Why is a processor limited in its frequency?

[echoSwe]

Hello!

This is my first post in this forum :). I have been wondering about this for a long time; why, if electricity is instantaneous, does a CPU in a computer have a maximum operational frequency?

//Henke

 

[russ_watters]

Well, first, electricity is not instantaneous, electrical impulses travel somewhat slower than the speed of light. That may seem fast, but at the speed of light, the impulese can only travel about 10cm in one cycle of a 3ghz chip. That's not a big limiting factor for a chip the size of your pinkie nail, but it is in transferring data through the motherboard or worse, to a hard drive 30cm away.

Second, every time a transistor switches, it dissipates heat. This has been the primary limiting factor for a couple of years and the reason, today, that the chip companies are no longer keeping up with Moore's law.

Third, the faster the clock speed, the less the margin for error in the manufacturing - meaning higher tolerances and higher quality are required.

Fourth, and this isn't really a clock speed thing, but still relevant - the more complex computers become, the more transistors are needed in each chip and the tigher they have to be packed. That's also mostly a manufacturing issue.

 

[echoSwe]

So what you are saying is that the length the electricity has to travel is the limiting factor? Then why don't they just reach the max?
How come that more quality is needed to gain clockspeed?
What's this about the 16x200 MHz chipset that are included in one physical processor? Wouldn't that be much slower then what you are saying - that a cycle is 1/3 000 000 000 s, but 200 000 000^-1 instead?
In theory - wouldn't AMD and Intel be able to add more transistors and just attach better cooling?
So, when you overclock - then you're actually making the CPU go faster. This by changing the bus speed and multiplayer - is this the same as the company itself does - change the bus speed?
If the speed can't be raised more - what are they doing instead?
What is a transistor? A switch being either 1/0, or what?
Thanks for the answer.

 

[Anttech]

the more complex computers become, the more transistors are needed in each chip and the tigher they have to be packed. That's also mostly a manufacturing issue.

On this note... They reckon that by 2020 we won't be able to get transistors any smaller and by 2007 Chip manufactures will already be implementing some means to get more speed rather than cramming more transistors on chips!

What is a transistor:
http://www.google.com/search?hl=en&lr=&q=What+is+a+transistor&btnG=Search

If the speed can't be raised more - what are they doing instead:
They add more transistors and cram them closer so the Electrons don't need to travel so far between Transistors... Also they are looking at different materials and usings Electron spin as binary

 

[echoSwe]

Hi,
I still have some questions in my last post which I wonder about.
Anttech - searching for "what is a" is meaningless - google excludes it because it gets too many hits, but the transistor word brought some history and Wikipedea. Still I refer to my last post.
Thanks
/Henke

 

[Anttech]

Anttech - searching for "what is a" is meaningless

Nonsence!

If you do a 'what is a' search it is the same as doing a 'define:' search and you actually get very good info IMHO!

http://www.google.com/search?hl=en&lr=&q=What+is+a+geek&btnG=Search

Click on the First link you see to find out what I mean

 

[Zantra]

I don't have an electrical engineering degree, which I suspect russ has or is working on, but in laymen's terms, the problem is size. They are basically unable to shrink the pathways any further to fit more transistors. They have been looking at different technologies. I've seen stuff on holographic storage being looked at. Even organic storage. I think we're going to top about before we hit 10GHZ, and then we'll have to find a new method.

 

[incognito-41]

transistors

I believe transistors as we know them are about to reach a point in which we will not be able to shrink them any further. currently there transistors that are on the order of several atoms in thickness. i don't know much about quantum physics but I'm willing to bet that it's going to be hard to build transistors out of protons, electrons, neutrons, or even quarks. i think zantra is right as far as where we're heading with this technology.

as for searching for the definition of a transistor. try going to www.howstuffworks.com[/url] or try this search: [url]http://electronics.howstuffworks.com/diode.htm[/URL]

I'd offer to show you my old textbook on solid-state devices but I'm sure it's more than you want to know. in short, transistors do act like switches and current controls. hope this helps.

on another note. i have to seriously question someone who implies that another is a geek as defined by google, when they themselves are checking physics forums at 3:10 in the morning. just making an observation. that's what geeks are good at.

