§1. Solders, creep, and the sprinkler-system case [00:03]
I want to finish up soldering and then go into microelectronics joining. What I haven't mentioned to you is this particular book, Soldering in Electronics, second edition, by Klein Wassink. It's published by the Electrochemical Society. I haven't looked to see if they've got a newer version — Klein Wassink was a researcher at Philips in the Netherlands, this is 1989, so he may be well retired by now — but if you've got an extra $300 or $400, this is a great book to have.
It actually has more details on solders and soldering from a scientific point of view than anything else. It's got creep data, it's got all kinds of stuff. Which is something I haven't mentioned, about another difference between soldering and brazing — this artificial difference — and that is, solders are always in the creep regime. If you just look at the temperature below 450 Kelvin, and they're at operating, if they're at a little bit elevated temperature like 100° centigrade, they're going to be operating at six-tenths of the absolute melting point, and so they're in the creep regime.
As a rule — and this is a very rough rule of thumb — I find that I can do a quick in-my-head calculation: if the stress on a solder joint is 1,000 PSI or greater, no good. Typically you don't want to see stresses more than 100 PSI or a couple hundred PSI, which is one megapascal, 145 PSI. So you don't like high stresses on solders. In fact, sprinkler systems — if you look up here, you've got this little white circle — if we had a fire here, that would pop down and start spraying. I don't know if it'd be a glass-bulb type of sprinkler system, or the old Wood's metal. Wood's metal melts at 165° Fahrenheit, and it would essentially melt and start spraying the room with water.
There's been a big controversy because historically, going back a hundred years, people have said don't use Wood's metal sprinkler systems, don't let them get above 100° Fahrenheit, because if you do, they will fail, and they fail by a creep mechanism. Wood's metal is just a solder-type compound, low-melting metal, and solders fail at low temperatures.
§2. Contaminants in the lead-tin solder bath [02:55]
I just handed out this summary which comes out of a book by a guy named Manko, who wrote a number of books on soldering. This is a summary of what's good and what's bad in lead-tin solder baths. Now maybe this is getting a little bit old, but people figure, oh, it's solder, it's not a big deal. I've told you about some of the things that are critical. The composition is critical. Remember I talked about wave soldering, the way we solder printed circuit boards.
[Tom puts up the wave-solder picture.] Here's the wave solder bath. You have a printed circuit board, you mount your components on here, and it used to be you always had all your little wires sticking all the way through. We're going to show you that's not the way you do it anymore — you actually take the electronic components and they have to survive the heat of the solder bath. You spray on a flux and you pass it across this wave of solder. This bath down here will slowly pick up contaminants. That might be a bath of several tons of solder that you're pumping continually, but you've got copper up here, you could have other metals on the components.
If you pick up very much contamination, you find that 50 parts per million of aluminum, magnesium, or zinc — zinc could be in a lot of things, we use it for corrosion protection, but it would easily get into that solder bath — will form intermetallics. Those intermetallics start to act like sand in the solder bath. They get in and interfere with the joint and the wetting and flow of the solder. We add up to three-tenths of a percent of antimony. Today we use antimony-silver solders, primarily antimony because it improves wetting. Antimony, like lead oxide, is a fairly stable oxide, and it will actually help do some of the cleaning off the surface.
However, you read a bunch of old specs and it will say you have to put the antimony in to avoid tin pest. Tin pest is the fact that pure tin will transform from white tin to gray tin with a maximum transformation temperature of -18° centigrade. It was once thought — turns out it's not true — that when it goes from monoclinic to triclinic, your tin will turn to powder.
It happened to the Russians in World War II. They had a big stockpile of tin outdoors that they were storing, and the winter came along, everything got covered with snow, and in the spring when the snow melted, they found they had a pile of not tin ingots but tin powder, because it got to minus 18° centigrade, or thereabouts, which is the maximum transformation temperature. This question was on my doctoral exam, by the way — not tin pest, but the maximum transformation temperature of gray tin to white tin, and going through the kinetics. If I'm a Russian general, I'm just irritated I've got a bunch of tin oxide now rather than tin ingots. But I could calculate it at that time in my life — at least I guess I could, I don't remember if I could or not. I got a pass, who cares.
