die space is shrinking rapidly. Something will have to change after we hit single digit nm, you can only shrink so much. We should reach this point in 10 years IMO. Great Article BTW

0
+ -

I think the most frustrating thing about this is the fact that these engineers know that the technology is out there, but for some reason they just can't find a way to get there. I don't know how much Moore's law really has to do with it, but I bet the whole money and financing thing really plays a large role. Regardless, technology increases at a steadily exponential rate, because we will never resort to an old way of doing something. That's what engineering is all about, revolutionizing the way we live out life, by making it easier. The very first example would be the wheel. 

This type of technology will be known as innovation. But we aren't there yet and that really concerns me, considering that the tech came into vision 40 years ago. I understand how RAM works now, but I can't really see how it would work inside these processors. It turns me to believe that a new motherboard would have to be made as well to accommodate. 

Another thing I was concerned about, which was mentioned before, was how users like us would be able to add memory to our existing memory. Right now everything just seems to work. Maybe that's another reason why the tech isn't out yet, engineers are having a hard time determining how to create this new CPU with existing computer layouts. 

However, thinking back to that article, where HP has created a new type of memory layout, I think it's possible. With the application of being able to stack memresisters, I can see how we would be able to upgrade RAM. But would that work the same way with the memory being integrated with the CPU? 

I greatly understand the potential for this. The more memory something has, the more responsive and intelligent it will become. The problem is getting there. Everything costs money. 

But I think the best possible advantage to having this new technology will be the sheer speed of the computers. I know a friend that has a custom-built over-clocked computer, that boots up in like 4 seconds. That's incredible. But HP is saying that this can turn on like the flick of a light switch? Phenomenal. 

0
+ -

The reason this sort of device has only recently become feasible is that it only works well at the nanoscale. Now that wafer lithography is down into the 20s of nanometers, these devices are getting more interest.

I think a lot of the posts here are getting ahead of the actual technology. Very little has been actually demonstrated. Most of it is mathematical models and speculation.

Even Dr. Stanley Williams, who is the research manager for the technology, only mentions the rather modest goal of having a flash-memory alternative in 3 years at only about twice the density of flash memory.

My guess is that the biggest issue for the technology is durability. HP has demonstrated memristors in titanium dioxide thin films on silicon. But the way these work is that the atoms in the film actually change configuration (i.e., move) when a voltage is applied. This is different than most solid state electronics where only charge carriers move, not atoms. To use these in something like flash memory, they can probably get away with a lifetime of around 10 thousand switches. But to use them to replace RAM, they would need to survive billions of switches. I suspect that is why Dr. Williams is only talking about a timeframe for a flash memory replacement, and not mentioning when he thinks memristors could replace RAM.

0
+ -

Even though they might have known this type of  technology could exist, it may not have been feasible at the time.  It's one thing for an engineer to develop something, but another thing for management to go with it. Cost is a huge factor when it comes to commercializing a product.   Of course if there are any problems with the research, then managers may also have delayed it - like the reading and writing times.

Hopefully they'll give it enough memory where I don't have to worry about upgrading the CPU to up my memory space :P

0
+ -

That would so suck if you had to do that (Get a new CPU if you want more memory :D) We will have to see how this goes. Or if this will happen :)

0
+ -

Looks like a great way for processor production to go. I just hope that if it is the only memory in the system, that it's more than enough for anything you may need to do, or there is an easy way to expand the amount without the need to buy a new chip.

0
+ -

They'll prolly do like Dell (when you try to build a comp on their site) and only give you certain options and never what you really want, meaning certain clock speed with certain amounts of memory, unless,like someone said, they give you more than enough memory to do all you need.

we'll see

0
+ -

ChevySSbowtie:
They'll prolly do like Dell

No, hopefully they'll develop it and then license the technology to other manufacturers so we can choose our own parts and build our own stuff.

0
+ -

License or sell it to AMD or Intel. They'll do something with it and then we can all benefit.

 

Either that, or they will keep it as a specialized solution for certain platforms. That way, they don't have to worry too much about the problem concerning the varying amounts of memory.

Login or Register to Comment
Post a Comment
Username:   Password: