these days i spend most of my time trying to trick my university into awarding me a degree by designing digital machines in hardware description languages and drawing them out in fancy CAD programs, transistor by transistor. designing CMOS circuitry is a fantastically hard problem necessitating a fantastically complex solution, and i am sick of dealing with things that are fantastically hard or complicated
let’s take a look at some ingenious, simple things
this is a tricam. it is comprised of exactly two components pictured above. it is no more complicated than it appears
i have never, nor do i ever plan on, climbing a mountain. i know nothing about mountain climbing
people who climb mountains frequently need to anchor and re-anchor themselves to the mountain so that when they fall, they stay stuck to the mountain and don’t splatter. this is problematic as they must carry numerous anchoring devices with them as they ascend which contribute to their mass. they also contribute negatively to the climber’s bank account as mechanically advanced anchoring methods such as spring loaded camming devices tend to be exceedingly expensive
the traditional (read: old) method of doing this involved pounding metal stakes into the mountain and hoping they’d hold. the people in charge of protecting the mountains often do not want weird hippies-crossed-with-X-games-athletes punching a bunch of holes in the mountains every time they want to climb the mountain, so this method has fallen out of favor
what do you do?
you use a tricam
the tricam works in two ways. let’s look at the first, boring way:
you shove this shit into a crevice in the rock. your dumb ass hangs from the rope-y part exerting a large downwards force (red arrow). the tricam, wedged in the rock, responds with an equal & opposite friction force as it cannot easily pass through the rock. yawn
the second, interesting way is ingenious:
we assume the previous setup: this shit is wedged in rock. the rope-y part attaches to the top of the cam and sits in a groove cut laterally in the cam. this allows the downward asshole force (red) to be translated into a strong torque on the came (blue). the cam “wants” to rotate per this torque, but cannot as it meets the mountain surface
what you get here is an anchor that digs into the mountain (securing you) according to how much downward force is applied. genius
this thing costs me $10 and weighs about half an ounce. compare to
moving on
it is only a very recent phenomenon, historically speaking, that european countries have ceased trying to exterminate each other (for the most part). volumes upon volumes of historical texts outline the brutal forever-war occurring just a mere century ago. what i’m about to talk about has to do with one specific incident in this anthology of bloodshed occurring in france during the middle ages
as we all know, france is full of assholes and really really good programmers. during this period there was also a population of disgusting slovenly burgundians and hideous dutch residing in some-fuckin-part-of-france that the french didn’t take kindly to. they were known as the flemish. just like any population at this time, you had a 99-to-1 mix of starving stupid peasants catering to genuinely evil aristocrats
the french jerks hated the flemish and wanted their dope cities so of course they waged a bloody, brutal war against them. on the french side, you had a relatively small number of well-trained knights on horseback clad in all sorts of fancy french fuckin armor. they were super french about it, is the best way i can put it
on the opposite side you had bumbling illiterate peasants who didn’t much care about borders and chivalry and all that crap and just wanted to protect their shit. they weren’t soldiers, were not in a military, were poor, and did not have access to any real armaments
they took to the primitive crap laying around their fields and invented the goedendag:
they owned the shit out of the french jerks with this simple spear-lookin thing. a bunch of idiot dirt farmers obliterated french knights who did nothing but train and do gay shit like read books all day
it was called the goedendag because “goedendag“ sort of meant “good day!” back then and i really cannot tell you how funny that is
just like the tricam, this shit is dead simple. you have a pokey spear part at the tip. behind that, you have a big fat flared base made out of iron or whatever they could find that was heavy. all of this sat at the head of a really thick wooden shaft
here’s what would happen: french jerk sees flemish idiot. flemish idiot brandishes what looks like a caveman speer. french jerk chuckles to himself arrogantly all french-like. french jerk charges flemish idiot
before they collide, flemish idiot genius plants back end of spear-thing into ground and aims it at the knight’s horse. the mass of the spear-thing plus its angle into the ground/towards the knight guarantees its not going anywhere. the horse slams into the pointy end of the spear-thing and gets fantastically injured. the big flared base thing prevents the horse from impaling itself further through the spear, which prevents the horse from mauling the fleming
at this point, the horse freaks the fuck out as horses tend to do. this throws the french asshole off the horse onto the ground, which is the worst place to be if you’re wearing a bunch of heavy armor. the flemish idiot proceeds to beat the knight senseless with the mass of the spear club-thing i guess, now. armor like that is really only useful against slashing and piercing, but when you start mauling someone in it you leverage the fact the armor is heavy and moves independently from the dick inside it. you break a lot of their bones real quick
this stupid spear-thing was ridiculously, ridiculously effective against the french. the french eventually won that war, but not before suffering embarrassing losses at the hands of, again, dirt farmers. this is analogous to my spindly scrawny ass taking out a team of navy seals with nothing but a football helmet, a sock full of quarters, and a doctor pepper
computers, in the context of this question, can be understood as networks of wires that connect a handful of simple components (capacitors and resistors, maybe an inductor here & there). electricity flows through these wires, but that flow isn’t exactly like the flow of water through a pipe. the drift velocity of the electrons is actually very slow and wildly fluctuates with temperature (which is simultaneously being generated by the electricity moving through a resistive substrate) which dictates you model drift velocity as an incredibly complicated differential equation that - yadda yadda yadda none of this matters. you can’t measure time reliably with simple electrical components is the take-away of this paragraphs
so how do computers keep track of time? a 2.3GHz CPU, which is around the frequency of the CPU in the computer you’re using to view this post, has a clock in it that ‘ticks’ 2.3 billion times a second with extraordinary accuracy. that is quite a feat!
how is this accomplished? with crystals. ‘crystal’, referring to any compound whose atomic structure follows an extremely precise and constant geometric lattice. look:
this is what table salt looks like on atomic scale. it’s made up of two atoms in alternating patterns to create a very consistent & simple shape. this pattern is repeated, with almost zero divergence, to form crystals your greasy ape eyeballs are actually able to see. it makes for an extremely reliable & calculable substrate that occurs naturally
now
these crystals exhibit something called the piezoelectric effect, which almost everything does. when you apply mechanical stress to (pretty much) anything, you elicit a momentary (or permanent) change in shape. that change in shape shifts around once-still geometries of atoms and disrupts the incredibly delicate orbit of electrons surrounding them. it follows that a disruption of a once-stable & homogenous body of electrons results in a momentarily unstable & uneven distribution of electrons. consider this as a dipole, or any system where there’s more negatively-charged particles in one region than another, creating a positive and negative pole. just like the compacted air molecules in a balloon “want” to explode outwards because of pressure, the electrons in a dipole “want” to move from an area of dense negative charge to an area of less dense negative charge. “less dense negative charge” is a fancy way of saying “positive charge”, by the way
this concept is known as a voltage differential and is the core driving principle describing how “electricity” can “flow”
so
to summarize, when you squeeze something in your hand you are mutating the atomic structure of that thing, leading to a dipole & voltage differential. inversely, when you apply a voltage to a piezoelectric medium, you get a momentary mutation of its atomic structure, which then causes another weaker voltage differential that causes another weaker mutation that causes so on and so forth
if you kick off this type of reaction with a single pulse of electricity, your piezoelectric medium just switches (or oscillates) between shape mutation & voltage generation over and over with less and less vigor each time until it finally settles back into a stead state:
however
if you measure the voltage produced by the piezoelectric medium, and wire it to a powered amplifier and feed the signal back into the piezoelectric medium, you arrive at a situation wherein the medium oscillates faster and faster due to the amplified feedback thereby creating a stronger voltage which goes through the amp and back to the medium to create and even stronger voltage and so on and so forth. the waveform doesn’t just taper off and die like in the image above
eventually, you reach an equilibrium where the increased voltage feedback into the medium no longer creates a faster oscillation & stronger voltage, and the voltage going out the medium into the amplifier doesn’t change, resulting in the output of the amplifier not changing easier
you have arrived at a steady state
the output waveform is no longer tapering off and dying. it remains constant so long as the amplifier is powered. it’s also oscillating extremely quickly and is dampened by the capacitance (”elasticity’) of the amplifier. so your output waveform looks like this:
instead of this:
cool, ok
this clean-looking waveform two images above is called a sin wave. it’s the most fundamental waveform and every aspect or variable describing it is constant and even. most piezoelectric compounds found in nature tend to be impure and uneven and extremely hesitant to provide such a waveform, even with the amplifiers “elasticity”, except for one category of compounds:
crystals
the extremely consistent and even atomic structure of crystals provides a correspondingly consistent & predictable piezoelectric effect which can be exploited to produce nice smooth sine waves like above
this is nice, but even with an analog sinusoidal waveform at an almost-perfect and stable frequency is not enough for a computer to be able to tell the time
the critical property of sine waves are that they increase from zero at a linear rate which decreases once that wave has reached it’s positive peak. for an infinitely short point in time, the waveform is flat (at the peak of the wave,) then rapidly falls back to zero & then the negative peak where the cycle is reciprocated. rising & falling, over and over again.
computers only understand ‘1′s and ‘0′s, though, which is way fewer things represented in a sine wave. we build a special type of transistor (light switch) that follows the waveform very precisely, and toggles its switch only once the analog waveform appears to stop heading in the direction it previously was heading and starts to go into the opposite direction. this is at each peak & valley of the wave:
in the image above, just replace “crest” with “peak” & “trough” with “valley”.
now we have a transistor (light switch) that toggles to one of its two possible states (on or off) at a very even frequency. with a 1Mhz sine wave (sine waves goes up and down a million times a second) you now transistor switching on & off a million times a second, or once every microsecond
cool! the “switch” being either on or off (with nothing in between) is where the concept of binary states comes from, a switch in the on position is analogous to a ‘1′ & off is analogous to ‘0′.
so now that we have a digital circuit that follows an extremely accurate timeframe because of shit happening in real life with magic crystals, we’re done right?
no
this consistently-alternating pattern of ‘1′s and ‘0′s is used to create an identical waveform, one that stays at a high voltage for an even, predictable period of time and then goes to zero, seemingly immediately, once that period is over. and then back to high voltage, then low, etc. looks like this:
this is called a square wave, or more contextually-specific, a clock signal. it’s amplified and provided to many hundreds of components on your computer with very strong impedance, meaning that despite it being provided to a bunch of shit that’s sucking up the strength of the signal, the signal doesn’t deform or de-amplify. these alternating “on-off” signal is what drives any kind of electrical component
such as a simple component that counts each on-off cycle. this is crucial. since you now have a signal that undergoes a quantifiable change at a very constant rate (which you know), and you have a device that can count those changes, you can use the sum total number of changes to describe the amount of time that has elapsed in the real world. for example, with a 1MHz clock (ticks a million times per second or once every microsecond) and a counter capable of holding any number from 0 to 1 million, you can now keep track of how many seconds has passed
we’re still not done because you still can’t keep track of amounts of time above a second, only the time between each second. so what do you do? add another counter of course. it counts the number of times your zero-to-one-million counter has rolled over from 1 million back to zero, effectively counting the seconds.
in reality, you have a counter that translates from billions-of-ticks-per-second to perhaps millions-of-ticks-per-second, then from millions-of-ticks-per-second to thousands-of-yadda yadda yadda
these hardware counters are super expensive though, so you don’t want to keep adding counters to describe longer and longer timespans. the computer i’m typing this long-winded answer on is capable of keeping track of time until the end of the god damn universe and a trillion-billion years after that. how?
