Atomic I/O letters column #85Originally published 2008, in Atomic: Maximum Power Computing Last modified 16-Jan-2015.
Is this really true, and why? Or is it just a case of current perceived wisdom that will, as have many others, be disproved by advances in technology?
And how does a 32nm transistor compare to the universal SI unit of small things, the width of a human hair? Or even the length of a football field?
This old four-inch chip-wafer boule top is about 101.6 million nanometres in diameter.
A human hair's about 60 to 80 microns in diameter; a micron is a thousandth of a millimetre. A nanometre is a thousandth of a micron, though, so a 70-micron hair is 2187.5 times as wide as a 32-nanometre object.
The pylon-to-pylon span of the Sydney Harbour Bridge is 503 metres. If you scaled up the hair so it was 503 metres wide, the 32-nanometre object would now be about a hand-span across.
I keep saying "object" rather than "transistor" because there are weasel words involved when semiconductor companies talk about transistor size. Nanometre numbers are always actually telling you the "feature size" of a given manufacturing process - basically, the smallest thing the chip fab can make on the chip. The smallest feature may or may not - and usually won't - constitute a whole transistor by itself.
There's not actually any standardised way to measure the size of the transistors on a chip, so you should take all such numbers with a grain of salt.
That said, there certainly are limits to the feature size you can create with photolithography, the current chip-fabrication technology. It's a process basically quite similar to the way most printed circuit boards are made, only much, much smaller.
When you're using basic photolithography - light shining through a stencil mask to print a pattern onto photoresist - the finest detail you can print is dependant upon the frequency of the light. The higher the frequency, the finer the detail. This is why the light used for chip photolithography started out in the near ultraviolet, and has moved to deep ultraviolet.
At the end of the ultraviolet spectrum, though, air blocks the light, so you have to start doing fabrication in a vacuum (this is called extreme ultraviolet lithography, EUV, and is so far still experimental). Go even further than that and your "light" becomes X-rays, which can also be used for chip lithography, but aren't commercial yet either.
Mainstream chip fabs have, therefore, been stuck somewhere short of the end of the ultraviolet for a long time. They've been using other tricks to keep pushing feature sizes beyond - FAR beyond - what previously appeared to be physically possible.
"Immersion lithography", for instance, is the hot new thing for current chips. It means doing your lithography underwater, to allow your lenses to have numerical apertures above one, which increases their resolving power. (This is also why oil-immersion microscopy exists.)
The constant shrinking of the supposed limit for photolithography has been kind of like the development of analogue modems. Various people sagely announced that, OK, it turned out that 2400 bits per second wasn't the modem speed limit, but trust them, nobody would EVER beat 9600bps, for sure. Similarly, it at first seemed to stand to reason that feature sizes could never be smaller than the wavelength of the light being used, which meant that anything below a THOUSAND nanometres was obviously impossible.
After a while, though, you run into serious information-theory limits, and you really and truly cannot go any further with a given technology no matter what you do. That's what happened with "56K" modems (which could only do 56 kilobits per second for downloads, not uploads, and which practically never managed to connect at 56K anyway), and it's going to happen, probably pretty soon, with photolithography too. There's no point worrying about what the actual limit number is, though, given that there's no standardised way to measure manufacturing process feature size anyway.
There's no laws-of-physics reason why feature sizes shouldn't shrink right down to atomic dimensions - a silicon atom is less than a third of a nanometre in diameter. To do that, though, photolithography is going to have to go away, and we're going to have to switch to one or another finer-grained form of "nanolithography".
These Colecos'll burn up on you like THAT!
I have a friend who recently bought a new Media Centre PC (from y'verygoodfriends at Aus PC Market). The usual stuff - dual-core Intel processor, nice hi-fi-styled case, reliable PSU, low-end ATI video card, two hard drives striped for speed, and a dual TV tuner card. It works splendidly and looks good on his LCD telly.
One problem - he's paranoid that if he leaves it on overnight, it will catch fire.
I would have thought this was, well, unlikely. Even if a CPU overheated or something, I would have thought it would immediately short-circuit and thus stop generating heat.
So, is he being overly paranoid, or am I being overly cavalier?
Um... does it often catch fire during the day?
Computers are pretty low on the fire-risk scale. A PC certainly will hang hard long before any processing parts of it get to ignition temperature, but the PSU will probably keep running, and even a small PSU delivers more than enough power to kick a flame out of a short circuit.
There's not much flammable material inside a PC, though, so all you get even in the most impressive failure situations - +12V line sheared off when you slammed the case closed, expansion card jammed into the slot wrong - is a brief flame from the problem point, followed by a bad smell and profanity.
I mean, think about it - what would happen if you put a couple of handfuls of dry leaves inside a PC, lit 'em, and closed the case? They'd burn out, the PC would be a mess, and that'd be about it. You might manage to get some embers coming out of the exhaust fans if the PSU was running at the time, and those could perhaps light the curtains or something, but this is what it'd take to get that to happen. There's nothing naturally inside a PC that's at all likely to do it.
A defective PC can also serve as an ignition source if you've got flammable vapour floating around the place. But your fridge clicking on, or a wall switch, is much more likely to do that.
There's practically no electrical device, down to battery-powered transistor radios, that can't in theory start a fire. But any normal house contains numerous more likely candidates than a PC, none of which keep most people awake at night sniffing for smoke.
Having recently modded up my "officyPCthatIgotoffthebackofatruck" (okay, so I put some lights in it and some shiny dials on the front), I seem to have run out of room for hard drives, due to a fat fan for the front intake vent. It's cooling my PC nicely with a little wind-tunnel type effect, however it does mean I can only fit ONE hard drive in my case...
Now far from doing the sensible and cheap thing (I am in my late teens after all) which would be to buy a thinner fan, I've got the idea into my head to construct my own external hard drive enclosure... out of an ex-military ammunition box!
Think of the wonders, opening the catches and swinging out the lid to reveal blinking lights and enough storage space for all my wildest needs! *Ahem* The only problem being I have NO idea what kind of devices I would need to support multiple hard drives, let alone hook em up to my PC in a simple manner.
Simplified, my plan is as follows;
1. Get hard drives, hard drive rack, ammo box.
2. Install rack into ammo box.
3. Install hard drives into rack.
5. Awesome looking external HD enclosure!
Any advice towards improving this plan would go a long way in my Conquest of The Earth, and should you assist me I will ensure you hold a high rank among the survivors when they are toiling in my vast plutonium mines.
I don't think you'd need to buy a whole rack device to get the drives in there - though if you want this to be a USB box, that could give you a quick one-step solution.
The ugliest but simplest way to run the drives would be some sort of external SATA arrangement, where a forest of SATA cables come out of the back of your drive box and into your PC. There's a real external SATA standard, but ordinary SATA data cables can be up to a metre long (a bit more if you push your luck), and they're thin and manageable enough that you could just about pull off wiring the drives all individually from a multi-port controller in your PC.
For power, you could either run a bunch of separate SATA power leads too, or wire up a multi-Y-adapter that runs from a single SATA power input, or something. I don't know how many drives you could expect to run from one SATA power lead; a beefier power source would be a good idea, and you could whip one up that took "Molex" input or something. Note that the standard SATA power plug pinout includes pins for 3.3V, but desktop drives don't use them, which is why (single-plug) Molex-to-SATA power adapters exist.
(After this page went up, a reader pointed out to me that SATA port multipliers would simplify the wiring a lot. I didn't mention them originally because they can be a bit expensive - this one for $AU143 delivered was pretty much on par - and you need a SATA controller that's compatible with them or they won't work at all. That used to mean some $500-and-up RAID card, but a lot of cheap SATA controllers these days now say they're multiplier-compatible, and you can get rather cheaper multipliers from the usual cheerful Hong Kong eBay dealers, too. I don't know how many motherboard SATA controllers support multipliers, but you're going to end up paying for controller hardware anyway when making a storage box like this. Multipliers are a great option technology if you don't need massive speed.)
