CS_F2012_02

Codes and Standards Fall 2012 Session · 6 sections 5 cases · Watch on YouTube ↗ all files
Layer 3 — readable edition

§1. LIGO: the most precise measurement ever attempted [00:01]

§1.p1

I have a short article on the history of codes and standards — lightning codes and standards. I'll hand it out because we talked about CSST and how lightning can perforate it.

§1.p2

Any questions? If not, I'll go back to what we were talking about, which was observables. We were going from the sublime to what I call the ridiculous. The sublime is the NIST F1 atomic clock — one part in 10 to the 15th accuracy, which is one second in 30 million years. So we have a pretty good handle on time today.

§1.p3

But I want to give you a case study on the most accurate measurement ever attempted: the laser interferometric gravitational observatory. LIGO. It's a joint MIT-Caltech project. Back in 1992, Ray Weiss, a scientist in the MIT physics department who had started as a technician and then got his PhD in physics, proposed to the National Science Foundation, together with two people from Caltech, what has become the largest NSF project in terms of cost ever attempted. They built two interferometers.

§1.p4

If you remember the Michelson-Morley experiment from freshman physics, they had a little interferometer where you split light on two perpendicular axes, send part of the light down one way and part the other, with mirrors at each end. The beams come back and you see the beating frequency, and you can measure the length very precisely along the interferometer arms.

§1.p5

These interferometers are 4 km long. One is in Hanford, Washington — which is a desert — and the other is in Lafayette, Louisiana, surrounded by a swamp. It's about a half-billion-dollar project. You can't take a picture of the whole thing except from space.

§1.p6

I read an article in Science saying this was supposed to measure one part in 10 to the 23rd. I'd worked with Ray on the welding of the beam tubes — about a 4-foot diameter tube, 4 km long. He came down to my office to explain this, and in his own handwriting wrote down what they were trying to measure: 10 to the minus 23rd per square root of hertz.

§1.p7

Anybody have an idea why square root of hertz? It's how frequently you sample the measurement. You could call it Nyquist frequency, but whenever you're dealing with a Gaussian distribution — remember the precision plot, the bell curve — you're trying to make a measurement. Here they're trying to measure gravity waves from space to confirm Einstein's general theory of relativity.

§1.p8

Gravity waves are created when a supernova occurs, sending waves out through space that change the weight of everything. So you're gaining and losing weight twenty times a second. If you could just combine all the weight loss and not the weight gain — anyway. What they do is square root of hertz because if you sample for a longer period, you get better accuracy; if you sample for a shorter period, you get more measurements. There's a trade-off. If you sample over too long a period, the waves average out.

§1.p9

So Ray said the length is 4 km — he wrote it as meters, he's a physicist, he doesn't put units down. The change in length you're trying to measure, if it's 10 to the minus 23rd over 4 km, is 4 times 10 to the minus 20 meters per square root of hertz. If you sample ten times a second, you lose threefold in accuracy because you're sampling faster than a second. So it's 1 times 10 to the minus 19 meters that you're trying to measure.

§1.p10

They have 12-inch diameter tungsten spheres with mirrors on them in the building where the two lines come together, and a mirror at the other end. They had to align these tubes within about 3/16 of an inch over 4 km. No one had ever surveyed anything that accurately. Civil engineers didn't know how to do it — at 4 km, you actually have to start accounting for the curvature of the earth.

§1.p11

They ended up using the GPS system with their own base stations to increase the accuracy. Civil engineers use GPS for surveying all the time now, but they had to develop the technique to even align these things on their foundations.

§1.p12

They had to do it in vacuum, because light changes speed in a vacuum. They're down at about 10 to the minus 12 atmospheres. The primary impurity is hydrogen, and one of the things they worked on — they had undergraduates working on this — was developing the right way to oxidize the surface of the stainless steel vacuum chamber so it wouldn't outgas hydrogen. At 10 to the minus 12 atmospheres, when you're dealing with 10 to the minus 19, it doesn't take much contamination to throw you off.

§1.p13

There's a little helium in there too. If they had done it the conventional vacuum technology way, they would have spent twenty or thirty million dollars a year just in energy cost pulling the vacuum. So they had to develop a vacuum system so hermetically sealed that they could do it with a couple of vacuum pumps that would fit in this room — for the whole beam tube. That was a whole new technology. So they pushed the limits on surveying, vacuum technology, and outgassing of surfaces.

§1.p14

This is when I learned, fifteen or twenty years ago, that some of the best engineering at MIT is done in the School of Science. I'd like to say some of the best science is done in the School of Engineering, but I'm not sure I can say that. When you get to these half-billion-dollar projects, the problems they have to solve are real engineering problems — as mundane as pushing the limits of surveying technology.

§1.p15

So that's LIGO — sublime measurements. 10 to the minus 19 meters is how much gravity waves change the location of that tungsten sphere holding the mirror. That's significantly smaller than the diameter of the nucleus — the nucleus is around 10 to the minus 13 meters. We're five or six orders of magnitude smaller than the nucleus. That's the distance they're trying to measure.

§1.p16

It's not something everyone wants to do. It's almost equal to the NIST annual budget. And then NIST has a precision measurement group, which is one or two percent of their annual budget, to make the most precise measurements of time and length.


§2. The ridiculous: spark testing for steel composition [10:01]

§2.p1

That's what we do for big science. Now let's talk about the ridiculous. Does anybody know what I'm going to do with my grinder and my steel rods? We're going to do spark testing. Chemical analysis à la 1900s steel.

§2.p2

Different compositions of steel produce different amounts of sparking. This is 1018 steel — regular old carbon steel. When I worked for a steel company forty years ago, which is almost an eternity, there were guys who'd been spark-testing in the yard for thirty years. They had big yards full of rolled steel bars. The bars were supposed to be labeled, but they didn't have fancy laser marking, and they didn't have nice battery-operated grinders forty years ago either.

§2.p3

[Tom takes a grinder to a steel rod.] If I take a grinder to this, I get sparks. Those sparks are different than if I take another piece of steel — this happens to be 440C. You can see the difference: hardly any sparks. This is a stainless steel; that was a carbon steel.

§2.p4

I have a bunch of other steels here. This is 1045, which has a little more carbon — medium carbon, that one was low carbon. If I grind this one versus that one, the difference is in the little sparks at the end. High-carbon steel: lots of sparks. Mild steel: very few secondary sparks. Do you see the secondary sparks on that one? There are some on this one too, but they're not as intense.

§2.p5

I can't tell the difference if you put two of these in front of me. But if you'd been doing it for thirty years, you could. These guys had 99.9% accuracy out in the yard. I have some alloy steels here too — tungsten steels, molybdenum steels. The most dramatic is stainless versus carbon. Stainless doesn't spark much because it doesn't oxidize well. This is all a question of how the steel oxidizes.

§2.p6

Pretty ridiculous way to do things, right? The people who could do this have all retired or died. What we do now is use a $15,000 gun. CMSE has one. It's got a little radioisotope that generates gamma rays, and you put it up to a piece of steel and it gives you a digital display of ten different elements within plus or minus a few tenths of a percent. So you don't need a guy who can look at sparks. But that's how people used to do it.


