Stanford University Designs New CNFET 3D Chip: RAM and CPU Combined

09Jul17

NOTE: For the less technical reader, some explanations are coming below. Persist!

For the past ten years or so, I’ve been thinking that RAM (a computer’s short term memory) and storage (what you think of as drive space, or “long term” memory) have to merge into one system. It’s very nearly happened in phones, tablets and modern laptops with the use of flash “NVRAM” or solid state disks (SSDs). In fact, a few years back I saw some strong support of this idea when HP revealed a new compute system they called “The Machine” and HP Discover 2015. Think of The Machine as a big server rack where each shelf is populated by multiple processors and nothing but RAM (instead of hard drives or solid state drives) to serve as both long term and short term memory. All of that RAM had interconnections between each shelf using very high speed fiber. In short, the memory lanes were slated to burst off the motherboard and into the interconnects.

HP’s design was predicated on the notion of a memristor with performance equal to standard RAM but with the ability to store data even when the power is off. The primary benefit was to be: similar cost to compared to standard RAM in spite of these new features. Unfortunately, the technology wasn’t quite ready in 2015. Many pessimists probably see this as a failure no real solution, where in reality, it was a setback. I think HP’s proposal is right (partially because it matches my notion of RAM and storage becoming one!). A computer that is no longer hampered by the distinction between storage and RAM would be able to do some very fast processing indeed. This distinction has always been a necessary evil from the days when disk was far less expensive than RAM as a long term memory solution.

Today I read about a bold new chip design from Stanford using carbon nanotubule field effect transistors (CNFETs). I believe this is one more lane being added to the road to the RAM and storage merge. First, a simple and inaccurate explanation of CNFETs:

You hear all the time that computers are “just ones and zeros”. But what does it actually mean? Believe it or not, any computer (smartphones, tablets, laptops, servers, even cheap media players) is mostly just a box of self controlled light switches. Millions of light switches:

buzz lightyear

The ones are when these switches are turned on. The zeros are when they are turned off. Right now as you use your web browser, there are switches turning on and off billions of times per second in your compute device’s processor, memory, graphics chip, and even the screen if you are using a flat screen of some kind. All those billions of ons and offs represent numbers which, in turn, represent the state of something: which letter of the alphabet you just typed, what color a specific pixel on Buzz Lightyear’s red button should be, or the packet of data coming in from the network to let you know that an e-mail just came in.

Those switches are special switches that can be turned on and off by applying power to them. They are known as transistors. One kind of transistor that works well in digital circuits is the field effect transistor (FET). Again, this is just a special kind of switch that can be turned on or off by applying electricity to it. The CNFETs talked about in the article are just the latest generation of FETs. What makes them different is that they are using carbon nanotubules instead of traditional silicon. That means, they can be smaller and much much faster. That difference in size at the chip level translates to smaller and faster devices in our future!

Digging deeper into the article, the most exciting proposition is that RAM (which they include as data storage) will be part of the processor itself. Today, when a processor needs to access storage (RAM or disk), it needs to do so over some kind of interface. Think of that as having to take a bridge (which is sometimes crowded) to get to the grocery store. Imagine how excited you would be if a new grocery store was built at the end of your street. That’s what moving storage onto the processor does. The “bridge” is gone. Besides being smaller and faster because of the move to carbon nanotubules, these systems will also be faster simply because some barriers are being removed. It’s a win-win for users of compute devices everywhere and it also supports the idea that RAM and storage are merging into a single structure, which is as it should be. Good bye to “disk”!

Advertisements


No Responses Yet to “Stanford University Designs New CNFET 3D Chip: RAM and CPU Combined”

  1. Leave a Comment

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s


%d bloggers like this: