of the early arcade games
Well its January 2009 and I thought I'd update this page for the first time in perhaps 20 years.
For example this picture no longer resembles me much. I don't have a newer one but if you imagine someone substantially older, fatter, greyer, with less hair above the ears and more below you will have the right idea.
People often ask me about that picture so let me talk about that first.
It was taken in 1987 at the castle in Milan. The statue is of the god Bacchus riding naked on the back of a turtle. I was travelling with Willy and Monica McAllister at the time and as we approached the statue Willy said, "Hey, that guy looks like you." I said, "No he doesn't." He said, "Sure he does. Stand there and strike that pose." And he took some pictures.
Well, he must have been right because no sooner do I stand like that and other tourists start taking pictures too."
I'm the guy on the right.
My First Paper
At this point most academics list their CVs.
Well, I've written some papers over the years. Some of them have even been published. Nothing like the output of someone with publish or perish pressure though. I worked at HP for almost 23 years and that took up most of my time. Some patents came out of that though. So rather than just a dry list of papers and patents let me tell the stories behind some of the things I did.
Besides, I'm in a talkative mood. So I think there may be a LOT of stories here before I'm done.
I was a freshman math major at the University of California at Santa Barbara in 1972. It was an interesting time both for me and for the world.
I was hungry to learn so I signed up for every math course I could fit into my schedule. And, as I knew I was headed for some kind of career in computers, I also took some computer courses. One of them was a numerical analysis course taught by Professor Sloss. While I did well in all my courses (I got straight A's that year) the fact that I was doing well came to his notice.
I remember the incident.
Professor Sloss was going through the error analysis for Newton's method. The analysis is usually given in terms of relative error. That is if the relative error of a given guess is O(epsilon) at some step it will be O(epsilon2) at the next step. One of the students in the class asked if there were any way to bound the absolute error. Professor Sloss said, "No. You cannot bound the absolute error."
Well, without realizing what I was doing I sat there shaking my head. Professor Sloss saw this, looked at me, and said, "You disagree?"
I said, "Well, since Newton's method converges so fast almost anywhere you start you will have a pretty good approximation of the actual root after just a a few steps. You could then take that root, go back, and subtract it from your original guess and have a pretty good approximation of the absolute error. You could then carry forward with a calculation of the absolute error."
I don't remember what he said to that. I do remember there was a long pause before he said anything. I think I had said something off the top of my head that was in a direction he had never considered before. I found out later there was a very good reason for that.
You see in 1972 it was still possible to be a professor of numerical analysis without knowing how to use a computer. Professor Sloss was an excellent mathematician but up to that time didn't even know how to log on to the campus computer system. Much less how to program it. His researches at the time were all of a theoretical nature. So the notion of using the actual result of a computer calculation to feed back into a theoretical argument was not in his experience.
I found all of this out because as a result of that incident the professor asked if I would help him with a paper he was working on. He wanted someone to help him make some plots on the computer to illustrate some mappings in the complex plane that were central to his topic. He said I would get mentioned by name as credit for my help. I said sure and over the next couple of days I programmed the computer to make the desired plots. He took pictures of the display with a poloroid camera and went away happy.
I transfered to Stanford the next year and never knew anything about what happened with the paper. Until a couple of weeks ago. (It just turned 2009 as I write this.)
| CERTAIN MULTIPLY CONNECTED REGIONS by James M. Sloss |
Now, 37 years later, I was having an email discussion with someone in France on an issue of interval arithmetic when I had cause to google myself in order to find some reference to some of my previous work. And, with google including more academic works these days, up popped Professor Sloss's paper with my name in it.
I had completely forgotten about it. I had never read it. I wasn't even aware of the title. But there it was. The first paper with my name in it. Not ON it but as I was only 19 at the time it was something of importance to me personally.
I did some searching around and eventually called someone in the math department at UCSB to see if I could get a copy. Any form would do. I'd even pay for it if necessary. Well with so old a paper there was some understandable difficulty but I was eventually able to buy a pdf of it. If it were anything other than this particular paper I wouldn't have bothered but I needed to see it.
You can find it here.
For full disclosure, I should tell you a couple of things about the state of my education at the time.
I was good with computers. From when I was first introduced to them at age 15 to when I graduated high school I had learned over 40 languages (I stopped counting at 44) on more than a dozen different computers by just about every manufacturer at the time. From BBN to Wang. (Both Apple & Zilog were still years away.)
I was always good with math. Even as a child. But by my freshman year in college I knew of calculus, complex numbers & functions, and Fourier series but not yet Fourier transforms. It wasn't until the following year that I learned of differential equations and things like Dirichlet problems.
So the topic of Professor Sloss's paper was outside my experience at the time. I knew only enough math and programming to make the diagrams he was interested in.
Even that has some historical context.
The NLS System
In 1972 computers were things owned by governments, universities, oil companies, and insurance companies. They were big things. They could easily fill your house. They needed to be installed by highly trained experts into air-conditioned rooms in buildings often designed just for that purpose. And they cost $3 million or so.
UCSB had one of the best. It was not the biggest or the fastest but it was one of the best. Actually they had two. But the other was a more or less conventional IBM machine that was accessed with punched cards & produced 14" fan fold printouts. I'm talking about the good one.
It was called NLS. NLS stood for oN Line System. The name was a historical artifact meant to distinguish it from the Off Line System which was, oddly enough, called OLS.
The reason NLS was thought of as 'online' was that you could walk up to a terminal, sit down, log on, and use it at the same time as several other people were doing the same thing at other terminals. This radically new concept at the time flourished during the time that computers were becoming more interactive and died away again when they became so cheap everyone could afford their own.