§3. The ball identification exercise [13:59]

§3.p1

There are complex ways to do things like LIGO, and simple ways to do things. So we're going to give you a test — an in-class lab. I have four sets of these balls. Each one of you can do this. There are about eleven or twelve balls in each bag, but there are nineteen things the balls could be. You've got four or five minutes to see how many you can figure out.

§3.p2

Last time I did this, people only got four or five out of nineteen. See how many you can do without the magnet, then use the magnet to see if you can figure out any others. Some of them you may only be able to narrow down to A or B — like, most of you can figure out you have a rubber ball, but there are two different types of rubber and you probably can't tell the difference.

§3.p3

If you think you've gotten as many as you can, then use the magnet. Make a note of how many you distinguished with the magnet's help. You can distinguish certain types of stainless steel by magnetism. There are some carbon steel balls in there too. Unless they're rusty, you can't tell the difference between carbon steel and stainless. There are no duplicates in the bag.

§3.p4

Okay, let's talk about it. It's not necessarily easy, but what properties were you using to distinguish the materials? That's a measurement. Everyone got the first one — brass. What property did you use? Color. Color is a property.

§3.p5

I bet most people got the second one. What property did you use to find the aluminum ball aside from color? Density. Most of you found it first on density. Some on color. Either way, that's another property.

§3.p6

Low-carbon steel — did anybody figure out which steel ball is low carbon? It's not easy unless it's rusty. If it's a little rusty, you got it by corrosion resistance. You couldn't tell it on density, and you couldn't easily tell it on color — unless you were really good, like the guys with the spark testers who'd been at it every day for thirty years.

§3.p7

Go into a machine shop, and a machinist can look at those balls and tell you whether they're 9/16, 7/16, or 1/2 inch. They all happen to be 1/2 inch. If I just put a 7/16 and a 9/16 ball on the table and asked you the diameter without telling you, the average person would be about 50/50. But a machinist who deals with these dimensions every day will get it right 98% of the time. So a lot of it is experience.

§3.p8

The stainless steels — you were able to distinguish that one of them was magnetic and the other two were not. That's because the 300-series stainless steels are austenitic, 18-8, non-magnetic.

§3.p9

[Tom takes a magnet to a Farberware pot.] I use this Farberware pot in a lot of lectures. It's non-magnetic down here because it's 18-8 stainless — actually 304. Up here it's quite magnetic. That little red magnet — whoever ended up with it — sticks great up here.

§3.p10

I use this in my forming course, deformation processing, and in other courses. Students complain I use the same props in different classes. Well, I only know certain things. We're all limited. Student: What were these bought for? I bought them for 3.014, and I don't know if we ever got to use them in 3.014, but I did buy them for 3.014. I may have talked in 3.014 about doing it, but the little labs always ran over. This was always going to be a fill-in if we had a couple of minutes at the end. I made these up.

§3.p11

If you know enough about stainless steels, you could separate the 440C. The tool steel — M50 — is also magnetic. So the low-carbon steel, the M50, and the 440C are all magnetic. You got three magnetic balls — I saw you picking up three with the cow magnet. By the way, that's a cow magnet. The two of you in the front row got the cow magnet.

§3.p12

Do you know what a cow magnet is? It's a stainless steel magnet — made out of 400-series stainless. You stuff it in a cow's mouth and make them swallow it. It goes into the cow's belly because cows tend to eat a lot of scrap metal, and the scrap metal can tear up their intestines. If you have something that holds the metal in the belly, the hydrochloric acid in the stomach will digest the steel before it gets to the intestines. So dairy cows get to eat a magnet. They're called cow magnets, they're cheap, they're stainless steel, and they don't rust. So I bought a couple. Don't worry — they've never been in a cow.

§3.p13

The tool steel — to identify that you'd have to do a spark test, something destructive. I was asking you to do this non-destructively and without a lot of tools. You used density, color, you might have tried hardness — and you used coefficient of restitution. Did you know that? You were bouncing the balls. That's the coefficient of restitution.

§3.p14

There were two rubbers. I can't tell the difference, but Viton is a fluoroelastomer. If I wanted to do it non-destructively and I had a scanning electron microscope, I could put it in the SEM and see which one has fluorine. Non-destructive. All I need is a quarter-million-dollar instrument. Easy.

§3.p15

Torlon — anybody get Torlon? That's color again, but you have to know something about plastics. Acrylic — you probably got acrylic. The acrylic and the polystyrene are a little difficult because they're both crystal clear.

§3.p16

There's a simple destructive test to tell acrylic from polystyrene. If I'd given you a plumber's torch or a match, you could have slightly burned them and told from the smell. Same thing for polyethylene. You wouldn't know what polyethylene smells like when it burns, but this is a standard trick — take a little razor blade, take a piece of the plastic if you don't know whether it's polypropylene or polyethylene versus some other milky white plastic, drop it in the flame. Polyethylene smells like paraffin wax.

§3.p17

Why does polyethylene smell like paraffin wax? It has the exact same chemical formula — C_n H_{2n+2}. It's just a longer chain piece of paraffin wax. Polypropylene smells almost the same — it's got methyl groups rather than hydrogens off the side.

§3.p18

Anybody get tungsten carbide? Heavy, right? Density. Also, if you had something to scratch it with, you wouldn't scratch the tungsten carbide. Anybody get aluminum oxide or silicon nitride? Two of you have silicon nitride and two have tungsten carbide. That's because those balls cost about fifty bucks apiece. They're high-tech ceramics.

§3.p19

The low-carbon stainless steel balls were like three dollars for a bag of 100, and I couldn't buy less than 100. The tungsten carbide balls you buy one at a time at fifty bucks a pop. These all came from McMaster-Carr.

§3.p20

Aluminum oxide — nobody got it? Hard, white, dense — but not very dense. Light, like aluminum. Aluminum oxide is a little heavier than aluminum, but not much. Nylon — some people might have gotten nylon by color. It's a different milky white. If you burn it, it smells slightly putrid.

§3.p21

How many did you get? Let's see the competitiveness of MIT students. Students: Five. Six. Seven. Three. Four. Did you count the rubber? Yeah, count the rubber. Student: I got five. One of you was getting about ten — that's because you've been working in a lab with a lot of these different things for a couple of years. So there's the experience factor.

§3.p22

The point is, we analyze things all the time without really analyzing them. When I'm teaching a communications lecture, I sometimes hold up a tensile bar — an aluminum tensile bar — and ask, "What is this?" A civil engineer will say it's a piece of metal, or a tensile bar. A mechanical engineer will say it's a piece of aluminum. A materials engineer might say it's 6061-T6. Everybody has a different lexicon and a different level of specificity, even though we're speaking English.


§4. Practical measurement of length [31:54]

§4.p1

Now you've learned a little about how we measure properties. Let's talk about the practical side. Mass, length, and time — let's start with length.

§4.p2

[Tom holds up a tape measure.] A tape measure might give you one part in 20. You don't do too much with a tape. A ruler — this type isn't much better than a tape. A machinist ruler with 64ths of an inch can give you about one part in 100 or 200.

§4.p3

A micrometer, or in this case a vernier caliper with a dial gauge, gives you about one part in 1,000 just from mechanical movement. An electronic one will give you about one part in 2,000 — half a thousandth of an inch.

§4.p4

Then there are gauge blocks. Have you seen gauge blocks? These are the good ones — chromium carbide, rather than tool steel, so they won't scratch.

