Slashdot Mirror
From Silicon To Microprocessors
prostoalex writes "Jim Turley from Embedded Systems Programming magazine answers the question of where microprocessors come from. While the public generally knows about the silicon and microprocessor vendors, few can describe the process of turning the beach sand into the latest and greatest several-hundred-dollars-worth CPU."
The microprocessor stork brings them.
Right, mommy?
I have been pwned because my
That, my friends, is a really unpleasant image.
Then it's sliced into exceptionally thin wafers about 6 to 8 inches (200 to 300mm) across, depending on the diameter of the ingot.
Owwww!!!!
or at least so I gather from the frequency with which the Silicone/Silicon mistake is made. Maybe if computer chips were warm instead of hot, and squeezably soft instead of hard, and bouncy always bouncy people would know more about them.
I'd always thought these materials were made in hot, dry climates, like Arizona, yet there was a supplier right in my backyard.
A feeling of having made the same mistake before: Deja Foobar
Hellacious spawning vats in the dark dungeons of Intel, AMD, IBM, and Apple.
...
*sqlorch*
*SQLORCH*
*Ding!*
The coolest voice ever.
The only thing I don't like about the process is the working conditions: annoyingly loud!
:)
For those of you that have never been in a clean room, there is a tremendous amount of ambient sound due to the very important air cleaning/circulation system. In order to make the clean room "clean", there can only be so much dust particles in the air. (e.g. 1ppm) (there are actually different classes of clean rooms)
The ramification of this is that one can hardly hear one's voice. Personally, I'm glad I'm not in the semiconductor field
A knowledge of history is almost always a Good Thing. I wonder how many programmers have never heard of Charles Babbage? ("Analytical Engine? What?") You should at least have a decent knowledge of the history of your craft. Call me old-fashioned, but my love of computer science isn't limited by EnterpriseJavaBeans and BiCapitalizedMumboJumbo and whatever buzzword happens to be out today. There's more to it than that.
I have discovered a truly marvelous
If you can visit Santa Clara USA then Intel's museum has a nice introduction to the process of turning sand into chips.
"Don't belong. Never join. Think for yourself. Peace." V.Stone, Microsoft Corporation
V I S A
vodka, straight up, thank you!
I read the article and find myself actually knowing in advance how silicon chips are made. You see, in the 80ies we had childrens books about computers that covered something more than how to start Word and update Winblows.
a couple of macroprocessors get drunk, start messing around... they wake up the next morning full of regret... next thing you know, there's a new microprocessor for someone to install, dress up in a nice case, feed it RAM, and reboot it when it makes a mess, which will be all the damn time for the first few months...
the latest and greatest several-hundred-dollars-worth CPU.
;-) The CPU's in these are a couple thousand dollars each.
Only if you're buying intel can you get the latest and greatest for only several-hundred-dollars-worth. We call the intel servers at work "tinker-toys" because they are wimpy and cannot get much real work done.
The Alphaserver GS160, the IBM RS/6000, and the Sunfire 12k. Those are the manly servers that do the real work around here. I don't think you can replace fans in these things for "several-hundred-dollars-worth".
I'd rather be a conservative nutjob than a liberal with no nuts and no job.
While informative on what it touches on, this doesn't describe what goes into making a chip. It describes how a chip is patterned. Then follows many many diffusion, oxidation, etch, and metallization steps that go between each photoresist mask step. I suppose it makes a good read for someone who wants just a general overview. But it makes it sound like making a chip is just a glorified film development process. I do microfab work, and the lithography steps are the steps we take for granted (mostly -- they still do take effort to get right, but are in general easier then what follows).
Having smaller die sizes is not good just because you can put more dies on a wafer. It is because your yield will improve. Dust/contamination is the real enemey, and bigger dies have an (exponentially or even worse) higher risk of having one dust particle destroying the chip function. Cutting the size with 10% may well lower the production cost by 50%.
.18 to .13) can be a real money saver (next to allowing higher clock rates).
And that is ofcourse why moving to a smaller technology (eg from
From the article:
For an example, let's look at a 200mm silicon wafer, which has about 986cm2 of surface area. That's about the size of a salad plate. Let's say your chips are square (most are) and they measure 10mm on a side?that's 100mm2 per chip. If the silicon wafer was also square you could fit 986 chips on your wafer. Alas, wafers are round so you can really only get about 279 chips on a wafer.
I guess the obvious question, since using squares on a round wafer wastes a certain amount of silicon, is why squares? Why not build a hex grid? That would seem to maximize the usage of the available area.
But then, I suppose cutting them out would be significantly more difficult.
What about triangles, then? Straight lines up and down, and in one (or both) diagonal directions.
On the other hand, someone's already thought of this:
Intel's old i960MX microprocessor was octagonal. It was so big its corners had to be cut off.
So my idea has an obvious flaw. The question is... what is it?
Stressed? Me? Of course not. Stress is what a rubber band feels before it breaks, silly.
I thought big screen TVs were "blurry" up close because they had fewer pixels per area. Besides... in this case, you wouldn't be making the image bigger, you would just be making a LOT of tiny images at once. Can someone either explain how his explaination makes sense, or what the real reason is?
Sometimes, when a CPU and a CPU socket meet in the middle of a back alley...
And the muscular cyborg German dudes dance with sexy French Canadians
Hmmm, and all this time I thought 200mm wafers were 8 inches and 300mm wafers were 12 inches. Maybe the author is a former NASA engineer...
And I agree, clean rooms are no fun. Ever trying typing on a plastic-coated miniature keyboard with two pairs of gloves?