Still the thing that made NLS great was its terminal. It was unique. I've never seen its like before or since. The display was a direct write Tektronix scope that held an image of what was written to it in the capacitor formed by the phosphers. Thus it was not scanned as TV displays are scanned. The way you wrote a letter onto the screen was to: turn off the electron gun, move the x & y voltages to where you wanted to start your letter, turn on the gun, move to the end of the stroke, turn off the gun, move to the beginning of the next stroke, and so on. So the letter A took 3 strokes. An E took 4 but 3 of them could be strung together. A G had to be approximated however you liked. And so on. Programming the display required its own language and I had fun learning that as well. When you wanted to erase the screen you pushed a button that momentarily grounded the phosphers, the charge was lost, and the display went blank.
You were never going to watch a movie on this thing but it was cool anyway.
There was something even more important that went into making NLS unique.
It was the keyboard. It was not like a conventional keyboard. It is even hard to describe. Imagine if you opened a laptop way out with the top half nearly 45 degrees from being flat on the table. Now imagine that you have a conventional keyboard where you normally have one below the hinge. But in this laptop the manufacturer has goofed up and rather than put a display above the hinge there is another keyboard there. With all new buttons that you don't have on the bottom keyboard. That is something like the NLS keyboard but it was molded into a solid piece of white plastic and looked really cool.
And what did you need all those other buttons for? Well the NLS designers decided to create a programming method (it was more than just a language as you can see) in which the operands were to be found on the bottom keyboard and the operators were to be found on the top. So if you wanted to take the sine of x the x was found on the bottom keyboard and the sin() was found on the top. Actually, it was parsed in an RPN style so the utterance would be "x sin". (RPN stands for Reverse Polish Notation, something that would later become popular with the HP-35 calculator. The Pole in question is Jan Lukasiewicz. Look him up if you are interested.)
And whatever you said appeared on the screen as you said it. Further, you could work on real numbers (which would appear as a decimal string in something like an E format), complex numbers (an ordered pair of decimal strings), real vectors (which would appear as a plot with the index position of the vector on the x-axis & the function value on the y), complex vectors (a plot of the vector in the complex plane), real arrays (a 3-D plot), and complex arrays (a plot in the complex plane even if a bit contrived).
Now as this building sized, multi-million dollar computer was VASTLY less complex, less capacious, and slower than even the cheapest throw away cell phone of today, computing with arrays was a bit of a stretch for it. But it did well enough anyway. Actually it was amazing.
One more thing about those plots.
There were no ink jet or laser printers to which I could have just dumped those plots. There was a pen plotter but Professor Sloss preferred not to use it. What he ended up doing was to sit behind me one night watching me make the plots on my Tek scope and when he saw one he liked he took a poloroid picture of it. Then he photographed the poloroids, blew them up, and had someone trace the plots BY HAND into his paper. As I remember the original plots I can see something was lost along the way.
Not much in the way of computer aided typesetting but at least he didn't have Microsoft Office to contend with.
Thus ends the story of the first time my name appeared in print in a peer
reviewed paper.
My Nerd Valentine
There is a minor epilog to that story.
Professor Sloss didn't use that plotter but I did.
For Valentine's Day that year, I sat down in front of the display & worked out a complex transformation to map the unit circle onto the shape of a heart. Then I plotted it out together with a short love poem by Kalil Gibran & sent it to my girlfriend.
It killed.
Don't tell me nerds can't be romantic.
The Disk Drive Company
During college at Stanford I worked at Westinghouse. Perhaps I'll make that the topic of another story.
But after graduating college I got a job at a disk drive company. It was one of the strangest places I've ever been associated with.
For one thing, during the 9 months I worked there I'm not sure we ever sold one disk drive. We repaired a lot but never sold one.
But this company was made strange from the top down. The president of the company was one of the most egomaniacal men I've ever met. He once told me in complete seriousness that the ultimate attainment of man was to reach a point where he could tell others what to do. And I'm completely sure he believed he had attained this personal satori.
The accountant was a bit off too. As I look back on it with the perspective of years it is possible that some sort of money laundering was going on.
Everyone else was great & I struck up many friendships but we all lived in this strange environment. I'll not go into the details but suffice it to say that the attrition rate at this company was unusually high. After a while I noticed that people we leaving at a pretty regular rate. After 6 or 7 months of this I began to notice a pain in my stomach as if I might be getting an ulcer. Perhaps I should have picked up on this earlier when the guy who hired me left the company right after showing me what I was supposed to do.
So I applied for & got a job elsewhere & told my boss I would give him two weeks if he liked but I was leaving.
A half an hour later the president called me into his office & asked me in a very confidential tone why everyone was leaving the company.
Well, I already had another job so for the first & last time in my life I burned my bridges. I said, "Why? A blind man could see it with a cane. Its you. They're leaving because of you." And I told him why in chapter & verse.
He took it & said, "I want you out of here right now. Don't speak to anyone. Just go." And he made sure I left.
I found out from my friends that after I left he exploded on the remaining employees & wanted to know if anyone else wanted to leave as well. I don't think anyone raised their hand then but I'm sure most of them got out in the next few months.
There are 3 interesting epilogs to this.
First, as I got in my car to leave & turned on the radio it was playing Take This Job & Shove It. I sang along at the top of my lungs & laughed so hard it was difficult to drive.
Then, I started my new job on Monday morning & not long into my day I was told I had a phone call. I thought this was strange because mostly no one knew where I worked yet.
It was from the president. (He evidently got the number from one of my friends at his company.)
He said OK, he'd cooled off & I can come back now. I said do you have any idea where you are calling? This is my new job. Did you think I was kidding or something? Evidently he had realized that I was the only guy at the company who knew how to do what I did. I said I kept good notes & its all on my desk & I would answer questions for my replacement when he was hired.