§4.p5

Gauge blocks have been used by machinists for about 100 years to precisely measure length. They're ground so precisely on the end that they actually stick together by adhesion — just the bit of grease on them is the adhesive. Take my adhesive bonding course and you'll learn about ringing gauge blocks together.

§4.p6

These come with a certificate of calibration. They're called Webber blocks because they're made by Starrett, but the traditional name is Johansson blocks, or Joe blocks, because those were the first.

§4.p7

The certificate tells you these were measured using the international inch, because there are different inches — different calibration societies. One inch equals 25.4 mm exactly. That's a defined definition of an inch back in terms of millimeters, which are defined in terms of the speed of light. So when you're talking about calibrations, you actually have to define what standard you're calibrating to.

§4.p8

Calibrated at 68° Fahrenheit, 20° C, 45% relative humidity maximum. The accuracy for 1-inch gauge blocks is about 0.02 microns, which is 0.8 micro-inches — eight tenths of a millionth of an inch. For 2-to-4-inch blocks, double that.

§4.p9

You can get them in different levels. If you pay enough, you can get master blocks calibrated back to NIST — the company makes them and sends them to NIST, and NIST measures them to their standards.

§4.p10

So there are all kinds of ways to measure length. If you control the temperature, you can easily measure to 50 millionths of an inch. You can get down to 10 micro-inches.

§4.p11

Why do you have to specify the temperature? The coefficient of thermal expansion. Steel has a coefficient of thermal expansion of about 10 to the minus 5 per degree C — go look it up, it's just material properties. That means if your room isn't constant within 1° C, that 10 micro-inches becomes another 10 micro-inches per degree.

§4.p12

Years ago I stole from my wife's sewing kit — she got it from her mother — a cloth tape measure on a roll. I liked it because I could take it with me and it rolled up the size of a dime. Now I can get small metal tape measures, and they don't stretch much because they're metal. That cloth tape, if you stretched it, you changed the length.

§4.p13

So whether it's temperature or just the modulus of the material, plastic rulers are not as accurate as metal rulers because they have ten times the coefficient of thermal expansion. But many times, one part in 20 is plenty. If you're cutting a dressmaking pattern, that's all my mother-in-law needed. Who cares if it's off by a 16th of an inch — you're going to sew a seam over it.

§4.p14

So you can improve things by a factor of two by going from a mechanical dial to an electronic dial. Electronic dials are cheap now — as cheap as mechanical ones — because of improvements in semiconductors.


§5. Chips to ships: extremes of manufacturing precision [38:30]

§5.p1

That's measurement of length. But let's think more about length. What are the most precise things we manufacture? After all, this is supposed to be a course on manufacturing. In parts per thousand or parts per billion, what are the most precise things we make? When someone asks a question like that, you have to go to the extremes. What's the extreme of a commonly manufactured product today in terms of small?

§5.p2

Student: Microprocessor. A microprocessor — a semiconductor chip, about 1 cm squared. What tolerance do I have to hold on that? Student: Nanometers. About 10 to the minus 8 meters. The chip is 10 to the minus 2 meters. That's one part in 10 to the sixth.

§5.p3

You have to worry about thermal expansion. They're actually doing lithography where they use masks to print the structures. Maybe it's within a nanometer, but really they're talking 20 nanometer feature lengths, or 13 — go look up on the Intel website where they are right now.

§5.p4

I just bought some jump drives at Costco yesterday — a pack of three for twenty bucks, 16 gigabytes each. You can get 64 gigabytes for eighty bucks today. I threw them on my desk with my 2-gigabyte ones, and somewhere on my desk I have a 128-kilobyte jump drive, probably ten years old. They keep marching down in precision.

§5.p5

Why are they limited to about a centimeter? Take my joining course and you'll learn — it's the same thermal expansion problem. If the chip gets too big, it has to be attached to a substrate with a different coefficient of thermal expansion, and they heat up 30, 40, 50 degrees C in service. You work out the delta T and you might crack them. The military has made them two inches on a side for super-duper applications, but you have to do all kinds of thermal management.

§5.p6

What's the other extreme? The next largest thing we build that has no expansion joints? Highway bridges go for 20 miles, but they have expansion joints, so that doesn't count. Chips at one end — and the other end is ships. Chips to ships. A 300-meter, 1,000-foot ship. When you weld these together, how closely do you have to line up the ship plate before you weld it?

§5.p7

That's three times 10 to the fifth. So who wins? It turns out Intel is winning now. A few years ago, ships were winning. Shipbuilding ten years ago was the most precise — in parts per million or billion — manufacturing we did. I now have to admit that the semiconductor guys are getting down there too. Build bigger ships.

§5.p8

I told you about Catherine Gebbie [Katharine Gebbie] yesterday — the woman who went to Bryn Mawr and Harvard and MIT, PhD in physics, who heads up the atomic clock group at NIST. I caught her at a break and asked her — because she's now in the manufacturing group at NIST, but she's an experimental physicist — I asked, "Do you know what the most precise manufacturing there is?" They like to talk in parts per million, billion, trillion. She said no. I said, "It's shipbuilding." She looked at me, and I pointed to these numbers. "Oh." She could relate to one part in a million. She said, "How do they do it?" I said, "If the sun goes behind a cloud, the ship changes shape by half an inch — and you have to align them within a millimeter." She said, "Well, how do they do it?" I said, "Great big 100-ton jacks. They just squeeze them together."

§5.p9

It shocked her, because she's been doing ultra-precise things. In shipbuilding, the practical answer is you just use a bigger hammer to pull it together. You don't use a bigger hammer when you're making semiconductors. So we're at two extremes — your quantum mechanical end and your astronomical end. Electrical readouts have improved our ability really only by about a factor of two.


§6. Weight, time, and the limits of measurement [44:31]

§6.p1

So that's length. We can do weight fairly quickly. A mechanical bathroom scale is probably good to plus or minus a pound — one part in 100. An electronic scale gets your weight to a tenth of a pound. Whether it's accurate or not, for precision and reproducibility, it's accurate.

§6.p2

If you're weighing for chemical analysis — gravimetric analysis, weighing your sample, digesting it, measuring concentration — that's still one of the most precise, at about one part in 5,000.

§6.p3

If I'm analyzing gold, they still use the fire assay technique. If you were in the materials department 120 years ago, you'd have taken a course in fire assay. There's a safe down in the basement where they used to keep the gold. The buildings were built in 1917, and they were probably still teaching fire assay as part of the undergraduate labs. You weigh the karat gold on a very precise balance — one part in about 5,000 if you do it properly.

§6.p4

Then you digest it by throwing it into a bath of lead. You oxidize away the copper, the zinc, everything, and you're left with gold and silver. Lead and gold are immiscible — look at the phase diagram — so you get a little button of gold at the bottom with the lead floating on top. Gold's more dense than lead.

§6.p5

You pop off that little button, roll it out in a little rolling mill, and put it in nitric acid. The nitric dissolves away the silver in this thin strip, and then you weigh the gold.

§6.p6

By this process you're actually refining the gold to 99.999% pure, and you weigh it to one part in 5,000. It's still the most accurate way.

§6.p7

Twenty-five or thirty years ago I was given the job of coming up with a better process than fire assay. We couldn't do it then. We're getting close now with inductively coupled plasma — probably one part in 2,000. But the standard, if you're doing a referee — a fight between people over the concentration of gold — is still fire assay. If you're dealing with fifty million dollars worth of gold, one part in 5,000 is $10,000. It's worth being accurate.