 

[chroot]

So what you are saying is that the length the electricity has to travel is the limiting factor?

The finite propagation velocity of electrical signals is one of the limiting factors, yes.

Then why don't they just reach the max?

That's generally the plan.

How come that more quality is needed to gain clockspeed?

If you make a processor out of large transistors, small defects in the raw silicon won't make much difference. If you make the processor out of small transistors, those defects become very important. The smaller the transistors, the more perfect your raw materials and process must be. Right now, Intel is selling processors made with 90 nanometer transistor channels. The -vast majority- of the finished units cannot operate at, say, 3 GHz. Some can only run at 2 GHz, for example, and are sold as such.

What's this about the 16x200 MHz chipset that are included in one physical processor? Wouldn't that be much slower then what you are saying - that a cycle is 1/3 000 000 000 s, but 200 000 000^-1 instead?

You can make a fast computer by just trying to make one huge processor run at enormous speeds. You can also make a fast computer by connecting lots of modest processors together and letting them share the computational tasks. The vast majority of supercomputers in the world work this way -- they expoit parallel or distributed processing.

In theory - wouldn't AMD and Intel be able to add more transistors and just attach better cooling?

The heat has to get out of the chip first for it to be taken away by a cooling system. The chip's packaging -- the black plastic or ceramic case you see when you look at a chip -- has a lot to do with its heat dissipation characteristics. The problem is not as simple as just "make better cooling."

So, when you overclock - then you're actually making the CPU go faster. This by changing the bus speed and multiplayer - is this the same as the company itself does - change the bus speed?

No, as has been said, the chip manufacturers use these methods to make faster processors:

1) They use newer fabrication technologies, enabling smaller transistors and smaller interconnecting wires.

2) They use newer packaging technologies to remove heat better from the chip.

3) They design better architectures to more intelligently schedule and execute instructions.

4) They exploit parallel processing, allowing a chip (or chips) to do the same operation to multiple pieces of data at the same time.

What is a transistor? A switch being either 1/0, or what?

Basically.

- Warren

 

[dduardo]

There are a bunch of reason why the frequency is limited. Here are a couple more technical reasons:

1. Parasitic capacitance: At high frequencies the internal capacitances of the transistor shorts, causing the tranistor to cutoff (aka stop working)
2. Transitor Dimensions: When you strink the size of the transitor the delays decrease but then you run into problems with electrons tunnelling through the polysilicon gate into the body of the tranistor and vice versa.

 

[echoSwe]

The finite propagation velocity of electrical signals is one of the limiting factors, yes.

But is it THE limiting factor?

That's generally the plan.

Sorry, shouldn't have asked this quiestion since it's exactly that question the whole discussion is about.

If you make a processor out of large transistors, small defects in the raw silicon won't make much difference. If you make the processor out of small transistors, those defects become very important. The smaller the transistors, the more perfect your raw materials and process must be. Right now, Intel is selling processors made with 90 nanometer transistor channels. The -vast majority- of the finished units cannot operate at, say, 3 GHz. Some can only run at 2 GHz, for example, and are sold as such.

How come? More detailed?

You can make a fast computer by just trying to make one huge processor run at enormous speeds. You can also make a fast computer by connecting lots of modest processors together and letting them share the computational tasks. The vast majority of supercomputers in the world work this way -- they expoit parallel or distributed processing.

No you can't. You can't just make a big CPU... Explain yourself. And it's not the samt. The new intels work with hyperthreading, making the OS think there are two processors, or two parts of it atleast. This is VERY different from how the super computer deal with load. In a supercomputer many nodes are connected, forming a cluster, and a mainframe so to speak hands out small portions of calculations to work on. These part are calculated by each computer/node and then sent back to the mainframe. This is different from how a single CPU works, since these supercomputers/clusters use multithreaded programs. And when a thread starts/is run it must have some primary memory and a cache file to work in. This needs to be done before the CPU can start processing the thread, and in one computer that's not done in the same way.

