Author
|
Topic:ðð How exactly are chips produced on those giant cookie sheets anyway?
|
Alcimedes Junior Member
|
posted 09-28-2000 10:49 AM ððð
ðððð ðð ðð
Ok, not sure if this is General Discussion, or Current Hardware, but this seemed good enough to me.How
do they make the chips on those sheets anyway? I know they grow them or
something, but I no idea other than the very, very basics, which are
probably wrong anyway. Anyone out there able to fill me in on this? Someone's gotta know. Thanks. -Alcimedes IP: Logged |
gEEk Moderator
|
posted 09-28-2000 12:45 PM ððð
ðððð ðð ðð
This was posted in the VLSI lab at my grad school... quote:
How to paint your room:
- Put an open can of paint in a paint-shaker.
- Turn on the paint shaker. Run out of the room (quickly!). Your room is now completely covered in paint.
- Put masking tape over everything that you actually wanted painted.
- Put an open can of paint remover in the paint-shaker. Turn it on and run out of the room.
- Clean away the sludge and remove the tape. Your room is now painted.
This
is a fairly accurate description of how circuits are built on silicon.
Eskimo is better suited than I am to describe this, but here goes.
(Eskimo... let me know where I screwed up: much of this info is based
on rather poor memory of 7 year-old tech) First,
deposit an insulator over your wafer. Then deposit polysilicon (the
gates of all of your transistors). Then use a mask to photographically
apply a photoresist layer over your poly where you want it to actually
remain. Then etch away the poly that you don't want. Then apply a
photoresist to etch away the insulator (but leave the poly) that you
don't want (these will be the sources and drains of the transistors).
Now you are left with polysilicon traces (Transistor gates, the width
of which is that 1.8um or whatever value listed for the process) and
insulator with holes revealing bare silicon. Now diffuse N+ into the
bare silicon if it is a P substrate or P+ if you are using an N
substrate to make the sources and drains. You have just made a
transistor. Now
lay down more insulator and repeat the process to put down as many
metal layers (aluminum or Copper) as your process allows. "holes"
called vias are etched in the insulator layers to connect one metal
layer to another, or a metal layer to the polysilicon or diffusions. Note
that most processes have several layers of polysilicon, areas of both P
and N substrate so you can make both N and P channel FETs (hence CMOS: Complementary Metal Oxide Semiconductor) and many metal layers (like six, or so) Close, Eskimo? ------------------ -gEEk There's such a fine line between stupid and clever. --David St. Hubbins, Spinal Tap IP: Logged |
gEEk Moderator
|
posted 09-28-2000 12:58 PM ððð
ðððð ðð ðð
By the way... this is a pretty general question, so I'm moving this to General Discussion. You were right! ------------------ -gEEk There's such a fine line between stupid and clever. --David St. Hubbins, Spinal Tap IP: Logged |
Eskimo Moderator
|
posted 09-28-2000 02:07 PM ððð
ðððð ðð ðð
Pretty good for an EE gEEk . I'll post on this when I get home from work. I'll even throw in some pretty pics maybe.IP: Logged |
Eskimo Moderator
|
posted 09-28-2000 07:21 PM ððð
ðððð ðð ðð
How are Chips produced?This
is a long description but I hope pretty easy to follow. IÌll break it
up into chapters to make it easiers. 1 chapter per post. Chapter1: Vacuum cleaners, telephones, and semiconductors?! In
the beginning there were vacuum tubes. The discovery of vacuum tube
technology by Lee Deforest in 1906 was the jump start of electronic
signal processing. These tubes were able to both switch on and off as
well as to amplify an electrical signal. They made it possible to build
ENIAC the worldÌs first electronic computer. This computer weighed 30
tons and took up 1500 square feet of area. It used nearly 20,000 vacuum
tubes. However powerful it was for the time however, it suffered from
the drawbacks inherent to vacuum tube technology. ItÌs uptime was very
short due to the fact that one or more of the tubes was constantly
burning out. Vacuum tubes are hot, prone to leak, require a lot of
power, and donÌt last very long. These
problems led three people from Bell Laboratories (Formerly of AT&T,
now Lucent Technology) to search for a better replacement for a vacuum
tube. Two days before Christmas in 1947 John Bardeen, Walter Brattin,
and William Shockley demonstrated the ability to electrically amplify a
signal formed from Germanium. They called this invention a
ÏtransistorÓ. It was smaller, took much less power, and had a very long
lifetime. In 1959 Jack Kilby of Texas Instruments took the next step in
the advancement of the industry by forming the first Ïintegrated
circuitÓ or IC. It was a complete circuit consisting of several
transistors, capacitors, and resistors in one solid piece of Germanium.
This was a VERY important step forward. Suddenly a company could
combine numerous electrical components on one small ÏchipÓ. Robert
Noyce and Jean Horni of then Fairchild Camera (now Fairchild
semiconductors) were able to take KilbyÌs circuit with its connecting
wires and make patterns, trenches if you will, in the natural oxide
that grew on silicon (see Ch. 2). Into these trenches they evaporated
Aluminum which conducts electricity. This circuit is THE model for ALL
integrated circuits up until today. Cost effective and able to be
miniaturized an industry was born. The race was on J Who could provide
the most functionality on one small piece of semiconductor? What
became of these pioneers? Well Robert Noyce left Fairchild to found a
small start up with Andrew Grove and Gordon Moore which they dubbed
Intel William
Shockley left Bell Labs and formed Shockley Laboratories in Palo Alto,
CA, it provided the beginning of what is known today as Silicon Valley. Two
of Robert NoyceÌs co-workers at Fairchild were Charles Sporck and Jerry
Sanders. Sporck left to found National Semiconductor (they made Cyrix
processors), and Jerry Sanders left to found Advanced Micro Devices.
IP: Logged |
gEEk Moderator
|
posted 09-28-2000 07:35 PM ððð
ðððð ðð ðð
quote: Originally posted by Eskimo: They
made it possible to build ENIAC the worldÌs first electronic computer.
This computer weighed 30 tons and took up 1500 square feet of area. It
used nearly 20,000 vacuum tubes.
