Proccessor form factor (shape)
Is there only one reason to make a CPU square in shape? If this square form factor could be overcome why not have a laptop CPU that is actually "BIGGER" than the desktop CPUs? Say a form that would be elongated and take up most of the long narrow side space or back space of a notebook like the PBTi. In a world where miniturization is a key goal why not go the other direction with mobile CPUs and spread out some of the heat generated by making them only as small as they need to be for a particular device. And have that CPU juxtapositioned entirely different than conventional "square" CPU horizontally mounted in deep in the bowels of the chassis where much of the heat is trapped regardless of heat sinking and fans applied to the problem? Maybe by "spreading" out the CPU and changing the form factor of CPUs more computing power could be obtained and built in heat sinks would be closer to the surface of the chassis of the unit and away from other critical components not to mention the keyboard. Also this frees up much room for killer standard sized chips for the onboard video!!!! Why go small when you don't have to. The neccessary ergonomic screen size is what dictates the size of a portable, that will only change when were all finally using heads up eye displays and wearable computers and that is way off in the future!!!
Comments
Or something like that.
In a way Apple is aldready spreading heat by using 2 CPUs.
<strong>Is there only one reason to make a CPU square in shape?<hr></blockquote></strong>
There are two good reasons. First is that silicon that's far from square has little structural strength-- it's considerably weaker in one dimension than the other (the short side is very strong, and the long side is very weak). Secondly, as has already been said, signals may propogate at close to the speed of light, but keep in mind that it's a finite speed. Getting from one side of a 1mm x 200mm chip to the other would be a long journey.
[quote]<strong>
If this square form factor could be overcome why not have a laptop CPU that is actually "BIGGER" than the desktop CPUs?<hr></blockquote></strong>
The square form factor has never been a problem for anybody. Rectangles up to a 1:5 aspect ratio are typical.
[quote]<strong>Say a form that would be elongated and take up most of the long narrow side space or back space of a notebook like the PBTi. In a world where miniturization is a key goal why not go the other direction with mobile CPUs and spread out some of the heat generated by making them only as small as they need to be for a particular device. And have that CPU juxtapositioned entirely different than conventional "square" CPU horizontally mounted in deep in the bowels of the chassis where much of the heat is trapped regardless of heat sinking and fans applied to the problem? Maybe by "spreading" out the CPU and changing the form factor of CPUs more computing power could be obtained and built in heat sinks would be closer to the surface of the chassis of the unit and away from other critical components not to mention the keyboard. Also this frees up much room for killer standard sized chips for the onboard video!!!! Why go small when you don't have to. The neccessary ergonomic screen size is what dictates the size of a portable, that will only change when were all finally using heads up eye displays and wearable computers and that is way off in the future!!!</strong><hr></blockquote>
Laptops are very densely packed. It would not make sense to have a huge CPU just to spread heat. If that were the topic, you could suggest a heat spreader (a hunk of metal permanatly attached to the silicon that's more thermodynamically efficient than a heat sink), but that's already done by some manufacturers. Alternatively, have you felt how hot those high end graphics get? Sure, you could put those in a laptop, if room isn't a factor, but you'd have no battery life.
The bottom line is this: No matter what aspect ratio you have, if you're not doing anything to decrease the power dissipation (CPU, GPU, what have you), the case will only get hotter and/or the blowers will only get louder. With both of those comes decreased battery life. If you're willing to sacrifice one of those for the other, you're in the wrong market, and you should be getting a desktop. Perhaps a cube is to your liking. Or an XServe mounted in an alternative way? Maybe an x86?
The reason I like the PowerPC is how elegant and flexible the architectures are. You just KNOW that very brilliant people are working on those. If something could be bettered by making a chip that's 3mm x 15mm, don't you think it would have been done by now? That maybe some of those brilliant enginers could have come up with the idea before an armchair adviser?
If you wanted, there's nothing to stop you making the package long and thin, but it would make no difference to heat dissipation - it would just get hot in the middle where the chip actually is.
CPU's are measured in microns, not millimeters.
<strong>You guys are aware I assume that the part of the chip you see is just a plastic package made big enough to fit reasonably-sized legs to the chip for connection to the circuit board. The actual silicon part of the CPU is the size of a pin-head and any changes to its aspect ratio would make very little difference to the size and shape of the package, and vice-versa.
If you wanted, there's nothing to stop you making the package long and thin, but it would make no difference to heat dissipation - it would just get hot in the middle where the chip actually is.
CPU's are measured in microns, not millimeters.</strong><hr></blockquote>
Heh, I have a pretty good reason to believe I'm better aquainted with what a CPU looks like than anybody else on this board and I can assure you that CPUs are much larger than pinheads. They are approximately the size of your thumbnail and are in fact measured in mm. 100mm^2 is a fairly typical die size for a CPU.