Each node has its own CPU. Are you saying my own computer works like this?

The heat has to get out of the chip first for it to be taken away by a cooling system. The chip's packaging -- the black plastic or ceramic case you see when you look at a chip -- has a lot to do with its heat dissipation characteristics. The problem is not as simple as just "make better cooling."

Then please correct me.

No, as has been said, the chip manufacturers use these methods to make faster processors:

1) They use newer fabrication technologies, enabling smaller transistors and smaller interconnecting wires.

2) They use newer packaging technologies to remove heat better from the chip.

3) They design better architectures to more intelligently schedule and execute instructions.

4) They exploit parallel processing, allowing a chip (or chips) to do the same operation to multiple pieces of data at the same time.

1) But isn't the limit reached now, when they can't make the transistors any smaller? Because if they do it leads to, what dduardo said, electrons tunneling through the polysilicon gate into the body of the tranistor? Btw - what is the gate in a transistor? Is it where the electric signal exits the transistor?
2) Isn't this what I said b4?
3) Do you have any tips on where to find more information about this?
4) And this is what will come?

Please say it technical. It's hard to ask questions if I don't know anything about it.

Basically.

And if you're to say it more advanced?

What's the difference between semi conductors and transistors? The whole, not the "basically" difference?

- Warren

Parasitic capacitance: At high frequencies the internal capacitances of the transistor shorts, causing the tranistor to cutoff (aka stop working)

Why does this happen?

Transitor Dimensions: When you strink the size of the transitor the delays decrease but then you run into problems with electrons tunnelling through the polysilicon gate into the body of the tranistor and vice versa.

Interessting. So there is a limit to how small you can make the walls of the transistor? Has this got anything to do with quantum mechanics?

I've heard some approximations about when the limit for our technology will be reached, and they were 4, 5 and 10 GHz. Now, I know this is approximations, so don't comment on these numbers, because they just give you a feel for it (...). What more can be done to increase the clock speed now? Since raised voltage can increase the frequency, will they do that, and start shipping the CPUs with water cooling, and in the end liquid nitrogen or something like this?
(And if they ship it with liquid nitrogen - doesn't the electrons freeze/the current stop?)

What makes could make the development come to a stop, at these frequences?

If you consider the spin of electrons - can you do anything from that?

 

[dduardo]

Look at this picture of a Field Effect Transistor:

http://www.pha.jhu.edu/~qiuym/qhe/MOSFET.gif

You'll notice that there is an aluminum sheet that is positively charged at the gate. Below it is a insulating material made of silcon oxide. And below the insulator there is a N+ region that is negatively charged. This is the formula for a capacitor. At high frequencies these capacitors are short, which basically connects the drain to the gate and the gate to the source. Your transitor is now a wire.

In the midband frequency range there is a tunnel of electrons created below the insulator. The electrons are able to pass from the source to the drain. By controling the voltage on the gate you control the rate at which the electrons pass from the source to the drain.

 

[echoSwe]

Look at this picture of a Field Effect Transistor:

http://www.pha.jhu.edu/~qiuym/qhe/MOSFET.gif

You'll notice that there is an aluminum sheet that is positively charged at the gate. Below it is a insulating material made of silcon oxide. And below the insulator there is a N+ region that is negatively charged. This is the formula for a capacitor. At high frequencies these capacitors are short, which basically connects the drain to the gate and the gate to the source. Your transitor is now a wire.

In the midband frequency range there is a tunnel of electrons created below the insulator. The electrons are able to pass from the source to the drain. By controling the voltage on the gate you control the rate at which the electrons pass from the source to the drain.

Thanks a lot. Nice to see a picture of it!
So at high frequencies they are short, and at mid range frequencies the electrons are tunneled below the insulator? So when an overclocker raises the voltage they are really just increasing the rate at which the electrons pass from S to D?