And the phone called ended.
The third thing to happen was he called a couple of weeks later to tell me about the new guy. Actually it quickly became clear that he was calling to BRAG that he had hired a bright high school student to do my job for much less money.
The conversation went something like this:
He said, "Do you remember what you got on the math section of the SAT?"
I said, "Yes."
"What did you get?"
"800."
"800? A perfect score?"
"Yes."
"Well, um, this kid got a 760."
"That's very good."
He kind of petered out after that & I never heard from him again.
In my ENTIRE life I think that phone call was the only time that getting 800 on the SAT was ever worth anything to me.
But I REALLY enjoyed it.
The Video Game Manufacturer
|
|
|
of the early arcade games |
The next place I went was a video game manufacturer.
Now you must realize that in the mid 70s video games were relatively new. Atari was one of the pioneers. There was no Sony Playstation. There was no Microsoft Xbox. There was no Wii. Everyone else in the business was already in the business of making pinball & slot machines.
You see at that time a video game was not something you played on your lap. It was not something you played on your home computer (which itself was a few years off). It was not even something that you attached to your TV & played on the couch.
It was not something ordinary people had in their homes at all.
A video game was something you stood next to, put a quarter in the slot, & pounded on it for as long as you could. It was something to be played with a beer in a very loud place like a bowling alley or a bar.
This was the pinball & slot machine business model.
The way it was described to me when I started working there was this: You know all that stuff that an arcade game (the retronym for an old video game) is doing when you walk up to it? That is called the Attract Mode. It shows all the most exciting stuff that can happen in the game as an enticement to get you to put your quarter in the game. When that was done, the game is over, you've won. Now all you've got to do is entertain them for 30 seconds or so & get them to put another quarter in.
The president of THIS company was a very pleasant fellow. He was kind of the hereditary president of a pinball, slot machine, & vending machine company that had been handed to him from his father &, I'm pretty sure, from his father as well.
He had a bunch of old pinball machines in his office the earlest of which was made entirely of wood & dated to something like 1911.
He was one of those people who knew nothing of technology & was proud of the fact. He was also extraordinarily talented in being able to identify a good game just by playing it. He had a big fishbowl of quarters in his office & insisted that everyone in his business (there were maybe 30 of us), at least once a month, come by, grab a handfull of quarters, & go someplace to try out the competition. (I only did it once. After all I only worked there for 3 months.) He told me a lot about pinball machines, et al, including an incredible depth of knowledge about just how tough coin boxes & joysticks had to be.
Well, as I was only there 3 months I barely had time to program one video game, a bowling game, the details of which he & I & my boss kicked around one day in his office.
My contribution was to make it a 3-D bowling game in that the user looked down the alley in perspective as the ball shrunk heading towards the pins. It took a guy like me to work out the math for the perspective as well as how the apparent motion of the ball changed as it moved away from you.
I got the last bug out of it literally on the production line.
There is one interesting story I think I'll tell.
We were having trouble in the then leading edge 8-bit 7 level stack microprocessor with 1024 bytes of RAM & 4096 bytes of ROM stone knives & bareskins technology we were using. It was hard to find enough memory to store all the positions & sizes the ball might take on as it rolled & hooked its way down the alley.
So I mused to my boss (the hardware guy) one day that it would be nice if I had a little 16x16 pixel array that I could move over the rest of the scene by changing the address of the upper left corner. He said, quite correctly, that the hardware at the time was not fast enough to merge that image with the rest of the image. I said, well you wouldn't have to merge it digitally, you could analog add it.
And I'll be damned if I came in the next morning & he was working on it. It was a little 2" x 3" PC board & he had it working that day.
I've never seen that done at any other company before or since.
Before that I never met anybody smart enough to do it.
And after that I was working at HP. I soon found out that NOTHING happens that fast at a big company.
Nothing. Not ever.
BTW, in later years, when the technology HAD gotten to the point where that
could be done digitally, it was called a sprite. Had I stayed at the company
I might have patented it. Ah, well.
HP & QA
Of all the jobs I've had in my life, Hewlett-Packard was the first I thought of as a career. And I knew it going in.
I started out in Cupertino in a software Quality Assurance (QA) group. They were about to have some floating-point hardware come through QA & they needed someone who could test it. Little did I know then that I would spend most of my career designing it.
I also didn't know that I would be testing a lot of transcendental functions software. They didn't either. It was what we later came to know as mission creep. What was a 3 month job turned into a 9 month job.
Still I enjoyed my time in QA enough to stay there for more than 2 years.
I learned how much fun it was to find bugs in other people's stuff & not have to fix them. And in doing so I learned how to avoid them in my own work. I learned to teach others under the nickname Dr Dan. (There were other nicknames, to be sure, but let's go with that one for now.) I was best man at a wedding, honorary uncle more than once, befriended, friend, & confidant.
That's what working at a good place is worth to you not even counting the work.
We had an adversarial relationship with the lab. That is they were trying to get something out the door & we were looking for reasons to stop them. It worked remarkably well. Too well for the lab managers. They noticed how much their projects were delayed. They did not notice how good the products were that got past us.
In fact in all the time I spent in QA I only got thanked by someone in the lab once. I was testing this guy's stuff & it was so good that it took me 2 of the 3 months I had on it to find my first bug, a minor one. He came 'down' from the lab, shook my hand, & thanked me profusely for finding it saying how terrible it would have been if it got out there that way. Funny that the only guy who ever thanked me is a guy who barely needed my services. Maybe not so funny after all.
I'll tell one story that illustrates how far we've come since then.
My first boss in QA was Nino Mateos. He was a wonderful guy to work for who drove an old 50s pink Corvette.
