Hacker News Reader

Comment by zerostar07 | Hacker News Reader

I actually find the simplicity of computing devices staggering and beautiful. If the earth was wiped out tomorrow, one would only need to remember what a turing machine is and how to make a semiconductor to rebuild them.

Also, the rant becomes too broad to have a point.

View on HN

Good luck with that.

Just making glass is going to take you a while.

Making semiconductors at a scale that you can start thinking of building a working computer will take you quite a bit longer.

It's simple in principle. It would be far easier to stuff a bunch of people in rooms with paper (assuming you could make that) and pencils (ditto) and have them execute your program.

but that's how it started! http://xkcd.com/505/

Among other things, "how to make a semiconductor" is like that joke about how to make a million dollars: "First, you take a million dollars..."

You're going to spend years just bootstrapping your metal refinery to the point where you can bootstrap your machine shop to the point where you can bootstrap your chemical industry to the point where you can produce undoped silicon of sufficient purity to make a decent integrated circuit. Sure, maybe you can slap together a handful of cheesy transistors to rival the first-ever transistor (Google up the picture of the first transistor; it's hilarious; you can practically smell the hot glue wafting from that picture) but it takes a bunch of transistors to make a computer, and you'd probably like your computer to get through three whole clock cycles before melting.

You might actually be better off just recapitulating history by starting with vacuum tubes. You'll need to master glassblowing and vacuum pumps to build semiconductors anyway, and tubes have lots of handy uses. (You could build an oscilloscope, which you will also need.) But you'll need to bootstrap your metal wire industry (generators aren't that hard to make if you can get hundreds of continuous feet of insulated wire...) and note that you'd better be very careful until you've had time to bootstrap your plastics industry; there are scary stories from the early days of TV when they had high-voltage circuits but nothing to insulate the wires with but varnish and paper...

Jeri Ellsworth managed to make transistors in a pottery kiln, starting with modern wafers. She said it took three years of research and interviewing graybeards who were working on it in the 1970s. Makes me hope she wrote that stuff down.

That sounds about right.

But "build a transistor, given a wafer of pure undoped silicon" really is isomorphic to that joke about using a million dollars to make a million dollars.

The wafer is the hard part.

That's why I suggested making glass first, it's a good primer on the kind of gear that might one day lead up to the tooling you'll need to make silicon pure enough to make a transistor.

You'd be much better off choosing a road that leads via relays, that's just metalworking and some coating. Vacuum tubes would still be an order of magnitude easier than silicon.

Think sandy beach + pit mine, before you turn that into a working transistor you're going to be busy for a while.

> You'd be much better off choosing a road that leads via relays

One Portland State University professor built a relay computer, with an ALU:

http://web.cecs.pdx.edu/~harry/Relay/

Quite impressive to see it in action.

Thank you for that link. I took the liberty of re-posting it separately, it is one of the coolest things I've ever seen built. Besides the memory (and that would just add another couple of 10's of thousands of relays for a usable machine) that looks like the shortest direct path between 'man on a beach' and having a working automatic computer with a practical degree of reliability.

"practical degree of reliability" seems somewhat debatable; it apparently burns out components fairly often when running. IIRC it uses quite a bit more juice than it needs to, which might have something to do with it, so perhaps it could be made more reliable.

Relay contacts tend to spark when they engage each other. You can get rid of most of that sparking by adding diodes but that would be cheating (semiconductors again), to a lesser extent using small neon bulbs across the coils (that will still take them up to 90V though).

With 'practical' I meant that you would be able to run programs to completion for for instance numerical problems.

Vacuum tube computers suffer from different problems (such as tube filaments burning out) and I really wonder whether a well designed vacuum tube computer would be more or less reliable than a well designed relay based computer.

Vacuum tubes got awfully small in their final days but the voltages involved and the heat make me believe that relays would probably be more reliable, but likely also more expensive and much slower.

A capacitor will deal with the sparking -- they've been used (under the name condenser) in conjunction with points and mechanical distributors in internal combustion engines for years.

This all presupposes a memory of methods to create reliable electricity, though -- including rectification and filtering, if that memory involves alternators. A purely mechanical representation of computing would probably be more practical in a forced Iron Age reboot of society. A treadmill-powered version of Photoshop would be a bitch, of course, but number crunching (at the level needed for serious societal progress) would come fairly quickly.

The relay computer does use a capacitor: a huge 1F capacitor.

That's to stabilize the power supply, not to help with avoiding contact erosion.

Relays!

I'm ashamed I didn't suggest those first. One of life's most precious moments was touring the MIT Tech Model Railroad Club during the last year of their old switching system's operation. It was built around some old relay-based telephone switch the size of a couple of refrigerators, and the whole thing made these awesome clacking noises. It also emitted this constant, periodic metallic click, like a metronome, which I was told was the clock.

It's only simple and elegant on paper.

Computer Science is beautifully simple.

Computer Engineering is a gory mess.

What about computer engineering do you feel is messy? To me, it all comes down to basic physics, arranged rather elegantly.

And yet it took 50000 years worth of accumulated Homo sapiens thought to come up with a formalism describing universal computation and a model of the Nature accurate enough to understand semiconductors.

yes, trial and error is a lengthy process

"In theory, there is no difference between theory and practice. But, in practice, there is."

One of the interesting points raised in the article is that even if we had the tech to shine UV light onto plates of silicon, we'd need to reinvent a previous generation of computers so that we could design modern chips to fabricate.

There are a few abstractions that can take you a long way towards understanding ... basically the standard computer engineering curriculum: TCP, Posix, C, the von Neumann architecture, digital logic.

In my opinion, though, a computer engineer is better served working in the other direction :) Or, perhaps, from both ends at once--that's how we did it at school, and it worked out OK.

This comment is for all the decedent comments worrying about silicon wafers.

Computers were possible long before electronics, see Babbage.

Computers were built long before integrated circuits and silicon wafers, see Vacuum tubes.

http://www.homebrewcpu.com/

User/data segments, 16bit data operations, user/supervisor modes, address translation, mmio and DMA. 8meg address space, 128K per process. Runs Minix (Precursor to Linux).

200TTL logic chips. How quickly do you think you could build that? My bet, is if you had a dedicated group of people within 2 years you'd have a mhz computer again.

It's easier the second time around.

EDIT: http://www.holmea.demon.co.uk/Mk1/Architecture.htm Homemade apollo guidance computer http://klabs.org/history/build_agc/

TTL was several steps up from ordinary transistors, I don't see how that would be a valid shortcut. You'd still have to make silicon first, as well as a process to create integrated circuits. Simple ones, but still.

Reply on news.ycombinator.com