<strong>Don't listen to him! What would a simple igloo dweller like him know anyway?
Don't anger the eskimo, he knows where your dog lives
Let me make clear I know nothing about CPU design engineering.
So that said;
"blabla", you talk of wider bus paths and I assume you mean a physical "distance" increase of the data traveling through the chip. At the speed of light making a chip wider will somehow slow the flow of electrons. BALDERDASH!
"Eskimo", you obviously have some knowledge of CPU design. so then; why on God's green earth can't a lithographic scanner image anything but a "square"? Then make one that does! You smart guys could explain to us armchair CPU designers what are the hurdles one must overcome to do these kinds of things!!!! You state a CPU is "thumbnail" sized but what prevents a CPU from measuring a "pinky" height x four fingers wide??? What prevints an odd aspect ratio like this?
"Chrys Robyn", firstly, the silicon isn't "structurally" sound for anything. I assume the millions of pathways are "lithoed" onto a much stronger substrate or backing if you will (once again, I know "nada" about these things). Also, at the "speed of light" do tell how far can a single pathway be from one side of a CPU to the other? C'mon! Will you be telling me it's from here to the moon and back!!!! Someone please step up and give us some sound engineering reasoning!
Secondly "CR", how is it "NOT" a problem to have to take a "HOT" CPU whose form factor only allows it to be mounted in a portable machine in "THEE MOOOOST" inconvenient, least effective (cooling wise that is), and worst place (under the keyboard)?#@! How is that "not" a problem? Yes indeed more power will reduce battery life and make the chip hotter! Fine, but maybe make a CPU that could be moved to the outer peremiter of the machine in question so that more efficient cooling could actually take place! HMMM...thats a novel idea. You make the case by stating that "Laptops are densely packed". Maybe moving the "hottest" part of the whole package "out of the way" so to speak, the whole "rig" would "breath" alot cooler.
Third "CR", I believe "armchair" thinking is HIGHLY underrated!!! the old adage "there's nothing new under the sun" is pure bullshit! Yes, it may have been thought of but did anyone actually attempt it? (whatever "it" may be). See "CR" I'm of the opinion that while there are alot of smart minds out there, hoards of good ideas get swept under the carpet either by shear economic reasons or closed mindedness. Forthright action upon many "armchair" ideas have brought about the most innovative solutions to technological challanges; two bicycle makers discovering flight and launching the aeronautical industry comes to mind. Bicycle makers Phhhhtt, who'd a thunk it?
BTW the 100th anniversary of that momentous occasion is comming up! I hate to tell you but "craftsmanship" not raw engineering is what makes space flight possible!
"Socrates", I am not talking about the "package" around the actual chip at all. I mean change the circuit layout of the chip routing it in such a way as to spread out the "layout" the CPU into a long thin form!
I restate my question: What of the odd form factor and alternative placement of the CPU I've asked about in portable CPUs is doable/undoable and why/why not?
Will some CPU engineer please enlighten us with a reasonable "layman's" explaination of the facts and not be condescending?
<strong>
There are two good reasons. First is that silicon that's far from square has little structural strength-- it's considerably weaker in one dimension than the other (the short side is very strong, and the long side is very weak). Secondly, as has already been said, signals may propogate at close to the speed of light, but keep in mind that it's a finite speed. Getting from one side of a 1mm x 200mm chip to the other would be a long journey.</strong><hr></blockquote>
Then in the spirit of unreasonable predictions, the famed IBM, 64 bit, save-Apples-ass chip will be spherical with pins radiating outward. Just like a sea urchin.
It would maximize strength and minimize electron travel. Piece of cake ;-)
Electrons travel at a finite speed, it's close to the speed of light, but it is still finite. For this reason if you want to make your chip faster, the best way to do it is to make it smaller because then the distances between the important bits in the chip are smaller and so messages can go between them more quickly.
If you make the chip anything other than square, you will be making one side longer than it needs to be, which means signals (electrons) travelling in that direction in the chip will take longer to reach their destination and so the chip will run more slowly.
This will certainly make it cooler, but at what cost? If you wanted to make chips cooler this way you could just make them huge, or put fewer transistors in them, but that would just be like going back to 1980's technology, it would hardly be an innovation.
In a square the longest length that any path might be is diagonal, which is only 1.4 times the length of the side, so a square is very efficient in terms of distances.
With regard to internal busses, you can compensate for distance with width to some extent. A bus is a group of wires carrying information. ech wire can carry one bit of information (either a current is flowing down it, or not). If you wish to send 8 bits of information to somewhere, you can either send then one after another down one wire (serial), or you could send them all at the same time down 8 wires (parallel). The latter is faster, but an 8 wire bus takes up more space on the chip. You could partially make up for your chip being longer by having a wider bus to carry information form one end of it to the other, but then it would be wider too, so in effect you are just making your chip bigger for no good reason.