This is good stuff! Just
a quick trivial aside: The accumulators in ENIAC were actually tall
vertical panels with a single column of lights (16), each indicating a
single bit. When several of these accumulators were lined up in a row,
you ended up with a large matrix of flashing lights, indicating the
values stored in the accumulators. That image of a wall of flashing lights became the standard way of indicating "computer" in movies and TV for years to come.
------------------ -gEEk There's such a fine line between stupid and clever. --David St. Hubbins, Spinal Tap IP: Logged |
Eskimo Moderator
|
posted 09-28-2000 07:59 PM ððð
ðððð ðð ðð
Chapter 2: From sand to your screen.The
basic ingredient in almost every chip or processor made is Silicon.
Atomic number 14 on the periodic table of elements. Silicon is sand
basically. If youÌve ever taken a walk on a sandy beach you are walking
on the basis of semiconductors. What is a semiconductor you say?
Everyone throws that around so lets try to explain it a little. One
important property of a material is its ability, or inability, to
conduct electricity. IÌm assuming youÌve heard of things like atoms,
and electrons? Well electrons orbit around the center (nucleus) of an
atom. In some elements such as metals like Aluminum, copper, and gold
atomic forces do not hold the electrons in the outermost orbits of the
atom tightly. So it is relatively easy to move these electrons away
from that atom and thus form electric current. A materials ability to
conduct electricity is known as conductivity. So metals are usually
conductors because they have a high conductivity. Opposite
of metals are materials such as glass which is an insulator. Insulators
have a very low conductivity. Do you know what glass is? Well your
plain run of the mill glass is basically referred to chemically as
SiO2, Silicon dioxide. Do you see Silicon in there? I do. Hmm, so SiO2
is a good insulator. WeÌll keep that in mind. Ah
now onto the meat, semiconductors. There are certain materials which
appear in column IV of the periodic table of elements which are known
as Intrinsic semiconductors. These are Silicon and Germanium. (There
are other III-V and II-VI semiconductors but IÌm not going into them)
These two elements have 4 electrons in their outermost shell (orbit).
Now IÌm not going to explain why exactly this is so special, just take
my word that this makes these elements suited to fit our needs for
integrated circuits. By themselves Silicon isnÌt that great in IC
applications. But, silicon is very easy to ÏdopeÓ. Doping is simply
introducing atoms of other elements with more or less than 4 electrons
amongst all the silicon atoms. This unbalances the siliconÌs 4
electrons and causes it to be slightly conductive. The more foreign
substances you dope into the silicon (commonly Boron, Phosphorus, and
Arsenic) the more conductive silicon can get. Cool so if we control how
much ÏstuffÓ we put into the silicon we can control how conductive it
is! This is very nice indeed. BUT, yes there is a but, darn it all .
See electrical current is carried by the movement of electrons (there
are opposite conductors known as holes but that is just more confusion ).
But if we ÏstuffÓ too many foreign atoms into the silicon it gets
pretty crowded. So all of a sudden its not too easy for the electrons
to move around and the conductivity actually decreases because the
electrons are stuck in traffic! So the engineers and scientists have to
consider all of this to reach a balance. WeÌll go into how they get
those foreign atoms into the silicon a bit later. Well
now that you are an expert IÌll try to explain how we turn that sand
into Ïcookie sheets as you so eloquently stated. Basically silicon
(sand) is taken from the earth raw. Reacting it with chlorides and
evaporating to form tetrachloride or trichlorosilane gas purifies it.
This removes metals in the ore that might have been present. We want
this stuff pure! The gas is reacted with hydrogen and reconstituted
into electrical grade silicon. This silicon is ultra pure, like
99.999999999999%. See pic below: It
is in a crystal structure (super important). See crystals by nature are
like rows and rows of legos that are all identical to each other in
shape and size. And they all line up with each other in very tight
rows. Now since making processors is a very very precise process we
want our starting material to be exactly the same. We want every chip
to be the same as every other chip so we want uniformity. Crystals work
great for this. Plus the crystals have big spaces inside of them to
hold our dopant material. So
they dump all that purified silicon in a big melting pot and heat it up
to really high temperatures. (Melting temperature is around 1400
degrees Celsius). Into the vat of melted silicon a small crystal of
silicon that is already arranged with a crystal orientation we
want(called a seed) is attached to a metal rod and slightly dipped into
the vat. This seed crystal is our ÏidealÓ lego block. Now silicon is
very good about copying another solid crystal when it is in its liquid
state. So the silicon swirls around and the atoms arrange themselves to
match the seed crystal. The seed is very slowly rotated and removed
("pulled") from the vat, as it leaves the silicon attaches and cools to
form solid. Because it is being rotated it forms in a cylinder. So
after a long pull you have this HUGE cylinder (standard is 200mm in
diameter and weighing 450lbs.) of silicon that is crystallized and
uniform. This method of
ÏgrowingÓ silicon crystals is knows as the CZ method. Next a super
precise diamond saw slices this huge cylinder into very thin slices
(think of a deli slicer cutting salami). These slices we call wafers! edited for some spelling [This message has been edited by Eskimo (edited January 07, 2001).] IP: Logged |
Eskimo Moderator
|
posted 09-28-2000 08:22 PM ððð
ðððð ðð ðð
Chapter 3: We need a special place to make our chipsSo
weÌve got these wafers now. They are real pretty too, very shiny like a
mirror because while you werenÌt looking we went and polished them
using a very advanced process called CMP. There were basically put on a
pad (like one of those auto buffers) and spun around in a slurry
(mixture) of some very special stuff. LetÌs leave it at that ok? While
they are spinning (there are usually 5 at a time spinning on 5 pads on
one machine) all 5 arms holding the pads are rotated in a big circle.
