Processors smaller = better?

Alright… so everyone tries to make processors smaller. I can understand that for micro technology items. But my Desktop has PLENTY of room for a Brick sized processor if need be, but evidently speed is not limited by size.

My thoughts on this… smaller = requires less power = less heat =faster speeds.

Somebody wanna set me straight?

On an integrated circuit (computer chip), the smaller the transistors, the faster they are. That also has the side benefit of letting them pack more transistors on a single chip, to do more complex things (or have separate circuits for different functions, resulting in faster overall performance).

There are practical limits on how big a single die (chip) can be. A lot of it has to do with how common defects are in the silicon crystal structure, so a chip with a large area would have a greater chance of containing a defect, resulting in unacceptably low yields.

The great advances in processor power over the last nearly three decades is due to shrinking the size of the individual transistors.

IIRC, Smaller generally means lower power consumption and less heat output (fewer electrons are required to trigger a smaller transistor, something like that anyway).

silicon and the plastic covering aren’t exactly great thermal conductors, either. Too large of a chip will cause the inside of the chip to have a very high temperature, even with cooling attachements.

Another issue is that

larger = more distance for data to travel = slower

Although I couldn’t tell you how significant it is. The important reasoning behind smaller processors is the smaller transistors bit that CurtC described.

Another factor which recently became significant is that it takes time for a signal to travel from one end of the chip to the other, due to the speed of light limitation. Current processors spend less than a nanosecond on each clock cycle, and light only travels a foot in a nanosecond. Toss in the fact that electric signals travel at less than c, and you’re down to a few inches per clock cycle. Now, admittedly, a signal doesn’t usually need to cross the entire chip every cycle, and they’re designed to minimize the distance a typical signal needs to travel, but overall size is still significant.

Grace Hopper also used to have an example that she used that I loved. Everybody in her computer classes were given a “nanosecond”, that is a piece of wire just over 11 inches long that represents the distance that light can travel in a nanosecond.

Smaller transistors can also be packed more tightly to remove the propagation delay of electric signals within the chip.


BTW: - She’s also credited with the first literal “debugging” of a computer.

…you beat me to it. Grace is still a cool lady, tho.

In a vacuum. Electrons in a conductor, like copper, only travel about 9 inches/nanosecond. :stuck_out_tongue:

No, more like 10 cm/hr. Chronos was on target, above (not surprisingly). :slight_smile: Notice he kept using the term “signal,” not “electrons.”

Look up “electron drift speed” in any college physics book.

Sorry, I wrote electron when I meant signal.
Here are the numbers.
Speed of light:
3.0 x 10e8 m/s in vacuum - about 12"/nanosecond
2.3 x 10e8 m/s in copper - about 9"/nanosecond
2.0 x 10e8m/s in optic fiber
2.1 x 10e8m/s in Silicon
See this search,and same for Silicon.
So the maximum signal speed between chips is about 9" per nanosecond. Within a chip things go even a bit slower.

A signal actually travels a very short distance within the chip on each clock cycle. The clock is just a triggering mechanism for gates and flip-flops to change state…the faster they change, the faster the stuff gets through…distance really isn’t an issue yet. However, distance IS an issue when fetching data from RAM to the processor. As the bus speeds get higher, the ram will need to get closer and closer to the processor…that’s a very real limitation.

Basically, the smaller the chip, the faster the transistors can change state, the faster the clock can be ramped up, the more transistors you can pack into the same area, the less power and thus heat produced, etc…

Squink, you can’t really talk about the speed of light in copper. Since copper is a conductor, light doesn’t travel in it. The speed of signals in electrical cables is determined not by the conductor, but by the insulating material that separates the forward and reverse current path. In the simple example of a coaxial cable, it’s the soft stuff in the middle that determines the speed, which is typically right around 0.7c. More specifically, the signal travels at c divided by the square root of the material’s relative dielectric constant. The way to think of this is that the signal is not really the movement of electrons, but the electric field that surrounds the conductor.

Also, metal lines on an IC are typically not copper (although copper is starting to be used) - they’re aluminum.

Even disregarding the finite signal propagation speed in lines, the switching speed of a transistor is strongly dependent on size. With constant voltage over a junction; by halving the dimension the electric field strength doubles at the same time as the drift distance halves. ergo quadrupled speed.
However, the field strength mustn’t be too high, so when decreasing the sizes the voltage has to follow. That’s why current processors use around 1.5V. (About 10 years ago they used 5V.)

And since nobody’s yet spelled out explicitly why “less heat” is important, I’ll point out that if you keep everything else the same, increasing the clock speed of a chip will cause more heat to be generated, pushing you closer to overheating it. If you can reduce the number of transistors (and hence the size), you’ve probably reduced the energy consumption and thus the heat output, allowing you to crank the clock up somewhat without it overheating.

So, yes: size, speed and heat are interrelated.

Several reasons:

  1. The smaller the transistors and diodes the faster can the electrons and holes move through them. So if you make a processor smaller you can make it faster.

  2. The smaller the transistors and diodes the less heat will be generated if everything else is kept the same.

Note that [1] and [2] are related since the higher the temperature, the slower the electrons and holes move.

Theoretically, since c is a constant, yes you can talk about it in the context of copper, at least for purposes of comparison.

First, what consistently gets smaller is feature size (size of individual transistors or metal connections on a chip), not always the chips themselves. I’ll list the main reasons here, which will duplicate some good points already made:

[li]Speed performance - smaller transistors (electrical gate width gets smaller) and connections (if you can decrease capacitance) lets the chip go faster[/li][li]Power performance - As the chip gets faster, it needs more power, which is a problem in the system for both power consumption and heat dissipation. Decreasing transistor size allows chips that run at lower voltages, and may use less power[/li][li]Feature creep(*) - There are limits to the total size of the chip you can make, so shrinking the feature size lets you put more transistors on, which means you can go from a 16 bit processor to 32 bits to 64 bits, add busses, graphics features, new commands, registers, all the latest and greatest nifty features for the programmers.[/li][li]Cost - Chips are made in “wafers”, which are a pretty consistent size (I’m oversimplifying), and you get as many chips on a wafer as fit. If you keep the transistor count constant and shrink the feature size, you get more die on a wafer that costs a comparable amount to make (more oversimpifying), and your costs go down. You either sell it for less and sell more at the same profit rate, or you sell it for the same and make more per chip, or somewhere in between.[/li][li]Maximum die size - There are several things that limit the maximum size a chip can be. The major two are defects and reliability. The bigger the chip, the more likely you are to have a defect on it which makes the chip fail. This is a cost issue. For a given wafer, you make fewer chips as they get larger, and even fewer of them work (are defect-free). Reliability is an issue with getting larger silicon chips to go into a chip “package” (the package is really the only part you see when looking at most chips). For instance, one problem is that the silicon has a different thermal expansion than the package, so as the chip heats up and cools down, it does not precisely match the changes in size. This puts stress on the chip, which limits how big you can make the chip.[/li][/ul]

I’ve oversimplified all over the place above, but I hope the gist is there.
(*) I don’t know this feature creep guy personally, but I understand he used to work at Intel, but mostly works for Microsoft now.