So it just transformed, and people thought that this would happen in the lead-tin alloys. It turns out not only does antimony inhibit it, but lead inhibits it, and it doesn't take very much to slow down that reaction. If you look at the lead-tin phase diagram, there's some solubility for lead in tin, so that's not the reason to put antimony in, but it does improve wetting, and nowadays we use it to improve strength of the solders.
Arsenic and iron — it might be 500 parts per million or 200 parts per million — form intermetallics. Bismuth, just like antimony, can help wetting; destroys recyclability of steel, but aside from that it's not a problem. Cadmium, a lot of people think, oh, it's cad-plated, no problem in soldering. Turns out 10 parts per million in lead or tin segregates in the solder bath and creates dross formation — all these oxides floating on the surface which interfere with the wetting.
You can have up to 800 parts per million copper or gold. Copper you don't care so much, although if you get above that, you actually start forming some copper-tin intermetallics and you do care. Gold — you idiot, if you've got 800 parts per million gold, send it out to the refiner and get your money back. You will slowly dissolve the gold on some of these integrated circuits, but you usually don't ever let the gold get that high. Nickel is rarely found but it's not really harmful. Silver we sometimes add to the solder bath but it's expensive.
Sulfur, seven parts per million, forms a copper sulfide. The copper sulfide is more stable than the copper oxide. Most of the copper ores we find in the world are sulfide ores, not oxides. Since copper sulfide is more stable than copper oxide, guess what — a lot of the fluxes, and particularly a non-activated rosin flux, won't clean off copper sulfide. So small amounts of copper sulfide just destroy wettability. This goes not just for lead-tin alloys but for some of the other things we use in soldering.
So even though you have this wave soldering system, about once a week or so, you take that two tons of solder, pour it out of the bath, solidify it, send it to the refinery and put some new solder in. If you're in a big shop doing wave soldering, you're doing thousands upon thousands of printed circuit boards a week, so it's not that big a cost to recycle it, because if you don't, you're going to find all kinds of defect problems showing up.
Student: How do they recycle it, how do they pull —
They go back to the way you clean up lead and tin to begin with. You don't have to go all the way back to the ore process. Actually, I'd have to ask Harold Larson how they would clean some of these things up. You're not going to put sulfur in there. Some of them you can oxidize out, like aluminum and magnesium, and you can filter them out — it's a low-enough-melting liquid that you can filter out these intermetallics and the oxides. So you could bubble some oxygen through it, I'm not sure if that's exactly what they do. I don't think you'd be using chlorides in most cases, but another way to clean aluminum out of a system is to bubble chlorine through it, through molten aluminum.
We used to do it all the time in the foundry — just exhaust it to the Cambridge air. You bubble chlorine through the bath and what comes up is bubbles of chlorine with hydrogen chloride gas. That was back in the days when you could pollute, it was all acceptable. Nowadays if they thought there was a hydrogen chloride molecule, even a single molecule, you would probably be threatened with a jail term for polluting. Back in the old days we just bubbled chlorine all the time through the aluminum to clean it up.
§3. The bond number and the size of a stable drop [11:26]
There is some other science you ought to be aware of, and this concept of a bond number, which I apply to soldering, but it applies to welding spatter, when you're welding steel or aluminum or anything else. Remember, metals have higher surface energies than other things. The bond number applies to any types of particles, even small aerosol particles if you sneeze. What size particle is stable, and what will form spheres and what will form egg-shaped drops?
The bond number is a dimensionless number. Are all of you familiar with dimensionless numbers? I know you are because you're a mechanical engineer, you are because you're a mechanical engineer. What about you materials folks — dimensionless numbers like the Reynolds number, the Peclet number? Dimensionless numbers are just an energy over an energy or a force over a force. In this case we're going to take the gravitational energy over the surface tension force. You can do it as forces or as energies. That's called the bond number — has nothing to do with joining, just happens to be called the bond number.