ItÌs hard to describe but it reminds me of that ride at the fair that
always makes me feel sick. Where you are sitting in a seat in a little
capsule, and that capsule is spinning but its on a big arm that is
rotating as well. Well all that stuff doesnÌt make the wafer puke, it
makes it really really smooth. I just want to explain CMP because it is
used a lot now of days on many steps as we go along. Now
there is a ton of stuff to be done to these wafers to get a G4 or
Athlon out of them. When we make an Athlon there are literally over a
thousand steps it goes through. The process of making a bunch of chips
on a wafer is known as, well, a process J. See each wafer is pretty
big, and chips are little tiny things (about the size of your
thumbnail) so we might as well fit as many on as we can. This will save
us money making a bunch at once! So throughout the process there are
basically a few Ïspecial departmentsÓ known as modules that the wafers
go back and forth between. Oh
let me stop and explain something else real quick. Since these tiny
little things are so complicated and tiny one little spec of dust can
wreck them. Dust may seem small to you and me but it is huge compared
to the stuff we are working with. So we want to keep all that dust away
from the wafer and the chips. This is why microchips are made in
special factories that cost BILLIONS of dollars called Fabs. I guess
Fab stands for Fabrication facility but everyone just calls it a Fab.
To fab something is synonymous with make it too. So all these modules
are inside of these expensive fabs. Now these fabs have millions and
millions of dollars of equipment guaranteed to keep the wafers as clean
as possible. Huge fans circulate the air inside through very expensive
filters. In fact, now try to picture this, ALL the air in the whole fab
facility is completely sucked out and replaced every 6 seconds!!! So it
is very loud in the fab from all these fans J. The air generally flows
out of the ceiling and out through grates in the floor. All the people
inside the fab have to wear special ÏcleanÓ suits. The suits arenÌt
there to protect the people from all the dangerous equipment and
chemicals but instead protect the wafers from all the dirt and other
nasty stuff we humans shed constantly. Fab25 (where I work) is considered a class 1 clean room.
This means the largest particle you should find inside is less that .1
microns!!!! On top of that there you should only find 1 particle that
big per cubic foot of air. For comparison I believe most hospital
operating areas that are sterile are around Class 10,000 (10,000
particles instead of 1 per Cu. Ft.). A Micron is 1 millionth of an
meter. 0.1 microns is the size of smoke particles!! Human hair is from
5-500 microns in diameter. We are talking very tiny. BTW, remember
micron because we measure a lot of stuff in microns. So
you have all these modules with all their very special equipment in
them. Each tool costs well over a million dollars to purchase. And each
module has hundreds of these machines. Now you see why it costs
Billions to build a modern Fab. [This message has been edited by Eskimo (edited September 28, 2000).]
[This message has been edited by Eskimo (edited January 07, 2001).] IP: Logged |
Eskimo Moderator
|
posted 09-28-2000 09:31 PM ððð
ðððð ðð ðð
Chapter 4: Mmmm, club sandwich on rye J So
we are inside our nice clean building with lots of fancy equipment and
these really cool looking wafers of silicon. What next? Well its time
to make some chips (sigh of relief settles over poster who has been
waiting for this part ) When
you build a device on a wafer basically it is a lot like making a
sandwich. You just keep adding layers and layers until you get the
desired result. Of course itÌs more complicated then that. Remember I
mentioned modules last chapter? Well there are 5 main modules in a fab.
Photolithography, Etch, Polish, Diffusion, and Implant. A guy could
write pages and pages about all the things each module does but IÌll
try to keep it basic and brief. Photolithography:
When a bunch of Electrical engineers get together and decide to make a
processor they figure out all the transistors and wires they will need
to do all the cool stuff that a processor can do. Well these guys draw
these designs just like an architect would draw blueprints. They
basically make the blueprints for the chip, thus they are usually
called computer architects. Now these blue prints for all these little
transistors and wires have to make their way onto the wafer. The way
this get done is known as photolithography. This process is actually
very similar to how a camera works and how film is developed. Kodak is
pretty big in the photolithography and it is a reason why a company
called Fairchild Camera that I mentioned in Ch. 1 could be big in the
industry. The
designs that are drawn up are taken and made into big quartz plates. On
the quartz is a layer of chrome. The chrome is removed in the areas
that represent all the trenches that the engineers want to appear on
the chip. This quartz plate is known as a photomask or just mask. It is
sort of like a negative is in photography. Well a wafer is taken and
covered with a syrup like polymer material called photoresist. This is
a very special liquid. Because ordinarily it is very hard to remove
except with something like acetone. But due to complex chemistry when
this resist is exposed to certain wavelengths of light it becomes
soluble (you can dissolve it more easily). So the wafers are placed in
a giant machine called a ÏstepperÓ. The
mask in inserted into the stepper as well and a light source is shined
through the mask and onto the wafer. So remember when you were a kid
and made shadow puppets? Well that is what happens to the wafer as the
chrome on the mask blocks out the light and only allows it to shine
through the clear spaces. So only the places under the clear area are
ÏexposedÓ. The exposed areas will be able to be removed by dipping the
wafer in some solution that is alkaline like sodium hydroxide or
potassium hydroxide. This gets rid of that resist in those areas and is
known as developing. Now there are all kinds of intermediate steps I am
skipping to keep this simple as well as all sorts of problems that have
to be overcome. One
basic one is that to make very tiny lines or features on our chip we
need some special sort of light. In fact you probally canÌt even see
this light. We are using deep ultra violet light (DUV). It has a
wavelength of 248nm (for reference humans can generally see from around
300nm (violet) to 650nm (red). Now 248nm = .248 microns. LetÌs just
round that off to .25 microns. This number might look familiar as many
devices including Pentium II processors and Athlon processors were
produced at this size. Now the light is limiting but using fancy
optical tricks we can make things even smaller using that same
wavelength of light, such as .18 micron like the new Motorola 7410 or
the AMD Duron processor. While
we are at this point we should discuss why we call a stepper a stepper.
Well the mask that is used is many many many times larger than the
actual chip. The light is focused through the mask and uses lenses to
shrink the image down. Think opposite of a magnifying glass. So it is
actually putting this design on a very small part of the wafer. So
after it exposes the place for that die (chip) it ÏstepsÓ over next to
it and exposes another and another and keeps doing this until the whole
wafer is filled up. This is very cost effective. So
what do we have now? We have a wafer covered in resist except in
certain areas. What should we do with it? Well the wafer will go
through photolithography MANY times throughout the process, so IÌm just
going to pretend we are making a basic transistor. To do this IÌm going
to send our wafer we exposed and developed and send it to the Implant
module. But
basically this is the fundamental step of making processors, layers are
added, and using lithography we can remove the parts we donÌt want
there and leave other parts intact at very small sizes.