Anyway I was moving the memory boards from one computer to another one day when he saw me. He said, "Be careful with that, you have a quarter of a million dollars under your arm there." I looked at the boards & looked at him & said, "I really wish you had let me put them in the computer before you told me that."
You see, they were signed out on his name.
Let me put that in a 21st century context.
These were brand new memory boards. They had the brand new 16k x 1 bit memory chips on them. Each board was 64k bytes & cost $25,000. They were about 8" x 10" & I had 10 of them under my arm going from one computer room to the next.
A year ago or so I bought a 2 Gig memory stick for $8.
Some people point out that we've made computers thousands of time faster over the years. Or that they are thousands of times smaller (physically). Or that they have thousands of time more memory of one sort or another. Or that they consume thousands of time less power.
But our REAL accomplishment over the past 3 decades, what we've REALLY done for the world is to make them 100 MILLION TIMES CHEAPER.
The cheapest cellphone today is faster, more complex, & has more memory than a $25 million Cray 1-S Supercomputer ever had. Our clothes have little computers in them (RFID tags) that cost a nickel & the industry is doing all it can to get that down to a fraction of a cent because they are STILL TOO EXPENSIVE.
And 5 years from now everything will be 10 times cheaper.
And another 10 times 5 years after that.
I don't know when it will end but it cannot end until cost is no longer the
issue.
The SOS Chips
|
|
|
When I left QA I went to work in the hardware lab making floating-point chips. The fact that there WERE no such things as floating-point chips at the time & that I was entirely unqualified to design a chip of any kind seemed not to be a hindrance. At least not to the guy that hired me to do it.
My new boss was Fred Ware. He had a lot of unusual traits not the least of which was seeming to see in me something that told him I could do the job. Another was his vision that floating-point chips were possible.
You see Fred knew something that no one else in the world knew at that time. In an era where other people were lucky to be able to put 5000 transisters on a chip Fred knew we could put 40,000 on a chip. We had SOS.
At the time the world was making its circuits in CMOS. Thats Complementary Metal Oxide Semiconductor but most of you will know it as CMOS if you know it at all. There were other technologies, to be sure, but nothing you could hang a few thousand transistors on. There were Bipolar, NMOS, various TTLs, some left over selenium & germanium, & GaAs (Gallium Arsenide), the technology that is & always will be the technology of the future. But we had SOS.
SOS stood for Silicon On Sapphire. It was pretty much the same as CMOS but for the fact that rather than grow your transistors on a chip of silicon, you grew little silicon islands on a chip of sapphire instead.
It sounds more complicated & it is. Some extra equipment & a few more steps. It sounds more expensive & it is. At the time a 2.5" silicon wafer cost $5 & a 2.5" sapphire wafer cost $50. Today a 12" silicon wafer costs $1500 & they don't make sapphire like that any more. But there were a lot of advantages you got for all that.
For one, SOS is rad hard. Thats military speak for hardened against radiation exposure. Some aerospace companies, mostly Raytheon I think, used it to sell circuits to the military to be used in things that someone might throw a nuclear weapon at. Also, I seem to remember RCA made some SOS chips to go into the Voyager spacecraft. Theres lots of radiation in space. But this was not a property that much interested HP.
SOS consumed very little power. This was of interest to HP in its primary application, that of hand held calculator chips. As a bonus, the same thing that caused SOS to use little power also made it fast. That was a low dielectric constant.
The dielectric constant is Electrical Engineer speak for the property of a material that makes it a good capacitor. It turns out that insulators, like sapphire, tend have much lower dielectric constants than semiconductors, like silicon.
Why does this property make for fast low power circuits? Well everytime you want to do something in an electronic circuit it involves changing the voltage on some tiny wire from a 'zero' (called the ground) to a 'one' (usually the supply voltage) or vice versa. The time it takes to do this is proportional to RC where R is the resistance (mostly in the transistor itself) and C is the capacitance (some from the wire & some from the transistors being driven). The energy required is CV2 where V is the amount of the voltage swing. That was 12 volts in those days, some 10 times what is used today.
Since the wires on a SOS chip were running over sapphire their capacitive coupling to the sapphire was far less than the corresponding coupling that wires running over bulk silicon experienced in ordinary CMOS circuits. So C was low. So RC was low & CV2 was low. Fast & low power.
So we had the speed & we had the power advantage. We could also put many more transistors on a chip than the bulk CMOS guys at the time.
You see another property of silicon wafers is that they are deliberately doped with boron to make them slightly conductive. The reasons for this are complex & don't really concern us here but suffice it to say that it makes the transistors perform better.
This caused a layout problem for the bulk CMOS guys. If they placed a P-N-P transistor too close to a N-P-N transistor it would look to the wafer as something like P-N-P--N-P-N. And even though current was not meant to pass through the middle, some did. If too much did then something called latch up would happen. It meant that current was going where current was not designed to go & there was no way to turn it off. The circuit stopped working & if you didn't turn it off right away the whole thing would fry.
This would be solved in later years by drilling trenches between the transistors made possible by the discovery of anisotropic etches & SOS would go the way of the dodo bird. But at the time the insulating sapphire meant we could put two transistors right up next to each other & put 40,000 transistors on the same size chip that others were struggling to get 5000 on.
So knowing all this, what did Fred know that the rest of the world didn't know? Fred knew it was possible to make a chip that would do an entire floating-point operation (add, subtract, multiply, or divide) on a single chip. Well, 3 chips really, one for add & subtract, one for multiply, & one for divide.
Now, actually this had been done already in the 8087. But the 8087 computed things (say multiply) as you would on pencil & paper. It was a little engine that would multiply the entire multiplicand by 2 bits of the multiplier (like base 4 arithmetic instead of base 10), then shift that 2 bits & add it to the product of the multiplicand by the next 2 bits, & so on until the entire product was done. Lots of steps that needed a clock.