You take some sphere that has some height above the surface, and you want to know if that's going to be spherical or if it's going to be an oblate spheroid, elliptical in shape. The gravitational pressure is ρgh — that's how much pressure down here from the weight of the metal up above, and ρ is the density of the metal, which is pretty high. The surface tension pressure is 2γ/R, if it's a sphere. If it's an ellipse it's not a single value of R, but they're close to the same value — this is just getting dimensional. R is about h/2, and if you go through all this and take the ratio, you get the bond number is ρgR²/γ.
So what? Surface tension or surface energy: if you put in one joule per meter squared, which is 1,000 dynes per cm squared, you put in a density of about 10 g per cubic cm, you put in gravity which is 980 dynes per cm, and you want to solve for a radius. If the bond number is one, at what value are the gravitational energy and the surface energy equal? That's when it's going to start to slump because of gravity. If you put in those numbers, you get R² equals a tenth, square root of a tenth is 0.3 cm, which is 3 mm. If you want to be exact, it's 3.16 mm — but I didn't use even one significant figure in doing my calculation, so I can hardly use three later.
So 3 mm or an eighth of an inch is sort of where you would start to think your solder below that should form nice symmetric geometric shapes when it's bonded on a circuit board, and above that it's going to start to slump. This comes out of the Klein Wassink book — these are joints they made on the old-fashioned wire-through-a-hole. Over here they didn't have enough solder, and they kept putting increasing amounts of solder, and they start to get a decent fillet right here. They put too much solder on and they start getting the whole thing clumping up, and depending on the size of this it may still be surface tension at this point.
But you can get to the point, when you're dealing with molten metal powders, whether it's spatter or doing powder metallurgy, that if it's less than an eighth of an inch in diameter, it probably is going to form a sphere naturally, and if it's larger than that it's going to slump and gravity is going to be the dominant force. Provided there's not — in welding, the drops come off with some velocity and they hit and they — it's actually called spatter when it comes off; once it hits some solid object it's called splatter. It splatters all over everywhere. Just like dropping a cat from a 100-story building — I guess they land on their feet, but who cares at that point anyway.
Actually, I don't know if I should say this on tape — I tried it once with a cat to see how quickly they can turn. It turns out it only takes about 8 or 10 inches and they can flip themselves around if you hold them upside down. [laughter] I was doing it on a sofa, so the cat really didn't mind, he wasn't upset.
Student: Pardon me?
I did do it several times to see how he basically turns. He rotates his torso, because he has no angular momentum, right, and he has to create angular momentum. He actually turns his rear legs this way to hit the ground and he turns the other way with his upper torso, because he has to conserve angular momentum. I'm sure he worked out all the physics equations to do that, but I was curious. Once they do land on their feet, in any case.
§4. Levels of interconnect and the copper-on-chip transition [17:35]
So now let's talk about soldering for microelectronics or interconnect. The folks doing interconnect, putting together electronic circuits, actually have a system for what they call the interconnect level. If I'm trying to make joints on the chip, on the piece of silicon — so I'm putting little aluminum or copper conductors on this semiconductor chip — that's called level zero. It's called level zero because the guys making printed circuit boards and assembling them probably forgot about that level of interconnection, so they went back to zero, where they found someone said, oh yeah, we make joints.
Does anybody know what we're using in the best electronic circuits now as the interconnect metal on the chip? It was a big announcement about 15 years ago by IBM, hit all the front pages of the papers around the world. They used to use aluminum, and the reason they used aluminum — aside from the fact it's easy to vapor-deposit aluminum onto a chip through the polymer mask, so you could vapor-deposit it and make little lines, connections between the little transistors — was because a little bit of aluminum contamination didn't destroy the semiconductor properties of silicon.
They would have loved to use copper, but they couldn't until probably the late '90s. IBM announced they had come up with this mystery bonding layer or inter-diffusion layer that would prevent inter-diffusion of copper into the silicon, so it wouldn't destroy the semiconductor properties. The mystery material turns out to be tungsten. Why would you use tungsten? You put on a little layer of tungsten — very heavy element, very high melting point, diffuses very slowly. This was a big secret for about 3 or 4 months until a couple of people who knew something about diffusion thought about it, said, well, it's obvious, you use tungsten. That's in fact what they used. But now they use copper.