[This message has been edited by Eskimo (edited September 28, 2000).] IP: Logged |
Eskimo Moderator
|
posted 09-28-2000 10:17 PM ððð
ðððð ðð ðð
Chapter 5: Shooting cannonballs and Fire! J Ah
implant, where little boys who still havenÌt grown up all the way can
live out their fantasies of shooting things while making a lot of money
J. Remember back in Ch. 2 I mentioned that Silicon was neat because you
could change its electrical characteristics by inserting foreign atoms?
Well this is how we get them in. We use giant tools known as Ion
implanters. An
Ion implanter literally blasts ions of a substance at the surface of a
wafer and the ions become implanted, or lodged, in the crystalline
structure of the Silicon. Using extremely high voltages and utilizing
some neat laws of Electro-Magnetic laws we can accelerate an ion
(basically an atom of a substance in a charged state) to a very high
speed. Then using magnets we can control the path of the ion and send
it hurtling at the wafer.
Not to make this sound overly simple but basically after using complex
equations and experimentation to determine how much ÏdoseÓ the wafer
should receive an operator simply dials in the concentration of ions
per cubic centimeter desired and presses the start button. The machine
does all the rest. IÌm going to shoot a nice healthy dose of phosphorus
into some clear areas IÌve made in our wafer. Now
you will see in the above picture there is some oxide on it. How did
that get there? Well I should have explained this before Photo but
didnÌt because I think photo is more important .
As I mentioned awhile ago glass is Silicon dioxide. It forms naturally
when pure silicon is exposed to oxygen. Like rust is to iron so to SiO2
is to silicon. But the oxide that grows on silicon naturally in the
normal air is not good for us to use. We can do better J. So we take
our wafer over to the Diffusion module where they play with fire.
Basically a high quality layer of oxide can be grown on top of our
silicon wafer by putting the wafers in a very hot (1000 degrees
Celsius) furnace and pumping in some pure oxygen. After a few minutes
we will remove the wafers and they will have a nice oxide on them.
[This message has been edited by Eskimo (edited September 28, 2000).] IP: Logged |
Eskimo Moderator
|
posted 09-28-2000 10:53 PM ððð
ðððð ðð ðð
Chapter 6: ItÌs all about stripping baby! Then we buff it upJ So
if youÌve been following this far you understand that basically a wafer
keeps passing back and forth between modules. If you canÌt tell IÌm
getting a little tired from describing EVERY step so now IÌm just going
to tell you what etch and polish do and then at the end IÌll show you a
cross section of what a basic transistor looks like built on top of
Chapter 5Ìs post implant cross section. Etch
is there to remove unwanted material. In the old days this was done by
using chemicals such as Hydroflouric (HF) acid. But commonly now of
days to achieve better results we use dry (plasma) etching. Plasma
is a chemical process that uses gases highly excited by high energy
until it forms into a plasma state. This is the 4th state of matter;
there is solid, liquid, gaseous, and then plasma. The power to excite
the gas is provided by a RF (radio frequency) field. The energized
plasma then attacks the surface of the wafer and literally chews it way
through silicon. As it does this it gives off some very pretty colors
and radiation that can be measured on spectroscopy tools. And stopped
when the emitted light changes to indicate that it is done going
through one material and has reached another. It is then stopped. A
good etcher is one that is able to etch almost completely vertically
without any horizontal damage as it tunnels down. Also it should be
highly selective and only etch away the material desired to be removed. After
the transistor structures have been formed using oxide, doped regions
of silicon, and a gate material metal such as aluminum or copper is
used to connect the different sections of the transistor into a more
complex transistor with the millions of other transistors in the chip.
I will cover copper since you are probably more interested in that and
it involves polish much more. Before the metal is added the wafer will
look similar to this: The
entire surface is covered in oxide first. Then the pathways for the
conduction lines are cleared using photolithography and etch steps.
Once that has been done copper is laid down over the entire surface of
the wafer. The remaining oxide keeps it from going where it isnÌt
wanted. So it settles into the trenches. But now you have all this
copper all over the place sort of splattered on the wafer, not very
precise huh? Well
along comes our buddies in polish. Using the CMP process I outlined
earlier they polish the entire wafer until it is smoothed down to the
point where all the overlapping residual copper has been buffed away
and only remains inside the channels. Then another layer of oxide is
grown and another metal layer is repeated and so on. Modern processors
use up to 6 layers of metal stacked on top of each other since it takes
that many layers to route all the wires necessary to connect the
millions and millions of transistors. Think of it as multiple
overpasses over overpasses on the highway to route all the traffic
smoothly. After all this you basically have a completed wafer like the
one pictured below except this one is using Aluminum instead of copper
metal. Cross section when done, 1 transistor. Now imagine you were making millions of these at once.
IP: Logged |
Insanely Great Member
|
posted 09-28-2000 11:27 PM ððð
ð ðð ðð
Duck and cover! We got an eskimo on a rampage here! Good god Eskimo... you either love your work or love to write.=-Insanely Great IP: Logged |
King Chung Huang Member
|
posted 09-28-2000 11:39 PM ððð
ðððð ðð ðð
This is great Eskimo! I'm looking forward to future chapters. IP: Logged |
iPad Member
|
posted 09-29-2000 01:35 AM ððð
ðððð ðð ðð
OMG, man, i'm printing this out, this is good stuff! Eskimo, u da man!------------------ Join Team GameDev.net's SETI@Home Team! http://setiathome.ssl.berkeley.edu/stats/team/team_79972.html IP: Logged |
bubba_nuts Member
|
posted 09-29-2000 01:41 AM ððð
ðððð ðð ðð
Very interesting stuff!I sort of think that this should have been kept in Current Hardware, though. IP: Logged |
Eskimo Moderator
|
posted 09-29-2000 01:56 AM ððð
ðððð ðð ðð
quote: Originally posted by King Chung Huang: This is great Eskimo! I'm looking forward to future chapters.