Rest assured that current processors already have the optimal space/bus width/speed layout that can be concieved by CPU engineers, so it is likely that speed critical tasks already have a wide enough bus, and so there would be no speed advantage to making it wider (you don't get any speed increase by sending 8 bits down 16 wires rather than 8). As a result, making the die longer would just be pure loss of speed.
And finally, as I stated before, though innacurately (Thanks Eskimo - I stand corrected), CPU's are not all that big, what takes up the space is interconnection, i.e. finding room for the wires to connect them to the other clever bits in your computer. A long thin CPU would not be much easier to place than a square one. If you put it in the screen or something you'd need to run hundreds of connection wires to it, and the speed issues apply to them too. Your RAM can't be too far away from the processor or you get slow RAM access. The GPU (graphics processor) can't be too far away or you get lower quality rendering as the graphics pipeline can't be fed fast enough, and people complaining about paltry 166Mhz system busses will get even more stroppy because a longer bus is a slower bus.
I applaud the armchair thinking concept, but this particular idea is going nowhere.
As for ideas that might go somewhere, there are several - see next post:
chip-on-board technology. At the moment, chips are made ina huge factory in very clean conditions and then put into relatively big plastic boxes with relatively giant metal legs sticking out. These are then shipped to people like Apple who solder them on to circuit boards.
What if the chip was connected directly to the circuit board in the factory, and then a blob of plastic was squirted onto it to protect it from the elements. This would be smaller and more efficient, taking up less room on the board and requiring less packaging which would help it to run cooler.
If you take apart most calculators you will find that they already do this, so it's not far off for CPUs.
3D chips. a square is more efficient than a rectangle, but less efficient than a cube. at the moment chips are made in layers and are essentially flat. They would be much more space efficient if they had wires that ran vertically as well as horizontally. Cooler, faster, you name it. This will undoubtedly be the future. I don't think you'll see it by January though, this may be a decade off or more.
Optical chips. electrons go at nearly the speed of light. In practice they bounce around a lot and so it is closer to 2/3 that. The same is true of light in an optical fibre actually, it goes a 2/3 that of light in free space. Light however does not have a charge (well, actually it is an electromagnetic wave, but that's too compilcated to explain right now). This means that two phtons can go right next too each other and won't affect each other at all, unlike electrons, which are a right pain in the ass. Once we get down to a few fractions less of a micron we'll find that the electrons start interferring with each other, but photons won't. Mkae chips optical instead of electrical and all kinds of new speeds will be obtainable, and they won't generate or be affected by radio frequency interference either, meaing that there is more hcoice of where to put them, and they can be put nearer the other components in the computer.
Electrical cooling. I think it was actually Einstein who came up with the idea of a Fridge with no moving parts. You can use purely electrical processes to cool stuff to below freezing. (You can even cool stuff by shining a laser on it using quantum effects.) This is much better than a fan, and it's silent. If this could be made small enough for a lap-top then it would solve the heat problem once and for all.
I've probably missed a few out.
Three-dimensionally built processors is a dead end - it would minimize the content/surface ratio and make cooling all but impossible.
Or what?
engpjp
[ 09-02-2002: Message edited by: RazzFazz ]</p>
"Eskimo", you obviously have some knowledge of CPU design. so then; why on God's green earth can't a lithographic scanner image anything but a "square"? Then make one that does!
<hr></blockquote>
The entire purpose of a scanner is to utilize the optimum portion of the lense elements in the optical column of lenses. At subwavelength resolutions the need for precise control over the tranmission of DUV radiation through the optical elements. Any unwanted abberations in the lense elements can lead to the inability to print the required features. Scanners work by moving the wafers stage at 4 times the speed of the mask stage in opposite directions. Since photomasks are made at 4 times scale this keeps the features being printed in a constant location at the center of the lens. This advancement in lithography took decades of research and billions in R&D. So to suggest we can simply whip up a new way of printing is a bit far fetched.
There are direct write vector scan tools which can write to a substrate much like you would use a pencil on a piece of paper. However the throughput of these tools leaves much to be desired and therefore they are not cost effective.
[quote]
You smart guys could explain to us armchair CPU designers what are the hurdles one must overcome to do these kinds of things!!!! You state a CPU is "thumbnail" sized but what prevents a CPU from measuring a "pinky" height x four fingers wide??? What prevints an odd aspect ratio like this?
<hr></blockquote>
Due to gravitational sag was are currently limited to producing photomasks on quartz substrates that measure 6" by 6". Due to the necessity of edge exclusion areas to reduce flare this limits the available field size that a design can be printed on. Thus while the horizontal dimension of a chip is a function of the lens, the vertical dimension is a function of your max field size on your mask.