Why do they use copper? Because back in the mid '90s — if you go back and look at how fast the semiconductor chip is — right now if you go buy a new laptop, it'll have an Intel chip in it, and it will be running at what frequency? A variable, but around what frequency? Three gigahertz, good number, I like it. Back in 1990 or so, these things were running at one-and-a-half, two gigahertz, and if you went from the 1980s up through the '90s you would just pay more to get higher speed. Does anyone know why we've sort of pooped out for the last 10 or 15 years, mired at 3 GHz?
We poop out because of the speed of light. If the speed of light is 3 × 10⁸ m/second, and wavelength is the speed of light times time, I do 3 × 10⁸ divided by 3 GHz, which is one over time, I get 0.1 m, 10 cm. That's how far light can travel at 3 GHz. You might say, that's a lot bigger than the chip. It is bigger than the chip, but there are things that slow that down. That's if electrons were flowing through the metal at the speed of light, and they don't actually go at the speed of light — they might go at a tenth of that. So instead of 10 cm, they may only go 1 cm, or maybe they go 2 cm, because of parasitic capacitance and resistance.
One of the things they do on chips nowadays is they almost always have a metallic back plane. This one has a layer of gold on it. They want to have that as a ground plane to prevent parasitic capacitance, because it will slow these things down. You don't usually have to worry about inductance at this size scale, but you do have to worry about capacitance. So another reason — not just thermal expansion — for making the chips only one centimeter in size, and I told you about bonding into dissimilar materials, but it's also: you make it much bigger, you've got to start worrying about all these speeds and how the electrons from one end of the chip get over to the other end to do their work.
Back at one point I told you the Pentium 6 had two chips on a tab tape instead of just one. One was about a centimeter square, the other was about six-tenths of a centimeter square, and I think the other was basically a bunch of memory. This was back around 2000, or late '90s. Nowadays they can get so many transistors on a single chip — we're up to almost 100 million transistors on a chip now — they can cram everything onto one chip. But in the old days, which is about the time you were born, they had a problem because the distances the electrons had to go were getting into a range that was limiting the speed.
What they do now is the chips have enough components on them that they can do a lot of their work on the chip and then just pass other information to something else. Now they have parallel processors. The world supercomputers are bigger than the size of this room, and they still have to communicate not just over 10 cm but over tens of meters from one chip to another. They might have 64,000 computer chips and cost billions of dollars, but they still have to communicate. How do you keep track of all these separate little parallel-processing calculations and bring them together to end up with something very quickly is an architect's nightmare. That's what electrical engineers do in that area of the business.
§5. Rent's rule, perimeter bonding, and the ceramic-package era [25:05]
In the old days, people used to worry about how many input-outputs you could get on the chip, and they still do, to a certain extent. There was something called Rent's rule, but I have not heard about it for about 12 years. Rent's rule was sort of like Moore's law — Moore's law says the cost of the semiconductor transistors goes down a factor of two every year and a half, 18 months. Rent's rule said the number of inputs and outputs on a semiconductor chip increased with the number of transistors on the chip to some power: number of circuits goes as capital N to the n power, where n was somewhere between 1.5 and 3, with 1.79 being the best fit. Well, three significant figures, that's great.
If you increase the number of transistors by a factor of 100, you'd have 10 times as many inputs and outputs. [Tom holds up a ceramic semiconductor package from the late 1970s / early 1980s.] If you look in the old days — the old days being when I was in my youth — you might have a semiconductor package that looks like this, from the late '70s, early '80s, where you have a little semiconductor that goes in here. You could count these on here, but it probably doesn't have more than three dozen connections, inputs and outputs. The semiconductor is in there, and this is a piece of 97% aluminum oxide filled with a bunch of gold or tungsten vias inside. This is a composite, a metal-ceramic composite from the 1980s.
This is your number of inputs and outputs that you're going to push onto a printed circuit board. Look at how much real estate that takes. They do call it real estate on the printed circuit board. That's 30-year-old, 40-year-old technology. [Tom shows a perimeter-bonded package.] Then people came up with little square things like this that had — that's called perimeter bonding, this is area bonding — so you'd have a chip that fits in about the same size as that, and you'd have the little wire bonds that would go from the chip onto the aluminum oxide. There's actually a ground plane on the back here, but it's a millimeter away. You couldn't have that today. This is probably mid-'80s type of technology, 286 computer or something.