What would you like in future chapters? IP: Logged |
Alcimedes Junior Member
|
posted 09-29-2000 08:28 AM ððð
ðððð ðð ðð
man, never thought i'd get such a great response to such a simply worded questin. i
do have one question though. when you take the seed crystal, and dip it
into the vat of molten sillicon, is there any way to have it come out
as a square rod rather than a round one? isn't there a fair amount of
wasted space on every wafer, since you can't make a square transistor
fit perfectly into a round base? ok,
my next question is, what happens so that some chips come out being
able to run at 700 Mhz, while chips from the same exact bacth will only
run at 500Mhz. is it a crappy seed crystal, poor etching, poor
polishing? is there any way to look at the good chips and tell what is
different with them in relation to the slower chips from the same batch? -Alcimedes p.s.
this has to be up there as one of the best replies ever. you might want
to think about just putting your replies somewhere on reserve around
here in case other people wonder. this is great stuff. i feel like i'm getting a mainline of knowledge here. IP: Logged |
gEEk Moderator
|
posted 09-29-2000 01:48 PM ððð
ðððð ðð ðð
WOW! Great post, 'mo!You put my feeble excuse for a response to shame! (although I still like that paint-the-room analogy from my VLSI lab ) My
gf back in grad school was a Materials EE and did Molecular Beam
Epitaxy and Ion Implantation. Your discussion of II brought it all back
. She made Christmas tree ornaments out of 4 inch wafers she had made! Let's
have the next chapter! I noticed no poly in those diagrams you
posted... does AMD use metal for gates? How about a discussion of the
layers and stackup of the process used for Athlons? (or is that
Athalons? ) ------------------ -gEEk There's such a fine line between stupid and clever. --David St. Hubbins, Spinal Tap IP: Logged |
CosmoNut Junior Member
|
posted 09-29-2000 03:56 PM ððð
ðððð ðð ðð
I'll
print that off and read it later when I'm at home and don't have to
spend 10 cents a page here in the computer labs on campus.Looks really interesting -- and thorough ------------------ iCan, and iWill. IP: Logged |
JGuidroz Junior Member
|
posted 09-29-2000 04:06 PM ððð
ðððð ðð ðð
My university has a class 100 clean room for chip
manufacturing(nothing big, just testing mostly). I went on a tour of
our solid state lab the other weekend, and it's pretty cool to hear
about how chips are made.IP: Logged |
Eskimo Moderator
|
posted 09-29-2000 07:51 PM ððð
ðððð ðð ðð
JGuidroz: Where do you go to school at?gEEk:
Thanks, yes those diagrams showed metal gate mosfet's since they are a
bit more simple than poly to explain. No we do not use metal gates
almost everyone uses some sort of poly. As much as I would love to talk
about layers in an Athlon (and could I ever ) I'm under an NDA that strictly forbids such things. Your paintcan thing did give me a good chuckle though I
might work on the next chapter tonight sometime if I get up the
ambition. I'm drained from work and at this moment don't want to think
about it. Alchimedes asked a fairly good question about what dictates
speeds. The next chapter will probally be on backend wafer processing
and testing. This is about as close as I have any personal experience
to speeds. If I have the time i'll also go back and add a bit to the
last chapter as I feel I cut it too short. Thanks for the positive replies ------------------ When you pirate MP3's you are downloading Communism. "LAN party and AMD huh? Aren't you just the ubergeek huh?" IP: Logged |
tonton Member
|
posted 09-30-2000 12:10 AM ððð
ðððð ðð ðð
If only
textbook authors used phrases like "...they are real pretty too...",
we'd have a hell of a lot better results in college.Good work, Eskimo! Ever consider writing "Microprocessor Fabrication Processes for Dummies?" You certainly should publish this work on the web, at least, and more than just in the AI forums! ------------------ "Extreme Mobile User No. 1" Apple, please give us a 3 lb. subnotebook! [This message has been edited by tonton (edited September 30, 2000).] IP: Logged |
Eskimo Moderator
|
posted 09-30-2000 01:16 AM ððð
ðððð ðð ðð
Chapter 7: Back end and loose endsWell
when we last we left our intrepid wafer it was basically finished.
Remember the breakdown I gave you was highly simplistic and modern
processors undergo hundreds and hundreds of process steps. Wafer fabs
operate 24 hours a day 7 days a week and it still takes on average
about 60 days from when a wafer is first ÏstartedÓ until it arrives at
the Ïback endÓ. The back end is simply where the completed wafers will
be tested, go through a connection process, and sorted. IÌll explain
what each of these is below. A
modern large-scale production fab facility will start on average about
5000 8Ó wafers per week. This means you will end up with 5000 finished
wafers each week. Now some wafers will be lost in the process. Wafers
are fragile if stressed in certain directions. Remember that Silicon is
closely related to glass, so essentially you have 8Ó thin slices of
glass. If you drop one it WILL shatter. Also from time to time the very
expensive and complicated tools mentioned earlier will eventually have
an error while processing a wafer. Many times nothing can be done to
erase damage done and the entire wafer has to be Ïscrapped.Ó In general
when a wafer is scrapped it is sold for a little bit of money to a
company that recycles them by melting them down and making new wafers
out of them. Just like tin cans. One term you will hear over and over
and over again in the semiconductor industry is yield. This is the
measurement by which everyone measures how successfully a fab is
operating. What good is it if you can make a 2GHz processor if you can
only make 2 for every 1000 you try to make? YouÌll never make any money
like that. So every company wants their yields to be as high as
possible. Yield can be measured in two ways. Outsiders and laymen often
like to express yield as a percentage. Thus take the number of good
chips, divide by the number of total chips and multiply by 100% and you
have a yield. In the industry a more common practice is to describe
yield in die per wafer. Die is just another term for a chip. Why do we
do it this way? Well the answer is that even engineers like to make
math easy on themselves. If you know the average die per wafer
produced, and you know the number of wafers you finished with, all you
have to do is multiply to get your total chips. So if we were averaging
200 dpw (die per wafer) and had 5000 wafers then we would have made 1
million chips that week. Now there are lots of different yields such as
line, sort, functional, and diegen among others. IÌm not going to
explain them all, they just describe the yield at different steps of
making a chip. It
is a sad fact of reality that not all of the chips on a wafer will
operate perfectly or even at all. There are many reasons why this is
so. Perhaps one of those ultra precise machines was not performing
perfectly. Perhaps a spec of dust managed to land on that chip. There
are endless possibilities. In my example above this fictional company
would be faced with quite a dilemma. They have 1 million processors
that need to be tested to make sure they operate properly. Suffice to
say that given the time needed to perform such tests that it would take
a massive investment in personnel and machinery in order to do that. So
what can a company do? Now we get to what I actually do at my job. You
see when the chips are placed on the wafers there are tiny little lines
in between them all. This gives us room to cut them apart when we are
done. Well we love to be efficient with our use of space on the wafers.