A huge amount. Many embedded systems have real-time requirements, tight memory-space limitations, and a much lower tolerance for failure than desktop systems. If you're talking about a comsumer embedded device (e.g. a cellphone), you have to deal with power management as well. There are multiple operating systems to choose from, several types of processor architectures (including the Harvard Archirtecture typified by Intel's old 8051 family that has entirely separate memory spaces for instructions and data), and several buses specific to embedded systems work.
Why should this matter? There are several embedded systems in your car, and I'm sure you'd be mightily ticked if your car just stopped working randomly. On a more mundane level, what about programmable thermostats or the security card readers where you go to work? That's not to mention the mission-critical embedded systems in aircraft and medical devices.
They don't use beachsand, that's silicon dioxide (SiO2), also known as quartz.
Pure silicon chunks are actually made from condensing a very pure Silicon gas called Silane. The chunks are broken up, and melted in a very hot furnace, with a crucible made out of quartz(usually). Any doping, or impurities to give the silicon it's different electrical properties are added at this point. Boron (B) is fairly common.
Then, a nice perfect seed crystal of silicon is dipped into the molten silicon which starts to crystalize around the seed crystal. The growing crystal is turned and slowly pulled out of the liquid silicon as it grows to help keep it regular. The result is called a boule, or "the bologna looking thing"
As a side note, the doping is usually too high at the top of the boule, and too low at the end of the boule, so only about the middle 25% is used.
Then it gets sliced into wafers. etc. etc.
There are more than a few nits...
.13u, .18u, or larger.
(1) Silicon is not sand. Sand is silicon dioxide (well, most sand). It needs to be reduced (the oxygen needs to be removed) and purified. And purified. And purified. (I believe Brazilian quartz is actually the preferred stock for silicon dioxide, rather than sand, due to its purity.)
(2) Photo-resist does not need to be electrically conductive. It does need to be capable of resisting attack by whatever chemicals are next in the step (especially the HF). Since they're usually polymers that are either polymerized or depolymerized by the exposure, they generally are not conductive.
(3) Current generation laser steppers are not EUV. (They are UV, maybe DUV, being slightly less than 1/2 the wavelength of visible indigo.)
(4) One could get the impression that each chip on the wafer is processed separately at each step.
(5) Fabs and foundries are related but distinct entities. (I personally have worked in a fab, but never a foundry.)
(6) It's the mask that is imprinted on the wafer's photoresist, not the chip.
(7) Moore's law is incorrectly repeated. This is especially bad because it claims to be correcting the common belief (which it probably is). Moore's law was about the economics of chip density -- the most _cost effective_ density doubles every 18 months.
(8) I've usually heard and talked about individual die and multiple dice. (And breaking up wafers into chips is called dicing.) Maybe others call them (plural) die, but not everyone.
(9) The 200mm wafer area calculations are wrong. A 200mm wafer has a radius of 10cm; the area is therefore (10)^2*pi ~= 310cm^2. So one won't get 986 die from a square wafer and only 279 from a round one.
(10) Lots and lots of companies don't build their chips on the smallest feature sizes possible. Very few can afford to manufacture 90nm chips at this point, so the bulk of chip _designs_ are manufactured at
There are probably many more errors...
RJ
More to the point, why are humans required at all in the manufacturing process. I would expect the entire manufacturing and testing process, from sand to plastic-encased chip, to be automated enough that people in bunny suits should not be needed. Maybe they are needed to replace the robots and fill up the supplies, but other than that, what do they do?
The article mentions that, with co-workers encased in bunny suits, you have to look at their eyes to tell people apart. When I worked in a fab, I noticed I became very attuned to people's body shapes and ways of moving. After working there for a while, I could subconsciously identify co-workers at the opposite end of a shopping mall, simply by the way they walked.
Unless you are talking about a clean room from the late 70s or the 80s, its more likely that the noise you are hearing is from the exhaust systems sucking fumes from processing equipment.
The materials used to produce semiconductors are extremely deadly to humans as are many of the process by products.
Pretty much every processing tool has multiple exhaust connections which remove potentially harmful fumes to a scrubbing system on the roof that removes the toxic chemicals which are then treated and disposed.
There are other noises from the tools and support equipment but I assume you thought it was the laminar air flow filtering system because it sounded like high volume air movement. They do move high volumes of air but you don't want the air moving too fast as it will stir up any particles that may be present in the room.
burnin
oh, I do work in a clean room, have since 1989.
* when cutting the ingots, people almost ALWAYS use a ring-blade; where the blade is on the inner edge of a ring larger than the ingot, and ingot is sliced. extra points for anyone who know why.**
* ingots are not always "grown." (think dipping candles) there is also a technique where you start off with a polychrystaline ingot and use localized heating to progressively monocrystalize it by localized melting. The technique is similar to one of the methods of removing impurities from iron bars.
* CMP is damn cool. I mean, it's nice and all hearing about "polish to within an atom" precision, but if you take a polished wafer, it would make the best mirror you'd ever own. Granted silicon is not the perfect reflective surface, but you won't get a mirror more accuratly shows every feature on your face. =) Otoh, when dusts and stuff DO get into the CMP machines, though, it scratches the wafer. Though you don't see it, when you trace failures on the wafer the failing gates would generally follow an arc shape (corresponding to the wafer and polishing head rotation), and from that you get the CMP machine checked out.
random junk I thought that was kinda neat.
** I used to know about 3 years ago but then I forgot. so don't expect like a correct answer or nothing.
My life in the land of the rising sun.