I have a couple of questions about the image too. It says the -green- stuff is p-Si (i.e. positively charged silicon with holes for the electrons to enter), but how then can the N+ region be negatively charged? Is it because the image shows the magnetic field and not the charge of the atoms?
Are the lines with an arrow on the diagonal the sign for the current only going one way, i.e. a semi-conductor? And what is Z?

Thanks for your reply. /Henke

 

[dduardo]

When a positive voltage is applied to the gate the electrons from the N+ region get attracted to the bottom of the insulator, thus creating a tunnel between the two N+ regions. The voltage necessary to create this tunnel is called the threshold voltage: Vtn. When there is a drain to source voltage electrons flow through the tunnel. The minimum voltage required for current to flow is Vds = Vgs - Vtn. This point is called the "pinch off voltage". If Vds is < Vgs-Vtn the transistor is in cut-off. For an NMOS (NPN Trasistor) the equation for the drain current is: Id = Kn(Vgs-Vtn)^2, where Kn is the process constant. This process constant is dependent on the capacitance of the insulator and the width and length of the gate.

Here is a good website that describes PN Junctions:

http://www.mtmi.vu.lt/pfk/funkc_dariniai/diod/

 

[Zeteg]

But is it THE limiting factor?

1) But isn't the limit reached now, when they can't make the transistors any smaller? Because if they do it leads to, what dduardo said, electrons tunneling through the polysilicon gate into the body of the tranistor? Btw - what is the gate in a transistor? Is it where the electric signal exits the transistor?

Interessting. So there is a limit to how small you can make the walls of the transistor? Has this got anything to do with quantum mechanics?

Okay okay, I don't know too much on this as of yet, but I will soon :)

From what I know, there is a limit of how small the 'chips' can be. Although it's much smaller than the technology currently available, it takes us into quantum mechanics.

Since computers use binary (1 and 0), in the process of gate circuts, you can really have a few inputs and outputs, according to either 0, or 1. However, when you get into the quantum world, a bit is no longer, a bit. It is a qubit. Instead of being a 1 or a 0, it'll be a 1 and a 0. It can be half 1, half 0, but it won't be 1/2. It can be 26.2851725% 0, and 100%-that, 1. Note that it won't be a decimal, only fractions of each value.

This makes quantum computing quite difficult. So far, the most advanced technology in quantum computing I believe, is a machine that can add, which was quite a feat in itself. :D

 

[chroot]

But is it THE limiting factor?

It might be for some designs, but not for others. There are a variety of limitations, and any of them can be the dominant limiting factor for any given design.

How come? More detailed?

Processors are built on silicon wafers, slices of pure crystalline silicon, which naturally have some amount of impurities scattered throughout them. A large transistor, which uses a large area of the wafer, would not be much affected by a very small spot of impurity. A very small transistor, however, could be totally ruined by it.

No you can't. You can't just make a big CPU... Explain yourself.

Of course you can make a very large, very high clock frequency monolithic processor -- that's the approach Intel's been taking for decades. An alternative approach is to build distributed computers, with lots of independent processors working in parallel. This is the approach of companies like Sun, IBM, and SGI.

I don't believe I ever implied that HyperThreading (which is nothing more than two CPU state machines sharing one CPU's worth of functional units) is anything like a distrbuted supercomputer. That was, after all, my point: there are two approaches currenly in vogue, monolithic and distributed.

Are you saying my own computer works like this?

No. Your personal computer is probably based on a single monolithic processor by Intel or AMD.

Then please correct me.

What I meant is this: even if you have amazing cooling systems that can remove heat at a ridiculous rate from the surface of a chip's package, that cooling system is worthless if the heat from the die itself can't make it out of the package efficiently enough.

1) But isn't the limit reached now, when they can't make the transistors any smaller? Because if they do it leads to, what dduardo said, electrons tunneling through the polysilicon gate into the body of the tranistor?

Process geometries continue to shrink. The limit has not been reached yet, but it is looming in the near future.