So inside of these lines we put all kinds of different electrical
testing structures. They are put in the plans of the chip so that as
the photo masks lay down patterns and the implanters implant while they
are building they chip they are also forming our test structures in
these Ïscribe linesÓ. What does this mean? Well this means that these
test structures were constructed under nearly identical circumstances
as all those hundreds of chips. So instead of testing each and every
one of those chips lets see how the wafer is looking by taking a sample
using these structures. There are structures representing everything
found inside the wafer. There are transistors, capacitors, resistors,
and some other special structures. If something happened that messed up
this whole wafer so that very few or none of the chips on it will end
up working we will be able to tell without having to test all the
chips. We run a series of electrical tests on the structure and look at
the results. At AMD we call this process ÏWafer Electrical TestÓ or
W.E.T, at other companies it is simply ÏElectrical TestÓ or ÏParametric
TestÓ. And it is the results of these tests that I personally
analyze and investigate. Because every failure is a clue as to what can
be improved or fixed inside of the fab for better future yields. So if
a wafer fails at WET we would simply scrap it rather than waste time on
it anymore. Now
that we have hopefully weeded out all the bad eggs in the bunch lets
get this puppy ready to call itself a real processor. A processor is
useless if you canÌt plug it into your motherboard. So something has to
connect the miniscule little connection places on the actual chip to
bigger things like wires or pins. Every company does this differently
but most including Intel, AMD, and IBM (probably Motorola as well) use
a technology called Ïflip-chipÓ or Ïbump.Ó Using some more fancy
process tools like photolithography and etch tiny little bumps of metal
are deposited on the surface in the specified locations of connection
points. When you get done you have chips covered in these little bumps.
The great thing about this is that when the bumps are on all you have
to do is literally flip the processor over onto a waiting package with
pin outs to the motherboard. (IÌm sorry if this is confusing, it is
fairly hard to describe and I donÌt know as much about this area). Here
is what a wafer looks like after it is bumped. Now
that our wafer has been bumped it is time to see how well this thing
works. This is done at Sort. Each wafer is taken one by one into a
large machine that applies probes to all those little bumps. Well over
a thousand tests are performed testing all sorts of variables. We want
to see how much power the chip is going to take, how fast it will
operate, and that this processor is reliable. AMD tests to ensure that
every processor will perform at peak efficiency for a minimum of 10
years. Most other companies have similar criteria. Now these tests are
quite basic. We arenÌt running Photoshop here.
It is digital with 1Ìs and 0Ìs being put in and waiting for the right
1Ìs and 0Ìs to come back. More detailed testing will occur after the
chip is packaged and it is taken, binned, and burned in. But this is
far from the fab. Not all of these chips are going to work. Some will
not function at all, some will require too much voltage for our needs,
and some will not operate fast enough to sell. I wish I could tell you
all the reasons things can go wrong but I canÌt. Usually the reason
falls into one of two categories. It is either defect driven or process
related. Defects involve all those little particles that can mess up
our chips or a slight scratch from some exposed metal. Process related
means that somewhere in the process something went wrong that was
slight enough that it was able to pass W.E.T. but important enough to
make our processor not run correctly or at all. Now in the old days if
a tester discovered a die wasnÌt functioning it would spit a big spot
of ink on it to signify such. Now of days we do it electronically and
just store the location of that die in memory. Sort
is traditionally the last step for a wafer to go through in a fab. From
here the wafer can be taken to a diamond bladed saw that separates all
the processors on the wafer apart from each other. The bad dies are
thrown out and the good ones are packaged and shipped to a packaging
plant. AMD operates several such facilities in locations such as
PeopleÌs Republic of China, Malaysia, and Thailand. After that it is
sent off to a packaging facility where it is hermetically sealed (means
it is protected from outside) inside of a plastic package. When you
open up your computer and look at the chips on your motherboard or your
main processor you will see that it is black. Well
when I showed you a completed wafer it was silver, what happened? Well
that is the black plastic packaging material that you see. Inside of
that plastic are metal pins that will connect to the bumps on the chips. Sigh,
I still havenÌt explained about speeds and such and IÌm tired of typing
again. Well maybe another time. Oh IÌd also like to say that of all the
subjects I studied in school English was by far my weakest. So go easy
if you see grammatical mistakes and such. I used a spell checker so
that should be ok, but I canÌt say the same thing about my use of the
English language. Eskimo
[This message has been edited by Eskimo (edited September 30, 2000).] IP: Logged |
Jaddie Junior Member
|
posted 09-30-2000 02:33 AM ððð
ð ðð ðð
Dear Eskimo and FriendsIt's very generous of you to educate us as you have with your many words in this thread. Thank you. If
you win a pile of dough, you might direct an instructional video
depicting how the typical personal computer actually works. For
instance, I'd like to see an instruction, or set of instructions,
travel through the CPU and see what happens to those instructions at
each point along the way. I'd like to know what the processor sees as
we type on our keyboards, as we send email, as we manipulate images in
Photoshop, etc. Thanks again for sharing your knowledge. Sincerely, Jaddie Cocky MDJ Reader IP: Logged |
Will D Member
|
posted 09-30-2000 11:57 AM ððð
ðððð ðð ðð
I am not kidding Eskimo... That was the best information on the fabriaction of micro-processors I have ever seen. You reall should write "Microprocessor Fabrication Processes for Dummies" just like Tonton mentioned.