§6.p8

The standard is still fire assay, which has been done for several thousand years. Over the last twenty-five years we've improved the other stuff by about ten times — the inductively coupled plasma is getting close.

§6.p9

What are the limitations on weight? Wind, vibration, and fingerprints. If you've ever had to weigh something as precisely as possible, the balance — even if it's electronic — is sitting on a marble table with three-inch-thick legs of solid marble. Lots of mass to cut down vibration. It's inside a glass enclosure. Professor Allen has an old pan balance in his office with a glass enclosure. You have to wear white gloves — no fingerprints — and ideally lint-free gloves. Your fingerprint can set something off by one part in a thousand. Those are some of the limitations on weighing, but weighing is still one of the best.

§6.p10

Time. What's the history of measuring time? We used to have mechanical clocks, and there are people who spend their lives buying antique clocks for fantastic amounts of money.

§6.p11

Why did the Navy build the global positioning system? Knowing your longitude has always been the classic problem in ocean navigation. You can get your latitude from the sun, but you need to know time precisely to know longitude. So time has always mattered for navigation.

§6.p12

A mechanical clock might be accurate to one second in a day — about one part in 10 to the fourth. Then quartz clocks came along — that's one part in 10 to the fifth. Quartz clocks didn't exist when I was born. We had mechanical clocks.

§6.p13

A mechanical clock divides time into about 100 hertz. An electronic quartz clock divides it into 1/100,000 or 1/1,000,000 of a second. An atomic clock divides it into one part in 10 to the 10th seconds. NIST F1 does one part in 10 to the 15th. Time isn't really a problem in measurement right now. We've made tremendous strides over the last fifty years. So the three fundamental quantities — mass, length, and time — those are sort of the limitations.

Cases referenced

  • LIGO gravitational wave detector construction §1.p3

    Tom's role welding the 4-foot diameter beam tubes, and Ray Weiss's hand-written explanation of the 10⁻²³/√Hz measurement target. Used as the canonical "sublime" extreme of precision measurement. Tom also uses it to make the point that the best engineering at MIT is sometimes done in the School of Science.

  • NIST F1 atomic clock §1.p2

    One part in 10¹⁵ accuracy — one second in 30 million years. Returns at end of lecture as the time-measurement upper bound.

  • Spark testing for steel composition §2.p1

    Tom's first-hand recollection from steel-company work forty years earlier; demonstration with grinder on 1018, 1045, 440C, and alloy steel samples. Used as the "ridiculous" (in the sense of low-tech, embodied-knowledge) extreme paired against LIGO.

  • Cow magnets §3.p12

    Aside during the ball-identification exercise — Tom uses 400-series stainless steel cow magnets as the magnet props. Explains the livestock-medicine origin.

  • Fire assay for gold analysis §6.p3

    Tom's account of the centuries-old gravimetric technique still standard for referee gold analysis: digest in lead bath, oxidize off base metals, separate gold-silver button, dissolve silver in nitric, weigh refined gold. One part in 5,000 accuracy. Tom notes he was tasked 25–30 years ago with developing a replacement; inductively coupled plasma is finally getting close.

Layer 2 — cleanup edit
p1 00:01

Actually has to do with codes and standards, but you can read it. It's a short, uh, relatively short article on some of the history of codes and standards. This is lightning codes and standards. I can hand it out because we talked about the CSST and how lightning can perforate it.

p2 00:21

Um, are there any questions? If there's no questions, then I'll go back into what we were talking about, which was basically observables, and we were going from the sublime to what I call the ridiculous. And this is the NIST F1 atomic clock, one part in 10 to the 15th accuracy, which is 1 second in 30 million years. So we have a pretty good handle on time today in terms of sublime.

p3 00:52

But I was going to talk about the most— give you a little case study on the most, um, accurate measurement ever attempted. And that is the laser interferometric gravitational observatory. Okay? It's a joint project. Nope, can't read it. Maybe I'll turn the lights off. Uh, let's see if we can— still can't really read it. But it's got MIT up in the corner here. This is the MIT website. There's a Caltech website. There's an MIT logo right there. You can barely see it.

p4 01:28

Um, but back in 1992, a scientist in the physics department, Ray Weiss, who had started out at MIT as a technician, got his PhD in physics and now is retired, but 20 years ago, he and two people from Caltech proposed to the National Science Foundation what has become the largest National Science Foundation project in terms of cost that's ever been attempted. And they built two interferometers.

p5 02:03

If you remember from your freshman physics, Michelson-Morley experiment, they had a little interferometer where you split light on two perpendicular axes, and you send part of the light down this way and part of the light down that way, and then you have mirrors at the other end. They come back, and you actually see the beating frequency as the wavelength of the light, and you can measure the length very precisely along the length of those interferometers.

p6 02:31

Well, these interferometers are 4 km long, and there's one located in Hanford, Washington, which is a desert, as you can see. And the other one is located in Lafayette, Louisiana, which is not a desert. It's surrounded by a swamp. Okay? But again, they look pretty much the same. This is about a half billion dollar project. You can't even take a picture of the whole thing except from space. This is the one in Lafayette, and here's the two 4 km long axes.

p7 03:08

And I'd read an article in Science that this was supposed to be one part in 10 to the 23rd, which is the accuracy they're trying to measure. So I had worked with Ray. I actually worked on welding of the beam tubes. It's about a 4 ft diameter tube, 4 km long. And this is actually in Ray Weiss's handwriting, okay? He came down to my office to explain this to me. They're trying to measure a length difference— that's the way they write it down— 10 to the minus 23rd per square root of hertz.

p8 03:44

Um, anybody have an idea why you do square root of hertz? This is the how frequently you sample the measurement. Well, yeah, you could call it Nyquist frequency, but whenever you're dealing with a Gaussian distribution, remember the precision plot? You had the bell curve. You're trying to make a measurement. Here, we're trying to measure a length, and they're trying to measure gravity waves from space to confirm Einstein's general theory of relativity.

p9 04:14

And gravity waves are created when a supernova occurs, and it sends waves out through space, which will change the weight of everything. So you're gaining weight and losing weight very frequently. Okay? As you know, several tens of— 20 times a second. If you just combine all the weight loss and not the weight gain, you know, anyway. If you can figure that out.

p10 04:35

Anyway, what they do is square root of hertz because if you sample for a long period of time, you can get better accuracy. If you short sample for a shorter period of time, you get more measurements. And so there's a trade-off of: do you want lots of measurements to get good statistics, or do you want more precision by sampling for a longer period of time? Well, these waves will actually go through there too fast. If you sample over too long a period, just average everything out.

p11 05:06

So anyway, he said, "Well, the length is 4 km." This is meters. I mean, he's a physicist. He doesn't put units down, you know. Come on. What do we need units for if you're in physics? 4 km, and if you're talking about 1 second, 'cause he could— even as a physicist, he could take the square root of 1. And the change in length that you're trying to measure, if it's 10 to the minus 23rd, would then be over 4 km, 4 times 10 to the minus 20 m per square root of hertz. If you're going to sample, make measurements 10 times a second, you lose three times— you lose basically threefold in your accuracy. Because you're sampling too fast, faster than a second. So it's 1 times 10 to the minus 19 m is what you're going to try to measure as a distance with these things.