Btw - what is the gate in a transistor? Is it where the electric signal exits the transistor?

As has been explained already, the gate of a field-effect transistor is the electrode which is used to selectively deplete the transistor's channel, changing its conductivity.

3) Do you have any tips on where to find more information about this?

Get a college textbook on microarchitecture.

4) And this is what will come?

Distributed computing is already the norm in scientific and numerical computation. It does not benefit the desktop user much, because desktop tasks do not lend well to parallelization.

What's the difference between semi conductors and transistors? The whole, not the "basically" difference?

You're asking questions that really require several years of education to "wholly" answer, so you're going to have to deal with some basic answers.

Semiconductors are materials which can act as either insulators or conductors, depending upon the electric fields applied to them. Raw, natural silicon is an insulator. You can dope the silicon with impurities like phoshorus and boron, which result in materials with an excess or scarcity of electrons. Placing these materials together results in junctions across which electrons can flow one direction but not the other. These junctions are called diodes. You can slap two such diodes together and build a basic transistor. You can build complex devices by wiring many such transistors together. Field-effect transistors operate on a different principle, but still involve the same physical principles.

So, basically, semiconductor is the bulk, raw material out of which transistors are made. If you want more detail, you'll either have to ask more specific questions, or consult a semiconductor device physics textbook (I suggest Neaman), or pursue a graduate education in electrical engineering. I cannot explain it all to you on a forum.

Why does this happen?

Because the impedance of a capacitor is frequency dependent. The displacement current is proportional to the rate of change of the electric field. The faster the field changes, the more displacement current you see, and the lower impedance the capacitor appears to have.

Interessting. So there is a limit to how small you can make the walls of the transistor? Has this got anything to do with quantum mechanics?

There isn't a specific hard limit like a brick wall. Instead, the limit is a fuzzy area where it becomes increasingly difficult (and decreasingly economical) to make smaller structures. It does indeed have to do with quantum mechanics. Electrons have characteristic wavefunctions which are not localized. If the size of your barriers is comparable to the size of the wavefunction of the electron, the electon can spontaneously appear on the other side of supposedly impenetrable barriers. You essentially cannot trap electrons in very small spaces. Much research is being done on the behavior of electrons in "low-dimensional" spaces, like quantum dots.

I've heard some approximations about when the limit for our technology will be reached, and they were 4, 5 and 10 GHz.

Try not to think of a processor's speed in terms of its master clock cycle; that's very deceptive. Novel processor architectures, compiler architectures, heat rejection systems, packaging, and more all contribute to the computational power of a computer.

Now, I know this is approximations, so don't comment on these numbers, because they just give you a feel for it (...). What more can be done to increase the clock speed now? Since raised voltage can increase the frequency, will they do that, and start shipping the CPUs with water cooling, and in the end liquid nitrogen or something like this?

That's one approach, but it's not very practical. There are better ways to build faster computers, as we've already explained.

(And if they ship it with liquid nitrogen - doesn't the electrons freeze/the current stop?)

Electrons do not freeze. On the grand scheme of things, liquid nitrogen is really not very cold.

What makes could make the development come to a stop, at these frequences?

The same limitations we've been discussing all along.

If you consider the spin of electrons - can you do anything from that?

Yes, spintronics is a new and very promising technology that uses the spin states of electrons to encode information.

- Warren

 

[echoSwe]

Hi and thanks for the answers.

I had a look at the page dduardo gave me, and it was very good at describing it. I failed to follow when the math talk started. Too bad for me. You don't happen to know a good site teaching that kinda mathematics :)?

chroot: What's novel processor architectures and compiler architectures? Also, what's a wavefunction of an electron? Read somewhere about wavefunctions of electrons in shells of atoms. Otherwhise I'd like to thank you for all the excellent answers you've given and your time.

Zeteg: Do you have any links to sites where I can read about this? I've read some about it, but from a quite non-technical perspective, and I'd like to read some more, from a more technical one...