Great Job! IP: Logged |
JGuidroz Junior Member
|
posted 09-30-2000 12:40 PM ððð
ðððð ðð ðð
Eskimo,Louisiana
State University. We have two different types of machines for
depositing metal. And I know definetly one, but I'm pretty sure about
two machines for etching excess metal away. The professor said we also
had the equipment for using copper on chips(process IBM developed). IP: Logged |
seb Administrator
|
posted 09-30-2000 01:42 PM ððð
ðððð ðð ðð
Nice job Eskimo! Have you been to see the original transistor, developed by Bardeen et al, on display at Bell near RIT? I
haven't but I saw a documentary about it. Bardeen used a paper clip
pressing down on the top to apply just the right amount of pressure to
control a signal. Pretty cool. Evidently it still has the same bent out
of shape paperclip pressing down on it to this day. IP: Logged |
Eskimo Moderator
|
posted 09-30-2000 02:24 PM ððð
ðððð ðð ðð
quote: Originally posted by JGuidroz: Eskimo,Louisiana
State University. We have two different types of machines for
depositing metal. And I know definetly one, but I'm pretty sure about
two machines for etching excess metal away. The professor said we also
had the equipment for using copper on chips(process IBM developed).
Yes
generally metals are deposited by means of sputtering or evaporation.
So the two types of tools you have probally fit along those lines. CVD
(chemical vapor depostition) is the method of applying most other
layers such as polysilicon and nitride onto a wafer. If you can do
copper at your school that is very impressive indeed! I didn't disucss
it but wafers can be etched not only by using a plasma but also by
using liquid etchants. It just isn't as popular now of days due to the
fact it is hard to etch in just one direction. seb,
no I have't seen that. In fact I didn't even know it was up there!
Where exactly is that? Is that in Rochester or Syr. or what? I'll have
to go see that when I go back in November. IP: Logged |
MacAddict Member
|
posted 01-07-2001 07:59 PM ððð
ðððð ðð ðð
I have
to say: That was the best-written, easy-to-understand, most interesting
information on how processors are made! I love it! Now I can mutter
words like "fab" and "micron" without embarrasing myself!Beautiful! ------------------ d IP: Logged |
Bodhi Member
|
posted 02-07-2001 01:17 AM ððð
ð ðð ðð
Okay Eskimo, lets hear about speeds and how this happens!Awesome post! ------------------ ~~Bodhi~~ ~~Peace~~ Homer (reading Internet for Dummies): Wow... they've got the Internet on computers now? IP: Logged |
wyntir Member
|
posted 02-07-2001 02:55 AM ððð
ðððð ðð ðð
Man, Esk, that was amazing. Do yall do tours at the AMD fab down there in Texas? That rules so hardcore. Whew. I want another chapter! wyn ------------------ "Your reality, sir, is lies and balderdash and I'm delighted to say that I have no grasp of it whatsoever." Baron Munchausen from "The Adventures of Baron Munchausen" IP: Logged |
Leonis Member
|
posted 02-07-2001 03:24 AM ððð
ðððð ðð ðð
I just bookmarked this thread! We should nominate this thread "The thread of the year" [This message has been edited by Leonis (edited 02-07-2001).] IP: Logged |
P-Dub Member
|
posted 02-07-2001 11:07 AM ððð
ð ðð ðð
Pheeew, it's 3:07am, but I couldn't pull myself away.*sombre tone* Eskimo, you are The Man And pictures too !!!!!! Oh
and don't worry about your writing ability - it rawks!!. Seriously,
that was so well written - so easy to understand, yet so informative.
Good Sh!t !! When I grow up, I wanna work in a Fab! IP: Logged |
Unreg!stered Member
|
posted 02-08-2001 07:36 AM ððð
ðððð ðð ðð
Mmm...Cookies!!!...Yum.8)IP: Logged |
Bodhi Member
|
posted 02-13-2001 03:02 PM ððð
ð ðð ðð
Hey Eskimo!Speed! How is this done!?! Next chapter!! ------------------ ~~Bodhi~~ ~~Peace~~ Homer (reading Internet for Dummies): Wow... they've got the Internet on computers now? IP: Logged |
Eskimo Moderator
|
posted 02-13-2001 05:46 PM ððð
ðððð ðð ðð
Ok, i guess I gotta do this in 2 parts as AI seems to choke on my whole response here. Good thing I wrote this reponse in Word.Sorry
Bodhi, I've been sort of busy. Let's see how I can explain some of this
in a brief manner. Motoman or gEEk or one of the other EE's would be
better at explaining speed due to logic delays and such. What I know
about is the physical means by which a transistor operates. Chapter 8: You
can think of a transistor as a switch, as this is what it is often used
for in digital logic devices such as a processor. The most common type
of transistor in use today is called a MOSFET. This stands for Metal
Oxide Semiconductor Field Effect Transistor. Don't worry about the big
sounding name. Basically the first 3 words describe the physical
structure of the transistor. You have a layer of metal, under which
lies a layer of oxide, under which lies your silicon substrate. The
Field Effect transistor describes the way in which the transistor is
"turned on". I'll try to explain this further down. You can refer to
the figure below as I explain this. The silver is the Metal. The brown
is the oxide, and the blue is the semiconductor. When
I was taught this I liked the water analogy so I'll try to explain this
using that. There are two doped regions of the silicon (if you remember
doping involved putting foreign atoms into the silicon crystal
lattice). One region is called the 'source', the other is the 'drain'.
As I skipped talking about holes earlier I will make this a NMOS
transistor. So you have these big reservoirs (lakes)filled with a bunch
of extra electrons. Electrical current is a measure of the movement of
electrons in a given amount of time. Like measuring the water flow in a
pipe, how much water passes a point in a given amount of time. Now the
level of the water in each of these "lakes" is determined by the
electric potential. Electric potential is another term for voltage if
that helps. So if you have more potential on your source "lake" than on
your drain "lake" it is like having two lakes being next to each other,
but one at a higher elevation than the other. Like Lake Erie and Lake
Ontario. Now what happens when there is a path for water (electrons)
from one lake to flow to the other? Well you can get Niagara falls if
you aren't careful . The
path for our electrons to travel along is called the channel. The
distance that the channel extends from the drain to the source is
called the channel length. This is the smallest geometric length that
is built into a chip. Thus when a company is producing a chip at .18
micron it means that the distance between the source and drain is that
long, basically. Now
the channel is not normally present in the silicon. This is a good
thing or else we would have those crazy electrons taking off on their
own accord. We like to be able to control things.