p12 06:01

And so they have these, like, 12-inch diameter tungsten spheres that have a mirror on them back in that little building where the two lines come together. They have a mirror at the other end. They had to align these tubes within about 3/16 of an inch, I think, over 4 km. No one had ever surveyed anything that accurately. Civil engineers didn't know how to do it 'cause no one ever had to do— 4 km, you actually have to start accounting for the curvature of the earth. Okay? When you start doing something like this.

p13 06:38

And they ended up setting up using the GPS system, which civil engineers use all the time for surveying now, but they actually had to do the little base stations, which increased the accuracy of the GPS and things to be able to even align these things on their foundations.

p14 06:58

And there were all kinds of things they had to do. They had to do it in the vacuum because when you're trying to measure something this small, I told you that light changes speed in a vacuum, and so they're down at about 10 to the minus 12 atmospheres, okay? And in fact, the primary impurity is hydrogen, and one of the things that they worked on— they had the undergraduates and others working on this— was taking the stainless steel and developing the right way to oxidize the surface of the stainless steel vacuum chamber so it wouldn't outgas hydrogen, very much hydrogen, at 10 to the minus 12 atmospheres. Because, you know, 10 to the minus 12 atmospheres, when you're dealing with 10 to the minus 19ths, you can get— doesn't take very much contamination to throw you off, right?

p15 07:46

And it's this 10 to the minus 12 atmospheres of hydrogen, and there's a little helium in there too that they can't— they have some diffusion pumps. If they had done it the conventional vacuum technology way, they would have been spending like 20, 30 million dollars a year just in energy cost to pull the vacuum. So they had to develop a vacuum system that was so tight, so hermetically sealed, that they could do it with a couple of vacuum pumps that would fit in this room. Okay? For this whole beam tube. And that was a whole new technology in terms of vacuum technology. So they pushed the limits on surveying technology, vacuum technology, outgassing of surfaces.

p16 08:31

And this is when I learned 15 or 20 years ago that some of the best engineering at MIT is done in the School of Science. I'd like to say some of the best science is done in the School of Engineering, but I'm not sure if I can say that. Uh, but nonetheless, it's amazing the types of things that— when they get to these half billion dollar projects, the types of problems they have to solve are real engineering problems, okay? As mundane as pushing the limits of surveying technology.

p17 09:00

In any case, that's LIGO, which you can say is sublime measurements. If you think about 10 to the minus 19 m, which will be how much gravity waves change the location of this tungsten sphere, which holds the mirror— that's significantly smaller than the diameter of the nucleus. Okay? The nucleus, I think, is like 10 to the minus 13 m or something. So we're like five or six orders of magnitude smaller than the diameter of the nucleus. Okay? And that's the distance they're trying to measure. So it's pretty interesting.

p18 09:40

But it is sort of sublime. It's not something everyone wants to do. It's almost equal to the NIST annual budget. Okay? And then NIST has this precision measurement group, which is one little, you know, one or two percent of their annual budget, okay? To make the most precise measurements of time and length and things like that.

p19 10:01

Um, so, um, that's what we do for big science. Now let's talk about the ridiculous, okay? And the ridiculous— well, does anybody know what I'm going to do for the ridiculous with my grinder and my steel rods? We're going to go to spark testing. Chemical analysis a la 1900 steels.

p20 10:27

It turns out that different compositions of steel will produce different amounts of sparking. And so I have— this is 1018 steel. This is regular old carbon steel. And when I worked for a steel company 40 years ago, which is almost an eternity, there were guys out there in the steel company who'd been working for 30 years doing spark testing out in the field, because they'd have a big yard where they had the steel they'd rolled— the steel— or had big bars of steel out there in the yard. And they're supposed to be labeled, but they didn't have fancy laser marking, permanent marking, and stuff. And these guys could take a grinder. They didn't have nice battery-operated grinders 40 years ago.

p21 11:09

[Tom takes a grinder to a steel rod.] But if I take a grinder, I get sparks off, right? And those sparks are different than if I take another piece of steel. This happens to be 440C. And you may not be able to see the difference here. Well, actually, you can on this one. There were hardly any sparks, right? This is a stainless steel. This is a carbon steel.

p22 11:42

But in fact, I have a bunch of other steels here. This is 1045 steel, which is a little bit more carbon. This is medium carbon, that's low carbon. And if I grind this versus I grind that, it's just the difference in the little sparks on the end, okay? There's a high carbon steel with lots of sparks. Mild steel B, very few secondary sparks. So compare this with that and see if you can see. Do you see the secondary sparks on that one? There's some secondary sparks on this one, but they're not as intense.

p23 12:36

I can't tell the difference if you put these two in front of me. But if you'd been doing it for 30 years, you could tell the difference. And these guys had like 99.9% accuracy in going out there in the yard. And I have some alloy steels here too. And they're all a little different. Tungsten steels, molybdenum steels. The most dramatic is stainless steel versus carbon steel. And you can see stainless steel doesn't spark very much. It doesn't oxidize well. So this is all a question of how the steel oxidizes.

p24 13:09

Pretty ridiculous way to do things, right? The people who could do this have all retired or died. And in fact, what we do now is we have little $15,000 guns. Oh, we don't have one in the department. The 911 [course 22] department, CMSE [DMSE], has a— oh, okay. I had to get one. Or maybe I can check it out when we get back. But it's got a little radioisotope in it, which generates the gamma rays. And you put it up to a piece of steel, gives you a nice digital display of like 10 different elements in it within about plus or minus a few tenths of a percent. So you don't need a guy who can look at sparks and tell you what it is. But that's one of the ways people used to do things.

p25 13:59

So there's lots of complex ways to do things like LIGO. And then there's lots of simple ways to do things. So what I wanted to do is— we're going to give you a test. And there's only three or four of you. So, unfor— I didn't give you a set of these balls, did I? I had four of these at one time. And I can't— oh, wait a second. Maybe I found my fourth. I just found my fourth. So each one of you can do this.

p26 14:37

We're going to take— we're going to do an in-class lab. This is— the balls tend to run away from you. So you can be number two, you can be number four, you can be number three, and you can be number one. I'm going to give you a sheet of paper and give you about four or five minutes. And there are— I think it'll be, like, 11 balls, 12 balls or something in there. But these are the 19 things the balls could be. Sorry. And you've got a few minutes to see how many you can figure out. Anybody need a pencil?

p27 15:27

I will tell you the last time I did this, people were only getting four or five out of 19. Oh, you can only get 11 or 12 'cause there's not 19 there. Okay. I have all 19 in the bag down there, but— don't use it yet. Yeah, after— see how many you can do without the magnet, and then you can use the magnet to see if you can figure out any others.

p28 17:09

And some of them you may not be— you may be able to say it's either A or B, number 10 or number 12. All right? Like, I think most of you can figure out you have a rubber ball, but there's two different types of rubber, you probably can't tell the difference, right?

p29 17:59

Well, if you think you've gotten as many as you can get, you can then use the magnet. I just want to know— make a note of how many you were able to distinguish with the help of magnet. I will tell you that you can distinguish certain types of stainless steel by magnetism.