Fred realized that with 40,000 transistors we could lay out the entire product in a single large circuit that computed combinatorially. That means that as soon as the operands appear on the input wires they start to flow through the circuit & the answer appears on the output wires as fast as they can go through the circuit. No steps. No clock.
(We actually made ourselves a little test rig with switches & lights where we were able to enter the input problem by toggling the switches & see the answer show on on the lights at the output. It was fun to play with.)
This meant that while the 8087 could clock something like 50,000 floating-point operations per second (50 kFLOPS), our chips could zip through around 1 million (1 MegaFLOP).
Today the floating-point circuits are distant decendents of our designs that occupy some tiny corner of a larger computer chip & operate in the GigaFLOPS range on chips with often more than a billion transisters on them. But this was a long time ago.
Well, now that I've spent so much time explaining what we did & why, rather than tell you about the 18 months it took to design them let me just say, we did it. My contribution was small on these first chips but it was enough to get my initials on them. (More on that later.) And I soloed my next chip. It had 153,000 transisters on it & I'll talk about it later.
I will tell one short story: We had a fab problem that involved a short
caused by a 4 micron square piece of aluminum that didn't belong. It caused
the chip release to be delayed while the faulty mask was repaired. The delay
cost HP something less than $1 million. Fred was moved to figure out the cost
of a 4 micron square of aluminum given the price of aluminum in those days. He
invented the unit femto-pfennig to express the result.
ISSCC That Year
After we got them working Fred made a big splash when he announced them at the ISSCC that year, 1980 or 1981 I don't remember. The ISSCC is the International Solid State Circuits Conference & is THE place for big new advances in chips to be announced. Well it was at the time anyway.
Everyone was impressed & in the question & answer period after his talk someone asked when he could buy the chips & how much would they cost. The moderator took the mike from Fred & said that this was a technical conference & no place to discuss business deals. Fred took the mike back & said that the chips were made in a captive HP technology (SOS) to be used for HP products (some computers that were also about to be announced). So if the questioner were in business we weren't allowed to sell them the chips. But if the questioner were an academic we might be willing to GIVE them to him if he made an interesting proposal.
Well, after Fred got back from the conference & over the next few weeks we got maybe 20 or so proposals. They filled a showbox next to Fred's desk.
Most of them were pretty boring. Vector & array processors of various obvious designs that were all along the lines of things we had on the drawing board anyway.
Except for two.
|
|
|
One guy, at the University of Tennessee I think, wanted to build a Navier-Stokes engine. The Navier-Stokes equations describe the behavior of fluids. And since fliuds behave in many interesting & complex ways, the equations are very difficult to solve. This guy wanted to build a special purpose computer designed to solve this problem. Since this is the very problem people try to solve when designing a new aircraft or ship or car or artificial heart & on & on it sounded like something interesting to do.
So we sent him 400 sets of chips. Well over $1 million worth.
And we never heard from him again.
The other proposal came from MIT. A guy named Gerry Sussman wanted to build a Digital Orrery.
Of course our first reaction was to look at each other & say: What the hell
is a Digital Orrery?
The Digital Orrery
|
|
instrument maker George Adams the Younger (1750-1795) |
You know that thing made out of brass balls that Kepler had that when you turned the crank moved like the planets in the solar system? THAT'S an Orrery.
It turns out it was named after Charles Boyle, 4th Earl of Orrery. This is not the Boyle from whom we get Boyle's Law. That is Robert Boyle. But I think he was one of many brothers.
Anyway what Gerry wanted to do was build a special purpose computer that would numerically solve the n-body problem for the 10 bodies in the solar system (9 planets & the sun).
This sounded very interesting. And he only wanted 10 sets of chips. So we sent him the chips.
And the next thing that happened is that Gerry shows up at our desk in Cupertino wanting to move in with us along with 2 other MIT professors & 2 graduate students. They were Hal Abelson, Jacob Katzenelson, Andy Berlin, & Guillermo (Bill) Rozas (Jinx to his friends), respectively. They would come to be known collectively as Da Boys to Willy McAllister & I while we helped them with their work. Later, back at MIT, we would also get to work with Jack Wisdom.
So we set about finding desks & lab space for them & got down to explaining how to interface with the chips. We basically dropped most of what we were doing & spent the next 3 months helping them design & debug the orrery.
|
|
of Scientific Instruments by Charles Mollan, Royal Dublin Society |
It turned out that Willy was much more useful in this design & debug phase than I was. Willy was an experienced hardware designer (who designed one of the chips) & I was a novice at the time.
I was an experienced programmer but I was entirely outclassed by Da Boys.
I knew more about solving things in floating-point than most people on the planet but when it came to solving ODEs (Ordinary Differential Equations) with TRILLIONS of time steps, they were breaking new ground that Jack Wisdom would later make important advances in.
But I thought it was cool so I helped as best I could.
It was a small computer for its time. It used much the same hardware as the PCs of the day. It was about the size of a toaster. And it was nailed to a 12"x18" piece of plywood along with a power supply & some interface cables.
It also had two handles screwed into the plywood so it qualified as a portable computer. It cost around $80,000 not counting our time, of course. And it ran like a bat outa hell. Way faster than the next nearest competitor which was a Cray supercomputer costing some $25 million.
It couldn't balance your checkbook worth beans but when it came to figuring out where the planets were going to be it was the best computer on THIS planet for the job.
|
|
|
The first thing Gerry set about doing was to compute the positions of the planets as far out as he could go.
How far could you go & how do you know?
Among the things Gerry included in his calculations were a number of side calculations to make sure that the main calculations were going correctly. Gerry is a VERY good programmer which makes him a VERY good scientist as well.