eventually, as your counter chains move from extremely fast ticks to slower and slower ticks, you reach a threshold where the circuitry driven by the fastest counters is suddenly able to keep track of the slower, longer counters and you cross a boundry on which my entire university education is premised upon:
the hardware-software interface
your 1′s and 0′s are no longer stored as inaccessible charges inside your counters. they’re stored in memory of which you have a lot of. in fact the amount of memory/bits/1′s-and-0′s taken up by the following group of characters:
12345678
is exactly equal to the size of memory used to keep track of the number of seconds between the beginning of time (january 1st 1970, a story for another time) and now. it has enough space to keep on tracking the seconds until the end of the universe (and long after that too)
once these time intervals become available in memory, you can do very easy manipulations & computations based off them and therefore keep track of time within huge timescales without the cost of adding more hardware counters. again, this is because the shit that does those calculations & manipulations works faster than the relatively-long increments of time. those calculations complete before another relatively-long increment of time has passed such that the internal circuitry is always a step ahead of what’s happening in the software realm. realistically, this means that while my computer operates physically on the timescale of billions of ticks per second, software i write can at most only deal with timescales of millions or tens of millions of ticks per second
you can take your boots off now
more dogs have been to space than people who genuinely love you
lets grab a rusty screwdriver and plunge into the chest cavity of the lips machine
i talked a little about the ridiculous power supply this thing has. removing it is not straightforward. i rooted around in the chassis for a while before i noticed two heavily anodized screws (relative to the other screws). aha. the enormous amounts of current going through this thing had to get around somehow, and the anodized screws would seem to be the culprit. they mount the busbar to the chassis, i popped em off and out slid the PSU:
what’s that PCB doing on the back and why is it as thick as a pencil? who knows. why you would need so much logic on a power supply is beyond me
900 watts! ians staying warm this winter.
the rest of the power supply is not very interesting. next we move on to the disks. they are secured by just two thumb screws on the back panel. twist those off & disks slide right out:
they are enormous. manufactured by maxtor, these two XT-2190 disks hold almost 200 MB each!
just in case you forgot we live in the literal jetsons future:
these disks operate using the same principle as modern hard drives do, inside these coffee can sized enclosures sit round, magnetic platters which spin under a read/write head
these disks are also dumb as hell & have no onboard controller. they simply pass an analog waveform resembling the one felt by the read/write head as it passes over the magnetic platter surface. it’s the job of the disk controller on the machine to decipher this waveform into 1′s and 0′s. these disks probably use MFM-style encoding which is a good thing for reasons i am not going to explain yet.
printed on the drive itself is a table associating byte counts to cylinders and cylinders to heads. the platters sitting just an inch below the sticker pictured above sort of look like double-sided CDs stacked on top of each other with about a quarter inch of clearance. each platter has two sides, and each side is considered a head. if you can imagine these CD-looking disks as vinyl records, each “groove” that circles 360 degrees around the center of the disk is a cylinder. and each of those cylinders contains a certain number of bytes. if i were to pop open openbsd’s disklabel(8) tool, my modern-day hard drive would indeed report similar cylinder/head/sector information, however, they would be lies. today’s hard drives (and pretty much any hard drive made since i was born) do their own housekeeping and report carefully-crafted disk geometries to the operating system to trick it into writing/reading data from it in a way that tends to be very efficient due to hardware design choices of the drive
the numbers in the above image are not lies, which is pretty cool as i could write data to the disk and then point with my finger where on the platter that data was encoded. those numbers you see above are also unique to this drive! why?
we used to be not very good at manufacturing perfectly round, evenly-weighted and chemically pure disk substrates. we were even worse at making & coating the disks with magnetically-sensitive surface films. every time you made a disk, you had to put it through a huge expensive machine that would test it & figure out all the places on the disk that weren’t totally FUBAR. out of the remaining potentially-OK disk surfaces, you then had to ascertain exactly how fucked up they were and make a judgement call about how much data you can pack into them safely based on the fucked-upedness
you see at the bottom “37 defects”. it came out of the factory with that printed on it. lol
okay now we get to the fun stuff:
“NAT/FEP CARD”. this is the front-end processor
this thing is sort of the analogue for modern-day motherboards. the disk controller, network controller, console controller, any kind of peripheral IO plugs into this
are you ready for what’s next? oh baby
the central processing unit of the symbolics 3620 lisp machine!
even as i type up this post, i’m speechless. not going to say much about this one & instead leave it to your imagination. the reason the ICs look like they’re capped in solid 24 karat gold is because they absolutely are capped in solid 24 karat gold