p30 18:23

Okay, well, we'll talk about it. So if you can sort out one of your stainless steel balls— but, you know, well, anyway, there's some carbon steel balls in there too. And you can't really— so unless they're rusty, you can't tell the difference between carbon steel and stainless. All right? By the way, there are no duplicates in there. Okay?

p31 20:50

That— it's not necessarily that easy, but what properties you were actually using— the properties of the material to distinguish the materials, right? And that's a measurement. I know that everyone got the first one. Brass. Right? And what property did you use? Color, right? Color's a property, okay? And in this case everyone got that first one.

p32 21:18

I bet most people got the second one. What property did you use to find the aluminum ball aside from color? It was density. Most of you found it first on density, right? Some of you might have found it first on color. But it was either color or density, right? So that's another property.

p33 21:38

Low carbon steel. Did anybody figure out which one of those steel balls is low carbon steel? It's not that easy unless it's rusty. If it's a little bit rusty, you figured it out, right? So you can do it on corrosion resistance. But you couldn't tell it on density and you really couldn't tell it that easily on color. If you were really good, you probably could tell it on color. If you were a guy who— just like the guys with the spark testers who'd been doing it every day for 30 years, you could probably look at the color.

p34 22:11

You go in a machine shop and you see a machinist, he can look at— he could look at those balls and he could tell you whether they're 9/16, 7/16, or 1/2 in. They all happen to be 1/2 in. You know, if I just put a 7/16 and 9/16 ball on the table and asked you what's the diameter of those and didn't tell you one was 7 and one was 8/16, I'd get about a 50/50 on the average person. But a machinist who's sitting there dealing with these dimensions and things of these dimensions every day can tell probably 98% of the time he'll get it right if you give him something unknown. So a lot of it is experience in doing things.

p35 22:53

Um, the stainless steels— you were able to distinguish one of the stainless steels was magnetic and the other two were not, right? That's what it should have been anyway. And that's 'cause the 300 series steels are austenitic, they're 18-8, they're non-magnetic.

p36 23:13

I need one of your magnets. [Tom takes a magnet, picks up a Farberware pot.] And you'll see— and if you take some of the lectures, I use this Farberware pot a lot of times. It's non-magnetic down here 'cause it's 18-8 stainless. Actually, it's a 304 stainless. Up here, it actually is quite magnetic. Okay? If I had that little red magnet, whoever ended up with that— that's the one I usually use. And that little red magnet sticks to this thing great up here.

p37 23:42

And that has to do— if you take my forming course, which is deformation processing, I use this and I use it in some other courses. Students complain I use the same props in different classes. Well, okay. I only know certain things, you know? We're all limited. Student: What we— ? I bought them for 3014, and I don't know if we ever got to use them in 3014, but I did buy them for 3014. Yeah, I think I may have talked about it in 3014 that we were going to do it, and then, you know, in the little labs we always ran over. But this was always going to be a fill-in in the lab if we had a couple minutes at the end of the day. I made these up. The rubber balls.

p38 24:27

So anyway, the stainless steel balls— if you know enough about stainless steels, you could separate the 440C. The tool steel is also magnetic, M50. So the low carbon steel and the M50 are both magnetic. The 440C is magnetic. You got three magnetic balls, which you found— three. I saw you picking up three balls with the cow magnet, right? By the way, that's a cow magnet. The two of you up in front row got the cow magnet.

p39 24:58

You know what a cow magnet is? It's a stainless steel magnet. So it's made out of a 400 series stainless. And it's designed to stuff it in a cow's mouth and make them swallow it. And it goes down into the cow's belly 'cause cows tend to eat a lot of scrap metal. And the scrap metal can tear up their intestines. And so, if you have something that holds it in the belly before getting to the intestines, eventually the hydrochloric acid in the stomach will digest the steel before it tears up their intestines and their udders. So dairy cows get to eat a magnet. They're called cow magnets and they're cheap. And they're stainless steel. They don't rust. Okay? So I bought a couple of cow magnets. Isn't that nice? They've never been in that cow. Don't worry about it.

p40 25:53

Okay. The tool steel— you'd have to do something like a spark test. You'd have to start doing something destructive. I was asking you to do this non-destructively and without a lot of tools around. You used density, you used color, you might have used— tried to use hardness on some of these other things. You were doing— you were using coefficient of restitution. Did you know that? You were bouncing the balls, right? That's the coefficient of restitution. Right?

p41 26:20

There were two rubbers. I can't tell the difference, but Viton is a fluoroelastomer. If I wanted to do it non-destructively and I happen to have a scanning electron microscope, I could put it in the SEM and see which one's got fluorine, right? It's non-destructive. All I need is a quarter million dollar instrument. Easy.

p42 26:41

Okay. Torlon. Anybody get Torlon? Yeah? Which one is it? Yeah. That's color again, okay? But you have to know something about plastics. Acrylic. You probably got— you think you got acrylic, right? The acrylic and polystyrene are— that should be polystyrene. Um, are a little bit difficult 'cause they're both sort of crystal clear. Okay?

p43 27:10

There is a simple test to tell the difference between acrylic and polystyrene that is destructive. If I'd given you a plumber's torch or a match, you could have slightly burned them and you would have been able to tell from their smell. Okay? That's another thing for polyethylene. Polyethylene— you wouldn't know what it smells like when it burns. And this is a standard trick. Someone will take a little razor blade and take a little piece of the plastic if they don't know if it's polypropylene or polyethylene versus some other plastic that just looks milky white. And they look at it— you can't tell by looking at it, but if you take a little piece and you drop it in the flame, and it smells like paraffin wax.

p44 27:58

Why does polyethylene smell like paraffin wax? It has the exact same chemical formula. C2N H2N+2. It's just a longer chain piece of paraffin wax. And polypropylene smells almost the same 'cause it's— I can't remember. Its formula is, you know, it's got— it's just a couple of extra carbons. It's got ethyl groups rather than methyl groups off the side type of thing. Or it's got methyl groups rather than hydrogens off the side. Anyway, I'd have to write it down.

p45 28:32

Anybody get tungsten carbide? You got it? Heavy? Okay, density. Also, if you could do it— if you had something to scratch it with, you wouldn't scratch the tungsten carbide. Anybody get the aluminum oxide? Or the silicon nitride? You got the silicon nitride? Two of you have silicon nitride and two of you have tungsten carbide. That's 'cause those balls cost about 50 bucks a piece. Okay? They're high-tech ceramics.

p46 29:00

And so I didn't buy— I mean, the low carbon stainless steel ball, the balls were like $3 for a bag of 100, okay? And I couldn't buy less than 100. The stainless steel ball or the tungsten carbide balls, you buy them one at a time at 50 bucks a pop or something like that. These all came out of McMaster-Carr.

p47 29:21

Aluminum oxide— some of you probably got aluminum oxide based on— nobody did? That looks like it's probably right. Hard, white, dense. Not very dense. Light like aluminum. Aluminum oxide is a little heavier than aluminum, but not much heavier. Aluminum oxide, nylon— some people might have gotten nylon by color. It's kind of a different milky white. If you burn it, it smells slightly putrid. Not putrid, but anyway.

p48 29:58

Anyway, how many did you get? We'll now do the competitiveness of the MIT students. Students: Five, six, seven, three. I don't know if you can count the rubber. You can count the rubber. Okay. Hey, look. I got like four. Four, okay. Did you count the rubber? I got five, but I didn't— yeah, okay. Five, three, four. 19. Yeah. You think. No, I had 16 balls, but I got 19. You got— actually, it looked like you were getting about 10. Yeah. And that's because you've been working in a lab with a lot of these different things for a couple of years now, right? So yeah, there's the experience factor, okay?