//Henke

 

[graphic7]

Chroot, was basically implying that a processor's frequency is not the largest factor when one evaluates a computer's computational power. Many, many factors can equate to more or less computational power of a system. Many of them, Chroot has listed.

Even if we do reach the limit of the number of transistors that are able to fit on a silicon wafer, one could spend quite a few years optimizing the processor for maximum computational power.

The manufacturing process of a processor also constitutes the processor's overall computational power, as well. One of the reasons the Alpha is (was) awesome at floating point operations and overall computational power is the fact that DEC designed the fabrication process for the processor. They meticiously tweaked each processor for maximum power, and they did a hell of a job.

 

[echoSwe]

hey graphic7, don't you think I understand that? What do you take me for, Honestly??

Could you give something about that? Anything just than a statement?

What's alpha?

Have you even red the whole thread?

 

[chroot]

hey graphic7, don't you think I understand that? What do you take me for, Honestly??

Your responses in this thread indicate that you don't understand that.

Like this one:

What's alpha?

You'd do well to try not to piss off the people who are trying to help you.

- Warren

 

[graphic7]

I apologize for insulting your intelligence. I assure you that it was not my intention. I read over the thread and "assumed" what needed to be said.

As Chroot has also said, many of the topics covered in this thread require a good amount of research. Myself and Chroot, have done this, therefore, you should, also. I don't mind answering questions, however, questions like "What is an Alpha?" can be easily Googled, and a much more descriptive answer can be found.

 

[echoSwe]

Sorry.
I understand much research has gone into the information you're giving me. Thanks

Chroot - Where in my responses do I indicate that? I beg to differ. I ask about what they are, accepting it's a list of things that limit the Frequency.

chroot wrote:
"Your responses in this thread indicate that you don't understand that."
That I don't understand that the processor's frequency is not the largest factor when one evaluates a computer's computational power? (what graphic7 replied)
I thought that was what we have been discussing all along? In this thread - so that must mean that I've been lolling all along?

graphic7 - You wrote:
"a processor's frequency is not the largest factor when one evaluates a computer's computational power. Many, many factors can equate to more or less computational power of a system. Many of them, Chroot has listed."
chroot wrote:
"Novel processor architectures, compiler architectures, heat rejection systems, packaging, and more all contribute to the computational power of a computer."
which basically means the same. This I though was obvious. I didn't try to piss anyone off or anything. So once more sorry for pissing you off chroot and graphic7.

Found this on alpha
http://www.lrz-muenchen.de/services/schulung/unterlagen/Meier-Fritsch/sld007.htm
http://www.cs.sandia.gov/cplant/

One "describing" it and one implementing alpha.

On compiler architectures:
http://www.ace.nl/compiler/cosy-technology.html

 

[sheldon]

It is all dependant on the switching speed of the transistors produced. The faster the transistors can switch the faster the cpu can process. The switching speed of transistors can be determined via many factors such as heat,resistance,design,materials, etc. make an optical cpu that would be pretty fast.
http://www.warp2search.net/modules.php?name=News&file=article&sid=15116

 

[chroot]

Transistor "switching speed" is mainly limited by gate capacitance, which in turn depends mostly upon size. Smaller transistors switch faster.

As it happens, transistor speed is less a limit of processor speed as is the speed of electrical signals propagating across the chip.

- Warren

 

[sheldon]

Transistor "switching speed" is mainly limited by gate capacitance, which in turn depends mostly upon size. Smaller transistors switch faster.

As it happens, transistor speed is less a limit of processor speed as is the speed of electrical signals propagating across the chip.

- Warren

agree, just need a switch that switches at the speed of electricity and we would be cookin:) lol

 

[hitssquad]

Found this on alpha
http://www.lrz-muenchen.de/services/schulung/unterlagen/Meier-Fritsch/sld007.htm

Wikipedia tends to be a good resource for distillations of technical items. Here is the Wikipedia page for the DEC Alpha:
http://en.wikipedia.org/wiki/DEC_Alpha