He calculated things like the total energy of the solar system. The sum of the gravitational potential energy & kinetic energy for all the bodies in the solar system. So long as nothing enters or leaves the solar system the conservation of energy says this should never change. Most of the energy in the solar system can be found in Jupiter anyway.
Similarly, he calculated the total angular momentum of the solar system. And, similarly, the conservation of angular momentum says this should never change. And, not suprisingly, most of it is in Jupiter.
Issac Asimov once said that the solar system is Jupiter plus debris. These calculations bore that out.
Gerry did another interesting check. After calculating the solar system forwards in time for a 110 million years or so, he would reverse all the velocities & calculate backwards for the same amount of time. Any errors in the physics would be reversed by reversing time, so to speak, & all the planets should end up back where they started. But any errors in the accuracy of the calculations would ADD to similar errors going backwards & the planets would end up in different places.
In this way he found out that the calculation was making an error of around 57 degrees in the position of Jupiter after 220 million years. So for the moment he could go no farther & count on the results. He published a paper.
|
|
|
The way Gerry summed it up was that he could predict how the solar system
would behave but he could not do horoscopes for dinosaurs.
The Solar System is Chaotic
Then one of those things that happens all too often in science happened to Gerry.
He went to a conference in Colorado & happened to sit down next to Professor William Kahan of Berkeley (he is known as Velvel to his friends & I am pleased to be counted among them).
He told Prof Kahan of his project.
|
|
|
If he had mentioned that he did the work at HP he might have found out that Velvel knew both Willy & I from our IEEE-754 work, but I digress.
He mentioned that he had done some experiments involving small changes in the time step which seemed to indicate he could drive the error to zero but he wasn't sure if that was real or merely the appearance of reducing the error.
It turns out he couldn't have mentioned it to a better person. Velvel knew off the top of his head that that method DOES drive the error to zero (at least to first order) & Gerry went home from the conference with the solution in his head.
His next calculation was carried out to 845 million years with less than a 5 degree error in the position of Jupiter. Later he (with Jack, using something we all call 'Jack's Map') was able to reduce that even further & go a few billion years.
I suppose he could now calculate the horoscopes of dinosaurs or even
trilobites but that was not what was on his mind.
What did he Discover?
Well it turns out that the planets pull & tug each other in ways that have resonances on 30,000 year, 100,000 year, 5 million year timescales & longer. It is as if all the planets were strings in a very large & slow piano & plucking one string causes the others to ring in harmonic resonances.
|
|
|
As you might have guess by now, Jupiter is the loudest string of all.
But Jupiter wasn't the interesting planet at first. It was Pluto. Or, more to the point, a resonance between Neptune & Pluto. You see every 3 times Neptune goes around the sun, Pluto goes around twice. Almost EXACTLY. This is called a 3:2 resonance or a 3:2 lock.
The reason this is interesting is that the orbit of Pluto crosses the orbit of Neptune. Indeed from about 1977 to 1997 Neptune was the farthest planet from the sun. (I will pass on the more recent issue of whether Pluto is even a planet or not. You will see it doesn't matter.) So ever since Pluto was discovered in 1930 people have wondered if it would ever get close enough to Neptune to get kicked out of its orbit.
Well, Gerry & company (by this time Jack was involved) managed to show that every time Pluto crossed Neptune's orbit, Neptune managed to be somewhere else.
But they discovered more. I said ALMOST exactly. It turns out that over tens of thousands of years Pluto creeps ever closer to a bad encounter with Neptune. And JUST as its about to get into trouble it slows & reverses course & starts to creep away from danger. Then after a few million years it starts to creep into an encounter with Neptune from the other side &, again, just before it gets into trouble it speeds up & walks away from danger.
| Numerical Evidence that the Motion of Pluto is Chaotic Gerald Jay Sussman and Jack Wisdom |
But theres more. This several million year long dance of approach & fall back is ITSELF not stable over a time scale of a billion years or more. Sooner or later something bad will happen to Pluto. But we can't predict when because the orbit of Pluto is chaotic.
Now when a scientist says something is chaotic these days he doesn't mean merely that it is confusing. It means something very specific these days that goes under the description of "sensitive dependence on initial conditions".
|
|
|
In most systems if you push on it a bit it responds a bit. If you push on it twice as much it responds twice as much. But in a non-linear system such as the solar system if you push on it just a bit it might respond MASSIVELY.
The weather is a chaotic system & people like to say that if a butterfly flaps its wings in South America today it will affect the course of a monsoon in the Indian Ocean 6 months from now.
Whether or not that is true Gerry was able to measure something called the Lyapunov exponent which is a measure of the chaos of the solar system. He could say that if we make a 1 centimeter error in the position of Pluto today, we will be unable to predict the position of the Earth more accurately than 1 Astronomical Unit (AU) a billion years from now. The radius of the Earth's orbit is 1 AU.
So as weak as the tiny little string is for Pluto WAY OUT THERE its sound is coupling back to the entire solar system.
| Chaotic Evolution of the Solar System Gerald Jay Sussman, Jack Wisdom |
The conclusion: As Pluto goes so goes the solar system.
If the motion of Pluto is chaotic so is the motion of the entire solar system.
Not a bad result from a plywood toaster humming away in the back of some office in MIT for months on end.
| The Supercomputer Toolkit: A general framework for special-purpose computing Harold Abelson, Andrew A. Berlin, Jacob Katzenelson, William H. McAllister, Guillermo J. Rozas, Gerald Jay Sussman, and Jack Wisdom |
A few years later we would replace the Digital Orrery with something called
the Supercomputer Toolkit
which would verify & extend the orrery's results.
The toolkit would be the faster of the two but the orrery was cooler.