p49 30:50

Anyway, the point is we analyze things all the time without really analyzing them. I sometimes will hold up a piece of aluminum and I'll say— when I'm teaching a communications lecture, I'll hold— this is a piece of, uh— what is this? This is polypropylene, I think, or polyethylene, one of the two. Anyway, I'll hold up a tensile bar and I'll say, "What is this?" And it'll be an aluminum tensile bar. And I always say a civil engineer will say it's a piece of metal, okay? Or they may say it's a tensile bar, okay? A mechanical engineer will say it's a piece of aluminum, and a materials engineer might say it's a 6061-T6.

p50 31:39

In that everybody has a different language, even though we're speaking English. They all have a different lexicon that they tend to use and a different level of specificity. Okay? So far as that goes.

p51 31:54

But anyway, so now you've learned a little bit about how we measure some of the properties. And a lot of these are very— well, I could call it the ridiculous, but they're very practical. If we start talking about the practical in measuring things, and we talk about mass, length, and time— let's start with length. We might start with a tape measure.

p52 32:22

[Tom holds up a tape measure.] And a tape measure might give you one part in 20. Maybe you can get a little better, but I mean, you don't do too much with a tape. A ruler— well, there's this type of ruler, which is not much better than a tape measure. There's this type of ruler, which is a machinist ruler, which has got gradations on here in 64ths of an inch. So this can give you about one part in 100 or 200 for a ruler.

p53 32:45

There's a micrometer, or this— I didn't have a— in my office, I only had a vernier caliper, but this has got a dial gauge and it gives you about one part in a thousand. Okay? Just mechanical movement. Uh, if I had— I did have in my office, but the battery was dead— an electrical one of these. And it'll give you about one part in 2,000. It'll give you half a thousandth of an inch or something like that. Okay?

p54 33:15

And then there's gauge blocks, which I just happen to have right here. Anybody know gauge blocks? Have you seen gauge blocks? You know gauge blocks, okay, as a machinist. Okay, gauge blocks. In this case, I got the good gauge blocks, which are chromium carbide as opposed to tool steel. And these are made out of chromium carbide, so no one will scratch them.

p55 33:44

And gauge blocks are used by machinists and have been for about 100 years to very precisely measure length. They're ground so precisely on the end that they will actually stick together— adhesion— just by the little bit of grease that's on there. That's the adhesive. Take my adhesive bonding course and you'll learn about ringing together gauge blocks.

p56 34:07

These gauge blocks come with a certificate of calibration. The— these are called Webber blocks 'cause they're made by Starrett, but the traditional name is Johansson blocks or Joe blocks. 'Cause those are the first ones that were made.

p57 34:27

But this one— it tells you that these measurements were made using the international inch, 'cause there are different inches, okay? Different calibration societies, right? 1.000 inch equals 25.4 mm exactly. That is a defined definition of an inch back in terms of millimeters, which are defined in terms of the speed of light, right? So you actually have to— when you're talking about calibrations, you actually have to start defining what standard you're calibrating to.

p58 35:02

Calibrated at 68° Fahrenheit, 20° C, 45% relative humidity maximum. Woo. And the accuracy of these things is about, um— for 1-inch gauge blocks, it's about 2.02 microns, which is 0.8 micro inches— eight tenths of a millionth of an inch. And for 2 to 4 inches, it's double that and other things.

p59 35:30

But you can get these in different levels. If you pay enough, you can get them calibrated back to NIST. NIST will have done— you can get them— what they call master blocks— and they will be calibrated back as if NIST measured them. The company makes them and they send them back to NIST, NIST measures them to their standards, okay?

p60 35:56

So there's all kinds of different ways to get measurement of length. And you can measure with these— if you control the temperature, you can measure these to easily 50 millionths of an inch. That's kind of the standard. You can get them down to 10 micro inches. Okay?

p61 36:19

Why do you have to specify the temperature? It turns out the coefficient of thermal expansion— the limits on these things are— at this end, it's the temperature. Steel typically has a coefficient of thermal expansion about 10 to the minus five per degree C. Go look it up. It's just the material properties. That means if your room isn't a constant temperature, you know, within 1° C, that 10 micro inches becomes another 10 micro inches for every degree centigrade. Okay?

p62 37:00

Down here at the tape— years ago, I stole from my wife's sewing kit, which I think she got from her mother, a cloth tape measure on a roll. And I liked it 'cause I could take it with me and it rolled up about the size of a dime, half inch thick. And I could wheel that out and measure inches. Now I can get small little tape measures and these are actually metal. They don't stretch much 'cause they're metal. But that little cloth tape, if you stretch it very far, you change the length. Okay?

p63 37:32

So whether it's temperature or it's just the modulus of the material you're making it out of— plastic rulers are not as accurate as metal rulers because they have 10 times the coefficient of thermal expansion. Okay? But you don't always need it. Many times, one part in 20 is plenty. You know, if you're cutting a dressmaking pattern, that's all my mother-in-law needed, okay? When she was making dresses, okay? Who cares if it's off by a 16th of an inch? You're going to sew it up and make a seam out of it anyway, right?

p64 38:08

So in any case, you can improve things, like I said, by about a factor of two by going from a mechanical dial to an electronic dial. Okay? And electronic dials are cheap. They're as cheap as the mechanical ones now. Due to improvements in semiconductors and stuff.

p65 38:30

Okay, so that's measurement of length. But let's think a little bit more about length. What are the most precise things that we manufacture? After all, this is supposed to be a course on manufacturing, right? In parts per thousand or parts per billion, what are the most precise things we manufacture? What products? When someone asks you a question like that— or if when I ask you a question like that— you have to go to the extremes. What's the extreme of a commonly manufactured product today in terms of small?

p66 39:12

Student: Microprocessor. A microprocessor. A semiconductor chip, which is 1 cm squared. Right? So we got the major length dimension. So it's just a little square piece of silicon 1 cm on a side, right? And what type of tolerance do I have to hold on that thing today? Student: Nanometers. Uh, 10 nanometers. Okay. Whatever. About 10 to the minus eight meters. This is 10 to the minus two meters. That's one in 10 to the sixth. Okay?

p67 39:59

You got to worry about thermal expansion. Okay? They are actually doing lithography where they use mask to print the structures. And maybe you're right, maybe it is within a nanometer, but really they're talking 20 nanometer feature lengths or 13 nanometers, depending— go look up on the Intel website what they're at right now.

p68 40:24

But I just bought some jump drives, a pack of three for 20 bucks at Costco yesterday, 16 gigabytes on a jump drive, right? And you can get 64 gigabytes if you want to pay 80 bucks today. So I threw them on my desk with the two gigabyte ones, and somewhere on my desk I have a 128 kilobyte jump drive, right? It's probably 10 years old. But they keep on marching down in greater and greater precision.

p69 40:51

Why are they limited to one centimeter in length typically? If you take my joining course, you'll learn it's the same thermal expansion problem. If the things get too big, they have to be attached to a substrate that has a different coefficient of thermal expansion, and they do heat up 30, 40, 50 degrees centigrade in service. And you figure out what the delta T is, and you might crack them. Okay? If you make them very big. The military's made two inches on a side. Okay? For super duper, you know, things. But you have to do all kinds of thermal management, they call it, in the semiconductor business.