So if there is no channel and there is no flow of electrons then we can
say that the transistor is OFF, or you could say it represents a 0, or
FALSE. Now maybe you are starting to see the basics of why a computer
thinks in 1's and 0's. So
how do we get a 1? Well we need to form the channel. We have our gate
in the center of our diagram. You can picture this as a valve on the
dam separating our two "lakes". To open up this valve we only need to
apply a small amount of voltage. When this voltage gets to a certain
point (known as Vt, the threshold voltage) it opens up the channel. It
does so because when you have a potential difference across a
dielectric (a sort of insulator) an electric field is formed. As this
electric field grows strong enough it is able to suck up some of the
electrons stuck in the silicon substrate and bring them near the
surface. The electrons don't have as much trouble traveling through
other electrons as they do plain old silicon. So they are free to try
to balance out the voltage differences between the source and drain by
flowing causing current. The
design engineers design their circuits so that when a certain level of
current flow is achieved they can consider that transistor to be ON, or
a 1, or TRUE.
IP: Logged |
Eskimo Moderator
|
posted 02-13-2001 05:47 PM ððð
ðððð ðð ðð
Chapter 8, Part deux:Now
how can we improve the speed at which this happens? Well for one, we
can reduce the length of the channel so that our electrons don't have
to travel as far. This is why when a semiconductor manufacturer
'shrinks' their process their chips can run faster. A .18 micron G4 can
run faster than a .20 micron G4. Another
way we can do this is by changing the level of the electrons in the
source and drain so that their is a greater tendency for them to flow.
This is like raising the level of one of our lakes. This can be done by
increasing the voltage difference between the two (a common tool of the
overclocker ) or it can be done by introducing different levels of dopants into the source/drain region. Another
method to increase the speed is to reduce the voltage at which the
transistor can turn. We can't increase the gate voltage totally
instantaneously, it takes a little time to do so. If we lower the
voltage we lower the time needed. To enable a lower threshold voltage
we can make the oxide between the gate and the semiconductor smaller.
This way it takes a smaller electric field. The problem with this is
that our oxides are getting VERY VERY thin. Right now Intel, AMD,
Motorola, and IBM are talking about putting down a layer of oxide only
3-4 ATOMS thick!!! This is incredibly hard to do. So what we are trying
to do is find a good material that blocks the electric field slightly
less than silicon dioxide. If we can settle on a good material to do
this than we can make thicker gate oxides out of this material. These
are referred to as "low k dielectrics". The only problem is that every
company and their brother claim to have found the easiest and best low
k material. Now we have to sift through all of them and figure out
which will do a good job and be as easy to work with as silicon dioxide
is. Now if an
electron is just minding it's own business traveling along in the
channel and suddenly hits the bottom it may become snagged. This would
lower your current and slow down your transistor. Some electrons do end
up doing this and this is known as leakage current. To try to reduce
leakage current one recent innovation is known as Silicon On Insulator,
or SOI. By putting some oxide right beneath the silicon surface the
electrons are less likely to stray from their path. With less leakage
current you don't have to force as many electrons through and you don't
lose as many. This makes chips run cooler AND faster. Not too shabby
(Thanks IBM! ). Now
we haven't talked to much about it but there is a whole network of
connections sort of like roads running over the transistors in order to
connect them. If our signals can't pass effectively through these
highways then it won't matter how fast our transistors can go. Some
things done to improve this is the use of alternate metals (such as
copper instead of aluminum). Another factor is that two conductors
traveling near each other will form a capacitor of sorts between
themselves. Capacitance will slow down our ability to pass electrical
signals. We try to isolate them with silicon dioxide. But as we make
these wires tinier and tinier they get closer and closer making it hard
to reduce this capacitance. So research now is being done looking into
high 'k' dielectric materials by which to separate the metals with. Well
I hope that answers some of your questions. The actual design of how
these transistors are used also has a huge effect on determining how
fast a chip can go. But when you hear of chips that aren't yielding or
are only running at say 500MHz even though they are designed to go
faster it could very well be because something mentioned above wasn't
produced exactly right.
IP: Logged |
Eskimo Moderator
|
posted 02-13-2001 05:52 PM ððð
ðððð ðð ðð
On a side note, this thread didn't even get one mention for "AI Thread of the Year" I thought it was a good thread
IP: Logged |
Alcimedes Junior Member
|
posted 02-13-2001 07:09 PM ððð
ðððð ðð ðð
hey Eskimo, now that i have a chance, could you answer one question i asked a bit earlier?is
there some reason why the wafers always come out as a circle? in all
the pictures, you can see chips that are cut by the curvature of the
circle. is there some reason why the wafers always come off of a
cylinder, instead of a square shape? is it possible to have the seed crystal grow into a rectangular shape, instead of a cylindrical one? -alcimedes IP: Logged |
Eskimo Moderator
|
posted 02-13-2001 07:17 PM ððð
ðððð ðð ðð
Unfortunately
we can't really do that Alchimedes. The reason is that due to the way
in which the silicon cools that the seed crystal and melted silicon
must be rotated as it is pulled out. The only shape you can really
rotate easily is a circle. We
COULD saw the wafers to make them square while they are still in ingot
form. But this actually leads to more trouble than it is worth because
it allows the corners and edges to be more easily damaged or chipped
during processing. And once a crack starts in a silicon wafer it
travels through the whole length of the wafer due to it's crystalline
structure, ruining the whole thing. Also all wafer handling machinery
is built to handle circular wafers so there is little incentive to
change the shape. While we could probably fit a few more die on the
wafers as we reach larger and larger diameter wafers the percentage of
"edge die" that aren't whole decreases quite a bit. IP: Logged | |