You have little wire bonds on here, and you look and say that's pretty close. Even for a wire bond, I'll talk about how close. On the back side they actually have a bunch of little dots — that's a perimeter array, where you bond actually not even a square perimeter, two lines. This is an area array, and this is basically a bunch of pins — pin grid array. You can see they have a bunch of wire bonds on here, and then they have the connections, and you would just plug this in, turn it upside down and plug it into the printed circuit board. That was all well and good until the mid '90s or 2000.
Because of miniaturization of laptops and cell phones, they couldn't deal with electronic packages where your whole component took up a bunch of real estate because of the aluminum oxide housing. Cell phones needed to be miniaturized. I wish I'd kept my old first cell phone. The old first cell phone wasn't a shoebox, but it was probably about that square and about that long. You could fit it in your pocket but it looked like you were carrying around a can of soda in your pocket, because it's about the size of a soda can. And it had range — from here to the building next door. It was great. If you walked underneath a power line, well, forget that conversation, I'm talking about dropping the conversation. Cell phones have improved.
It cost me $3,500. The first cell phone I ever had — a guy from Motorola, a vice president of Motorola, was visiting, I had him over for dinner, and he had one of the first cell phones, and at that time if you could buy one they cost about $10,000. This was about 1990.
§6. Tab bonding, Intel, and chip-on-board [30:04]
[Tom passes around a tab-bonded chip.] Here we have what they call tab bonding, tape automated bonding. It actually has about 400 inputs and outputs around the perimeter. What they did — the reason Intel would do this — you have what you call inner lead bonds, which Intel would make and make the chip to the tab tape, and they would sell you this little cartridge with the tab tape. By the way, about 1995 when Intel gave me these things, the tab tape cost about $50 a piece. I had a bunch of high school students coming in for a Saturday at MIT, and I called up a friend who was one of the — his badge number was less than 1,000 at Intel, and your badge number was your hire date. So he was one of the first 1,000 people. He had taken me over to meet Craig Barrett, who became the CEO. Craig Barrett was a materials science professor at Stanford who left in his youth to go to this company called Intel that Gordon Moore was founding.
So Gene Meieran, who's a graduate of this department and is an Intel fellow — I was down at Intel in Chandler, Arizona, he said, I want you to meet Craig Barrett, who at the time was president of Intel. We go rushing through all these buildings about three-quarters of a mile, and we get to Craig Barrett's office, and he had a cubicle just like all the other people with cubicles, and his secretary had a cubicle. His cubicle was about 50% larger than the average person's cubicle, but this was Intel's culture at the time. They did have conference rooms, but everybody could use them for meetings of two or three people, so you could have a private room if you were going to fire somebody. So I met Craig Barrett at that time.
I called up Gene when I was department head, and I needed to have some touchy-feelies to give the students. We ended up giving them a turbine blade — I got Pratt & Whitney to give me a couple hundred turbine blades that were old turbine blades — and I got Intel to give me some of these, and there were about 500 of them. I didn't realize at the time, at $50 a pop for the tape. These were actually test pieces, the ones that don't have a chip. I only have one that remains that actually has a bad Pentium 5 chip — it's got a little dot on it. Intel would make the inner lead bonds, and you could make your own outer lead bonds directly on a printed circuit board. This was called chip-on-board. Chip-on-board technology allowed people to get rid of all those aluminum oxide packages, and they don't really exist much anymore except in kind of low-end electronics, because they're expensive to make those packages in general.
[Tom holds up a printed circuit board.] So the packaging — when you're bonding from the chip to the package, all those things I passed around, I'm kind of giving you a history — and then you go from the package to the printed circuit board. You know what a printed circuit board looks like. This is a printed circuit board, it's got 21 layers, and it's basically a bunch of plastic sheets that have different patterns on them. You have to put them together in the right array, and you press them together and adhesively bond them in a vacuum furnace under pressure. So there's a big adhesive bonding problem. You also have to do some things to make connections between each one of these holes. And you can put your chips directly on there without those big packages.