Option two would be to use a collection of USB- or FireWire-to-ATA adapters. The delivered price for those items on eBay these days is no more than $US15 each.
Ideally there'd be one out there with a couple of SATA ports on it that you could use simultaneously and no other rubbish, but I don't know. The standard type has two or more plugs for different types of drive, but you can only use one connector at a time. The universal connectors are nice, though; they'd make it easy to slap in any old parallel-ATA drives you've got lying around at first, then swap them for nicer SATA units later.
These adapter cables always come with their own separate power adapters to power the drive, plus extra plug adapters and such, which are very nice but would of course become a bit unmanageable if the drive count rose above about four. Frankly, I'd put all of the adapter units to one side and use a single PC PSU to power the whole shebang; with a big ammo box you could easily fit a PC PSU inside, and its fan could provide enough ventilation for the whole case.
Option three would be to find some fancy drive box that's suffered horrible cosmetic damage, scrap its casing and use its guts. Great if you can pull it off, but you probably won't be able to. Small cheap boxes seldom have much expandability beyond the internal bays they come with.
Option four would be to build a whole PC inside a medium-to-large ammo box. Any old motherboard will do for a storage device, and there are Linux and BSD Live CDs that let you make a network-attached storage box pretty easily and very cheaply.
Ammunition boxes, of course, come in a large variety of sizes, but I'm presuming you're not considering starting with one that originally held 72 RPG-7 grenades.
I've got exactly one ammunition box here (I keep a few small but very flammable items in it; thanks to your e-mail I checked its contents, discovered my flash-cotton had dried out, and re-wetted it for safety - thanks for that!).
Mine originally held 300 linked 7.62mm NATO cartridges, but I think there are boxes of very similar size and design that held 100 .50 BMG rounds or 800 5.56 NATOs. (I painted this one Danger Red a long time ago, but I didn't strip the original paint off first, so I just spent a couple of minutes lighting the box from a steep angle to try to read what it used to say on the side.)
Anyway, this size of box is usually pretty easy to find at surplus stores, and its dimensions would be just fine for quite a large drive array.
The internal dimensions of this box are about 142mm wide by 280mm long by 180mm deep, so you could get about nine 3.5 inch drives in there, standing on their noses or tails like dominoes. There'd be clearance for at least SATA and probably even PATA cabling, and a tolerable amount of air space between the drives. Eight drives could be quite well ventilated.
This type of box even has a rubber seal inside the lid - apparently they're quite popular with geocachers, for this reason. The super-tough latch mechanism wouldn't be good for drives, though. You can't even open a box of this type without jarring its contents, and the latch goes BANG when you close it, which is obviously bad.
A bit of brute force and ignorance could be employed to bend out the tight-fitting side flaps on the lid a little and similarly loosen up the catch, though. Buy two boxes - ones like this seem to only cost about $10 each even if they're completely rust-free; grotty-looking ones can be much cheaper - and it won't matter if you wreck one.
(Ammo boxes are, by the way, easy to find on eBay, but low-priced large objects like this can of course be expensive to ship. Definitely check out your local army-surplus stores first.)
The box is made from 1mm mild steel plate, which is easy to work on, and it's wide enough that the clearance on each side of each drive for mounting hardware would be about 20mm. That's plenty of room to do something with hex standoffs, long screws, or what-have-you.
One simple mounting option would be to use the 3.5-inch cages from old and busted PC cases, which can be had for free (along with various other enormously useful things like keyboards that've been rained on and 3.5-inch floppy drives full of dog hair) at council cleanup time. The width of this type of ammo box is also pretty close to that of a 5.25-inch bay, so you could also try the cheap adapter brackets, available from any decent computer shop, that let you install 3.5-inch drives in those bays.