Archiving our Work
Well after the orrery was finished & Da Boys went back to MIT we got on with our work & they got on with changing the solar system. (Remember: don't buy land on Pluto.)
We were already working on the next set of chips (in NMOS, as it happens) when we got a call from Colorado saying that they'd like us to gather up all our notes & technical specs on the floating-point chips & ship them to a company archive they have in some sort of bunker out there. It seems that HP had plans to make calculators even in the case of a nuclear disaster or massive earthquake or something.
So we gathered up our work, put it in a big box, & sent it to the bunker. Along with a few choice gags for future generations. How many times do you get to fill a time capsule?
By this time SOS was running out of steam. Actually SOS was being slowly starved to death. It turns out that all it takes to kill a chip process is to not buy new fab equipment for 2 or 3 years. If you don't have new toys you can't keep up. If you can't keep up you can't play. And if you can't play they shut you down. So one day someone decided he wanted to get rid of SOS & it took 2 or 3 years to die.
Actually, it took a bit longer than that. When it was eventually announced that they were going to shut down the SOS line several divisions around the company said you can't do that because our entire line of products uses SOS chips & there are no replacements anywhere in the world.
Was this mere reality enough to keep SOS alive? No, once someone decides to kill something like this his career is riding on the fact that it will die.
So these divisions were asked: How many chips will you need for the entire future of your products before you can retool into another process? And they said: Millions upon millions.
And they were told: OK, we will make them all now & you will pay for them now & then we can shut down the SOS line. And they replied: Pay for them NOW? Well, maybe not so many millions after all.
And it was done & the parts were put on the shelf & the SOS fab was no more.
And the stage was set for future disaster.
But first...
The Smithsonian Called
The Smithsonian had called Gerry & asked if he still had the digital orrery. If so, they wanted it. It turns out that it was the first special purpose computer that discovered an interesting & important result in physics.
So Gerry called us & said since the orrery couldn't have been built without our chips, they wanted everything we had on them to be sent to the new Museum of Science & Technology they were opening.
We had already sent the most technical stuff to the bunker in Colorado but we looked around & it turned out we still had a lot left over. We had plots & photomicrographs & publicity photos & lots more that seemed more suitable for a museum than for a hole in the ground. So we sent it to them. Along with some left over chips.
And that's how my stuff ended up in the Smithsonian.
I haven't been there & I don't think my stuff is on display but I've been
told if I ever want to go there the curator would be happy to show me around in
the back where they keep all the stuff they haven't shown to the public.
We can safely throw it all away
One day, maybe ten years after all of this, Willy is moving his desk about 30 feet down the hall. (In this business you move your desk every year or two. I think it keeps them from sticking to the floor.) At the time I was working about 50 feet in another direction & he asks me to come by his desk so he can show me something.
So I walk over to his desk & he shows me yet more schematics, plots, & photos left over from the SOS days. Since he is moving he is planning to throw them away on the grounds that after all this time they can't be worth anything to anyone. He says, "I'm sure we can safely throw this all away now, don't you think?" I say, "Sure, but you KNOW that as soon as you throw it away we'll get a call from someone wanting to know if we still have anything on them." We both get a laugh out of that & he throws it all away. And as long as it was on my mind I went back to my desk, found a few old SOS items & threw them away as well.
Less than 3 months later, LESS THAN 3 MONTHS LATER, the phone rings & its someone from corporate wanting to know if I'm the Dan Zuras who once worked on the SOS floating-point chips & if I had anything left by way of documentation.
Naturally, my reaction was to laugh uncontrollably.
To which she wants to know why this is so funny to everyone. (It turns out that she had called Willy a few minutes earlier & he reacted in the same way.)
After I regained control of myself I ask her why she cares as there is a lifetime build sitting on some shelves somewhere. See Archiving our Work above.
And she gives me a long story about that.
First, since everyone had to immediately pay for all the chips they would ever need in the entire lifetime of their products, they were motivated to low ball their predicted future sales.
Next, since they had all they would ever need sitting on shelves already neatly tested, packaged, & burnt in, they gave themselves plenty of time to figure out how to replace them. They needed it as it probably took ordinary CMOS 3 years or so to catch up to SOS in what was possible.
Finally, over the years they started to notice that the SOS parts were dying as they sat quietly on those shelves. It turned out that a chemical used in the glue to attach the top of the package to the bottom was very slowly eating away at the metal pins that stuck through it. It was never noticed in parts that went right into products because when the chips were used the heat they generated would drive off the chemical & the pins were safe.
By the time she called me there were something like 2% of the parts that were dead already. That combined with the low ball estimates in the first place meant that they were just months from running out.
I say OK, I understand. I say you should go to the archive in Colorado, all our technical schematics & artwork are stored there.
And she says that they were all thrown away.
I make some intelligent comment like, "Huh?!?", or "What?!?", something along those lines.
She replies that they are in the habit of routinely throwing away anything older tha 5 years.
I say , "Do they understand what the word ARCHIVE means?"
She say, "Yes, but they were running out of space to store things."
I say, "Then they should have bought more space."
Then I point out that this is all moot anyway as there isn't any way anyone can make a SOS chip anymore without spending $20 or $30 million to recreate the fab. (Fabs were cheaper then. Now its measured in billions.) And even THAT wouldn't do it because all the people who knew how to do it have since left the company.
She says they are going to try to recreate something similar to the characteristics of the SOS process in a modern CMOS.
I say, "Good luck with that."
The rest of the conversation consists of her saying that they have popped the top off one of the chips & they are in the process of microphotographing the entire chip in the hopes of recreating the masks from that. Then me trying to explain that it won't be as simple as that. And on & on.
Willy & I met later that day, looked at each other, & just laughed.
She called him some time after me & he must have said roughly the same as I did.