[Tom holds up a smaller board.] A smaller one looks like this — this is from about 1997, Cisco Systems says on here. If you're trying to run the internet, this is going to go through one of these. They used to make these up at a little plant in New Hampshire. I had a student work up there, and they had lots of quality control problems, because you make something this complex — this one's not 21 thick, this is probably only 13 or 14 layers thick — but can you imagine how many tens or hundreds of thousands of dollars of electronic components are going to go on here, and every single one of these vias, every single one of these holes, has to make a connection. If it doesn't, you've got a horrendous quality control problem. How do you repair it, and you really don't want to throw it out, after you've put everything on it.
This is an excellent example of where I told you joining often comes late in the process of the manufacturing of a component, when most of the value is already in the part. They had automatic scanners of each one of these sheets. I should have brought one of the sheets — they're not Mylar, but one of the little sheets is kind of flexible, and they'd stack these up. I do have one back in my office that I'll try to remember to bring next time. By the mid '90s we're getting rid of the aluminum oxide packages, we're going to chip-on-board, and we're getting things smaller and smaller. The number of inputs and outputs is going up.
Except it didn't keep going up. Back in the early '90s, the defense department made some 2-inch chips that were 1 to 2 inches on a side, where they were integrating everything all on one chip to beat this time limitation of the gigahertz, because it still takes time, you've got parasitic capacitances. You can't go 3 GHz from one corner of this board to the other corner. If you're going to be operating at 3 GHz, you've got to do everything in a very small area. The classified military chips at the time were manufactured at Motorola, who had a facility that was about 10% of their business, and all they did was make classified military electronics, special chips for the military.
§7. Levels of interconnect and AMP [36:48]
Let's finish telling you about the different levels of interconnect. We talked about, on the chip — we now use copper because of its lower resistance, so we can go to higher frequencies. We wouldn't be getting 3 GHz today if IBM hadn't figured out how to put a layer of tungsten down as a diffusion barrier for copper. We would still be using aluminum. Package to the printed board — well, we don't really do that. We actually put the chip directly on, so we sort of eliminated level one, because we go directly from the chip to the printed circuit board, which is level two.
The board to the subsystem — that's within a cabinet. You've got an audio amp, you buy it and you put it into your audio system. That's that little thing that's 18 inches across by 10 inches deep by 3 or 4 inches high. You've got a bunch of interconnects in there, and these things plug into little connectors. There's a company called AMP, anybody heard of AMP, out of Harrisburg, Pennsylvania? You ever heard of the Whitakers — Whitaker College here at MIT? Uncas Whitaker started a company called AMP, the world's largest manufacturer of level-three packaging — all those little plug-in connectors. You pull a printed circuit board out nowadays, if you're lucky enough to have a repairman come in — all he does is, he's got a full set of circuit boards, he just pulls them out until the thing starts working. When it starts working again, he leaves that board with you, charges you $5,000 for the replacement board, takes the other board back to the factory and someone repairs it. Then they sell it to somebody else for $5,000.
Subsystem to the system — you put a bunch of these into a relay rack, which stands tall as a person and has a whole bunch of these things. Or it could be cables running from this part of the supercomputer to that side, that part of the supercomputer. Now you've got all kinds of complex cables. If you want, you can think of an automobile — you've got all kinds of electronics that have to communicate with each other.
Automobiles today, about 25% of the cost of an automobile is what?
Student: Healthcare.
A third of the cost of an automobile is electronics. Ten percent of the cost of an automobile is steel — about $500 for the steel that goes into an automobile. Sort of amazing. So those are the different levels of interconnect technology.
[Tom holds up a textbook, Introduction to Surface Mount Technology.] I didn't bring the book with me — there's a book that's getting a little bit old now — but anyway, this is Introduction to Surface Mount Technology, which is chipped directly on the printed circuit board. Through-hole, which is that DIP package, that dual-inline — the first one I handed out with two rows of a couple of dozen things — that's a dual-inline package. PGA, pin grid array — I passed around that thing with a bunch of pins, that's a pin grid array, that's about early 1980s. Then there's multi-chip modules, which had these little gold wings coming off the side, but they're still in these aluminum oxide packages. There's flip-chip — I passed around a flip-chip — which is around 2000. That's as far as this book went, because really after 2000 you get to chip-on-board.