[quote]
"Chrys Robyn", firstly, the silicon isn't "structurally" sound for anything. I assume the millions of pathways are "lithoed" onto a much stronger substrate or backing if you will (once again, I know "nada" about these things). <hr></blockquote>
There is indeed the necessity for some physical strength when we seperate the die from the wafers using dicing saws.
[quote]
Also, at the "speed of light" do tell how far can a single pathway be from one side of a CPU to the other? C'mon! Will you be telling me it's from here to the moon and back!!!! Someone please step up and give us some sound engineering reasoning!
<hr></blockquote>
Actually electrons travel very slowly, on the order of 10mph roughly. The actual signal being transmitted is the propagation of the disturbance of electrons which does take time. When you are operating at GHz speeds these signals must transmit from one side of the chip to the other in pico seconds.
<strong>10mph? <img src="graemlins/surprised.gif" border="0" alt="[Surprised]" /> </strong><hr></blockquote>
[Edit: layman's translation: e- = electron, c = speed of light in a vacum.]
Yep, I believe its called e- drift. That is how fast a specific e- would actually move through a circuit. As Eskimo said, however, the signal that MULTIPLE e-'s propagate together is on the order of c (though not really close to c, more like 1/3*c, and up to about a max of 2/3*c; still fast, though).
The classic example is the line of train cars starting. When the engine starts moving, it jolts the line of cars into motion. Once the entire line of cars is moving, each car is moving probably less than 1mph, but the last car in the line started moving only a second or so after the first car behind the engine started moving. If you calculate out the speed of the signal (cars starting to move) propagating from the engine to the last car, on a long train that could be 10's or even 100's of mph (guessing here), but at any rate, much faster that the initial 'drift' of the cars once they are all moving.
So, replace train cars with e-'s, engine with EMF (i.e. an Electric field, or Voltage differental), and you've got an electric signal.
That and a little pratice, and you too can design a CPU! <img src="graemlins/lol.gif" border="0" alt="[Laughing]" /> <img src="graemlins/lol.gif" border="0" alt="[Laughing]" />
Oh, and if no one else metioned it, if you did manage to make a long, skinny CPU die, and the litho didn't distort, and somehow avoided all the other potential problems, you would also have a hell of a time packaging and wire bonding the thing with out breaking it!
Oh, and to nit pick, someone said that the longest distance on a square CPU would be 1.4... * the length of a side, that would be true if you used diagnal traces. Now, I don't purport to have seen every CPU design out there, but I've never known a CAD tool or CPU design that used anything other than straight (i.e. right angle) traces. Eskimo? Am a wrong here? If not, then the worst case path would be 2*length of a side, but any CPU designer that put a trace in like that would be fired. <img src="graemlins/lol.gif" border="0" alt="[Laughing]" />
[ 09-02-2002: Message edited by: Transcendental Octothorpe ]</p>
<strong>
... I can assure you that CPUs are much larger than pinheads. They are approximately the size of your thumbnail and are in fact measured in mm. 100mm^2 is a fairly typical die size for a CPU.</strong><hr></blockquote>
Oh, and while we're talking about this and everybody's hip on the POWER 4 these days, I took a look and some of the Spinacker (POWER 4) dice, and they are over an inch square! (single die, not the whole MCM, of course; there's four in those.). But, granted, these are the highest end chip in production, with two cores and a butt load of on-die cache. The 75xx's are smaller, about half that, I think (and somewhat rectangular, IIRC).
Oh, and I promised in another thread a couple weeks ago to ask if some of the Spinacker team had been working on any single die versions, but the answer was no (except for chips that only had one die working). Guess I shouldn't be suprised, really. I'm sure that this new babyPOWER 4 would have a different code name and be an entirely different project.
Oh well. It was worth a shot.
[ 09-02-2002: Message edited by: Transcendental Octothorpe ]</p>
<strong>
Now, I don't purport to have seen every CPU design out there, but I've never known a CAD tool or CPU design that used anything other than straight (i.e. right angle) traces. Eskimo? Am a wrong here? If not, then the worst case path would be 2*length of a side, but any CPU designer that put a trace in like that would be fired. <img src="graemlins/lol.gif" border="0" alt="[Laughing]" />
[ 09-02-2002: Message edited by: Transcendental Octothorpe ]</strong><hr></blockquote>
Correct, there are some initiatives to attempt to integrate 45 degree interconnects in future designs but for now even the cutting edge designs all use 90 degree traces.
You've set me straight in a proper fashion.
Now would anyone care to elaborate on moving a CPU with conventional dimensions to a more advantagious spot in a portable such as near the outside perimeter of the unit's chassis? Would that be a worthy design goal or am I WWAAAY "armchair" engineering "OUT THERE"!!!!!