We both offered to help, if only for the fun of it all. But we never heard from her again & to this day I don't know how that effort went.
If I had to guess they probably tried their best but, in the end, were forced to find another solution.
SOS had to die way back then & 200 people had to lose their jobs. I'm not sure why. Even if it had been properly developed I don't think SOS would have survived to this day. But I could be wrong. At least it would have been interesting to see what those 200 very clever people could have done with it.
A tiny epilog: There is something very much like SOS around these days. It
is called SOI for Silicon On Insulator. It has much the same advantages as SOS
but instead of Sapphire they use Silicon Dioxide (SiO2) as the
insulator. Like SOS you pay for the advantages of SOI with it being more
complicated. Unlike SOS it is much cheaper in dollars.
Graffiti
|
|
|
There is something about designing chips I haven't told you about yet. It's not really a secret (well not any more) but we chip designers like to keep it to ourselves.
We sign our work.
Right on the chips.
You see designing a computer chip a hard. It can take 18 months to 2 years out of your life. More & more of the detailed grunt work is being taken over by computer programs these days but in 1980 you had to personally lay out every last rectangle in every last layer of a dozen or so layers that described the exact arrangement & connections of tens to hundreds of thousands of transisters. Not to mention all the electrical design to make sure that the circuit does what you want as fast as you want it.
Modern chip designers have it much easier in that a lot of what we did by hand is done on the computer. But rather than use that advantage to make chips as simple as ours over a long weekend, they must design chips thousands of times faster with thousands of times more transisters with much tighter tolerances than we ever thought possible.
|
|
|
I don't envy them their task.
So I presume they still sign their work. I'm sure they are every bit as proud of their work as we were of ours. So they'd have to sign it.
This signature takes the form of tiny writing. If 'graffiti' is latin for 'little writing' we take that to an extreme the Romans could never have conceived.
So we scribble our initials. Or maybe our whole names. Or maybe we draw a little picture. Anything to say 'Killroy was here'. Graffiti has been getting more complicated over the years but this has been going on as long as people have been making chips, probably longer.
My first experience of it was with the Tosdata Chip. This was the multiplier chip used in the digital orrery. Fred drew a little 'full adder' in the lower right hand corner of the chip & put all our initials next to it.
A full adder is a circuit you use in arithmetic chips. It is pretty much the binary version of an add with carry.
|
|
|
The snake is Fred's pun on this. Fred loved his puns. The chip name itself, Tosdata, is something of a pun. I forget what it really means.
The stack of initials are: FW (Fred Ware), WM (Willy McAllister), HS (Howard Shishido), DS (Dan Sun), & DZ (Dan Zuras).
The other chip that went into the orrery was called the Addchilada, a further pun which doesn't make sense without the Tosdata. And the graffiti was a 'half adder', another pun that doesn't make sense without the other one. (A half adder does an add without carry.)
The next generation of chips were done in NMOS & I was in charge of designing the multiplier. I called it the Roadrunner chip in homage to that most fast of cartoon characters. I drew it freehand with a child's coloring book as a guide. As I remember it was 150 microns tall by 450 microns wide. I cut a hole in a superfluous ground plane in the upper right hand corner of the chip & put it there. The names of all who worked on the chip are just around the corner to the right but cropped off of this photograph.
|
|
|
This chip was briefly the biggest & fastest multiplier in the world but as anyone who works in this business knows, such records are fleeting.
Willy was the head guy on the NMOS add chip & he took the idea of signing his chip literally. He traced in his signature & the signature of Valerie Wilson, his mask designer. Everyone else's name appears in the list below which is the same as I had on my chip.
The list is: Willy McAllister, Theresa Corsick, Valerie Wilson, Mike Asturias, Dave Graubart, Dick Vlach, Kathy Clark, Balbir Pahal, Ken Holloway, John Carlson, Dan Zuras, Fred Ware, Mark Reed, Cory Chan, & Dan Sun.
In case you were wondering, the list is ordered by number of letters.
This chip was also briefly the fastest in the world.
The third chip in this set was designed by John Carlson. John is also a man of puns so he called his the Buffalo Chip.
|
|
|
As I remember he found a picture of a Buffalo in a National Geographic & traced it for his picture. Then he duplicated & resized it to make a mommy & daddy buffalo with 3 little ones. They appear in the top middle of his chip beneath an I/O bus & above a large ground plane.
He wanted something else to go with the herd & he had plenty of space to put it in. So we kicked it around over lunch one day & I said I thought I could do a crossword puzzle with all our names in it. So I did that & it appears just to the right of the herd.
Across: Valerie Wilson, Balbir Pahal, Ken Holloway, Dan Sun, Dick Vlach, Theresa Corsick, Dave Graubart, & Willy McAllister.
|
|
|
Down: Mark Reed, Cory Chan, Kathy Clark, Fred Ware, Dan Zuras, & John Carlson.
A couple of years later Willy was working on a CPU chip with another group. The computer was called Cheetah so he wanted to have a cheetah picture.
He found this one on the cover of a computer magazine & thought it would do well for his purpose.
This time he enlisted the aid of his wife Monica. Monica teaches interior design and knows her way around a drawing table. So as you can see the final image was not just traced from the magazine but redrawn longer & sleeker. Relatively speaking, that is. The cheetah is the largest of all of our drawings & it is STILL smaller than the smallest speck of dust.
Graffiti indeed.
|
|
|
|
|
|
|
|
|
The engineer is here
K-9
Damn spots
archeological dig
The Silicon Zoo
Combinatorial Divide
The ASPLOS-II Paper
An old unpublished paper. An old published paper. Another old published paper. A resume. Some nearby Hipparcos stars. Taps. Taps & drills. Taps & mills. Some mysteries. Some visual puns.