p70 41:32

So what's the other extreme? What's the next largest thing we tend to build that has no expansion joints? You can think of— highway bridges that go for 20 miles, right? But they have all kinds of expansion joints. So that doesn't count. These are chips. The other ones are the other end. They're called ships. Chips to ships, okay? Okay. So a ship 300 meters long— 1,000 foot long ship. And when you weld these things together, how closely do you have to get that ship plate to line up before you weld it?

p71 42:16

And so that's three times 10 to the— 10 to the fifth. So who wins? Well, it turns out Intel is winning now. A few years ago the ships were winning. Shipbuilding 10 years ago was the most precise— in parts per million or parts per billion— was the most precise manufacturing we did. I now have to admit that the semiconductor guys are getting down there too. Build bigger ships.

p72 42:42

So I told you about Catherine Gebbie yesterday, I think— this woman who went to Bryn Mawr and Harvard and MIT and PhD in physics and heads up the atomic clock group at NIST now. And I caught her at the break and I asked her, "Do you know what the—" 'cause she's now in the manufacturing group at NIST, but she's a physicist, right? Experimental physicist.

p73 43:06

So I asked her at the break, I said, "Do you know— do you know what the most precise manufacturing there is is?" And she says— you know, because they like to talk in parts per million, you know, or parts per billion or parts per trillion— these kind of one part in so many pieces. And she says, "No." I said, "It's shipbuilding." And she looked at me and I pointed out these numbers to her. "Oh." You know, so she could relate to, you know, one part in a million or so. And she says, "Well, how do they do it?" I said, "If the sun— if the sun goes behind a cloud, the ship changes shape by half an inch. And you got to have them aligned within a millimeter." She says, "Well, how do they do it?" I said, "Great big 100-ton jacks." Okay? They just squeeze them together.

p74 43:52

It kind of shocked her because she's been doing— with little micro, you know, ultra precise things. And the practical, in the case of shipbuilding, is you just use a bigger hammer, basically. Okay? To pull it together. You don't use a bigger hammer when you're making semiconductors. So we're at two extremes. Okay? So you got your quantum mechanical end and you got your astrological [astronomical] end. Actually, we don't even— we're not in astrological end like— okay. But electrical readouts have improved our ability by really only about a factor of two.

p75 44:31

So that's mass and— length. We can do time— we can do weight fairly quickly. Yeah, we can, 'cause that clock is behind or it's too fast. Weight. A mechanical scale— a bathroom scale that's non-electronic, with the old mechanical scales— is probably only good to plus or minus a pound, right? Unless you get a really good one. Okay, it's got compensations and stuff. But okay, one part in 100. If I use an electronic scale, boy, you can get your weight to a tenth of a pound now, right? Whether that's accurate or not, but for precision reproducibility, it's accurate. Okay?

p76 45:15

If you are weighing something for a chemical analysis in a chemistry lab— gravimetric analysis, basically weighing your sample and then digesting it and measuring, you know, what the concentration is— is still one of the most precise, at about one part in 5,000.

p77 45:35

If I'm analyzing gold, they still use the fire assay technique. And if you were in the materials department 120 years ago, you would have taken a course in fire assay. There's a safe down in the basement where they used to keep the gold and stuff. I mean, this is, you know, the buildings were built, you know, in 1917. They were probably still teaching as part of the undergraduate labs fire assay, where you basically weigh the carat gold or whatever the gold is on a very precise balance and you can do this to one part in about 5,000 if you do it properly.

p78 46:10

And then you digest it. In this case, you digest it by throwing it into a bath of lead. Okay? And then you burn off— oxidize away the copper and the zinc and everything— and you're left with gold and silver. And the lead and the gold are immiscible if you look at the phase diagram. So you got a little button of gold at the bottom with the lead floating on top. Gold's more dense than lead.

p79 46:39

And you take off— it just— pop off that little button. You then roll it out in a little rolling mill, and you then put it in, um, an acid. It's not hydrochloric. Maybe it's nitric. I think it's nitric. No, no, you dissolve the gold. I think it's just straight nitric. And you dissolve away the silver in this very thin strip, and then you weigh the gold.

p80 47:05

And you can get gold that's— by this process you actually are refining the gold to like 99.999% pure, and you weigh it and you get one part in 5,000. It's the most accurate way of doing it. It's still the most accurate way.

p81 47:21

We are getting— I actually was given this job 25 years ago, 30 years ago, to come up with a better process than fire assay. And back then we couldn't do it. We're getting very close now with some of the inductively coupled plasma. You probably get one part in 2,000. But the standard, if you are doing a referee— a fight over people about what's the concentration of your gold— and let's face it, one part in 5,000, if you're dealing with 50 million dollars worth of gold, you know, one part is $10,000, right? It's worth being accurate, right?

p82 47:57

So in any case, the standard is still fire assay, which has been done for several thousand years. Okay? But in fact, over the last 25 years we probably improved the other stuff so it's probably 10 times better— the inductively coupled plasma. And it's starting to get close.

p83 48:19

But in any case, what are the limitations? The limitations on weight: wind, vibration, and fingerprints. If you've ever had to go in a lab and measure something as precisely as possible, the balance, even if it's electronic, is probably sitting on a marble table that's like three-inch-thick legs of solid marble and three-inch— you've seen these, right? They're called weighing tables. Lots of mass to cut down the vibration. It will be inside a glass enclosure. Professor Allen, if you look in his office, he's got an old balance and it's got the glass enclosure for the pan balance. And you have to wear white gloves. No fingerprints. In fact, it's best if they're lint-free gloves. Your fingerprint can set something off by one part in a thousand. Okay? So those are some of the limitations to weighing, but weighing is still one of the best. Okay?

p84 49:18

Um, time. What about time? That was what Frodo— no, it wasn't. That was the Hobbit asked, right? Time. Time heals all things. So what's the history of measuring time? We used to have mechanical clocks, and there are people who spend their lives and buy antique clocks for fantastic amounts of money.

p85 49:44

And the ancient mariners— why did the navy build the global positioning system? It's always been a problem of knowing not your latitude— you can get that from the sun— but your longitude. Okay? Has always been a classic problem in navigation of the oceans. So you need to know your time very precisely to know what longitude you're at in the day, to know where you are in the ocean.

p86 50:11

So mechanical clocks— you know, you could get something that's accurate to, I don't know, 1 second in a day or something like that. And then we got quartz clocks. Before you were born, but well, during my lifetime— quartz clocks didn't exist when I was born. We had mechanical clocks. Well, quartz clocks came along and that's one part in 10 to the fifth. Whereas a mechanical clock might be one part in 10 to the fourth.

p87 50:46

But quartz clocks divide time up— well, actually, I guess if I think— a mechanical clock divides time up into, like, 100 hertz. An electronic quartz clock divides time up into 1/100,000 or 1/1,000,000 of a second. An atomic clock divides time up in one part in 10 to the 10th seconds. And NIST F1 does it one part in 10 to the 15th, right? So time is not really a problem in measurement right now. We have made tremendous strides over the last 50 years. Okay? So the three fundamental quantities of mass, length, and time— those are sort of the limitations. Okay.