§8. The transistor's origin and area-array bonding [41:13]
Whether you're doing chip-on-board or putting into an aluminum oxide package, you still have to worry about perimeter bonding or area-array bonding. People were worried about this for the last 40 or 50 years, because IBM would predict and say, where are we going to be? In fact, that's why we have a transistor.
In about 1925, AT&T, which was called Bell Laboratories back then, the Bell System, was looking at all the mechanical switches they had. People always say that the transistor came out of basic research. No — AT&T in about 1925 was looking at how many mechanical switches they had in their system, and they started something, a reliability group, a bunch of statisticians who were looking at how to make the system more efficient. They projected that sometime by about 1960, 35 years hence, the whole system would come to a screeching halt because the reliability of mechanical switches was about 1 in 10,000 per 10 years. They call it FITs, failures in time. If you're in the semiconductor business, they talk about FITs, failures in time.
It was these guys in the 1920s and 1930s at Bell Labs that were looking at this type of stuff, and they came to the conclusion they needed to come up with an electronic switch, because the mechanical switches were not going to have the reliability 30, 40 years hence. Bardeen and Shockley invented the transistor not because they were doing basic research, but because they were trying to come up with an electronic switch. That's exactly what they were doing — applied research. So forget all this basic research stuff about the transistor coming out of basic research.
IBM was looking at the same things in the 1960s, when you only had a thousand transistors on a chip. So you've got a 1 cm chip, four-tenths of an inch in diameter. If you're doing wire bonding, the closest you can get together of these little dots around the outside is about 7 thousandths, because the wire is 1 thousandth, but that little needle head — I told you it looks like a sewing machine needle — has to have about three-and-a-half thousandths on either side. With four sides you can get 230 inputs and outputs around this thing.
A tab-bonded tape like the one you have is about 360 rather than 400, but if you've got 4 cm on the perimeter, and you've got 4 thousandths of an inch, 100 microns between them, you basically have 400. That was all you could really get unless you went to area bonding, which the military did in the early '90s. But now everybody does area or array bonding, invented by IBM in the 1960s, never used because it wasn't needed.
IBM saw they were going to need more, and some folks got together and said, let's do C4 — controlled collapse chip connection. You'd have these little solder pads, just like that thing I passed around, which didn't have 2,000 of them on it, but your connections will be on these little pads, and you'll have a solder ball. You can get up to 2,000 theoretically before you start running into thermal expansion problems, of these things just shearing themselves apart. That's the way most of the chips are done today — not for 2,000 inputs and outputs, because I think Rent's rule has sort of pooped out. When we can get 100 million transistors on a chip, we don't need as many inputs and outputs.
I used to say, back when I first learned about Rent's rule 15 or 20 years ago, that it can't last forever, because at some point all you want is about 10 inputs and outputs — everything will be on one chip and so you're not going to need 5,000 inputs and outputs. Other people said, oh no, you're going to need 5,000 inputs. I said, that's ridiculous — at some point you go over the peak and you start needing fewer inputs and outputs. So there's area bonding and perimeter bonding.
[Tom shows a wire-bonding tool.] Here's the little sewing machine for wire bonding. I think I've shown you something like this before. This was ultrasonic — it's talking about a sonotrode, so it's ultrasonic, says ultrasonic wedge bonding. It makes something on one little pad here and then makes another bond here. These are the types of bonds — about 50 trillion of these are made a year, is what I estimated about 10 years ago.
The reliability has to be huge, because one bad joint out of 400 — actually one bad joint out of 800, because there's an inner lead bond and outer lead bond — one out of 800 in the whole thing is junk. You might try to repair them, but it's not that easy to repair things that are only 100 microns across. People do it when you're talking about a $500 chip, but if you're talking about a $50 chip you can't afford to do it.
Here's the encapsulation I was talking about. You basically have to protect that chip from any corrosion and moisture. No other questions? Probably a good stopping point, because I'm starting to see the vultures out there.