Bit of an aside but I wonder how far tube technology might have advanced, without semiconductors intervening. In the late 1950s and early 1960s, GE, IBM, and RCA, probably other companies, were working on "integrated tubes" with many components in a single envelope, as well as techniques for easier and more automated manufacture. For example, introduced in 1959: https://upload.wikimedia.org/wikipedia/commons/b/ba/Nuvistor...
> Bit of an aside but I wonder how far tube technology might have advanced, without semiconductors intervening.
There were Compactrons. There were subminiature vacuum tubes.[1]
A one piece printed circuit board of glass, with multiple tubes, might be possible.
A glass plate made with lots of recesses, electrodes and wiring created by photo-etching like printed circuit boards, a glass plate on top, pumped down to vacuum and sealed. A low-density integrated tube.
That's what a plasma panel display is. It's an integrated array of neon lamps. In a vacuum fluorescent display, each illuminated element is a triode vacuum tube. So it's quite possible to fabricate a big array of tubes.
Maybe something like a ball grid array would work for external connections.
Probably could have been done if necessary. Density probably would have maxed out around the density of elements on on the most dense vacuum fluorescent displays. Maybe devices at 1mm scale, or 1,000,000 nm. Good enough for mainframes and minicomputers, but not microprocessors.
Would tube reliability have limited how much you could scale it up? As far as I know, tubes have a limited lifetime and burn out eventually. If you have a million or a billion of them, they might fail so fast that your computer simply doesn't work.
I don't know whether reliability is a solvable problem. Tube-based devices were once very common, so that suggests it would have been solved if it could have been.
Tubes have a limited lifetime, but high-reliability, long-life tubes were developed specifically for digital circuits.
One such development was a change in alloys for the filaments. It turns out that the filaments were made from a tungsten alloy containing silicon, and the silicon evaporates and is deposited on the cathode. The cathode has a special coating to help it emit electrons, and the silicon deposits would interfere. From what I can tell, the alloy for filaments had silicon in to make it easier to draw through the dies necessary to construct the filament in the first place, so there is some tradeoff between the lifetime of the tools used to make the tubes and the lifetime of the tubes themselves.
This is not entirely unlike the problems faced by semiconductor manufacturers—problems with impurities, solid state chemistry, and vapor deposition. I can imagine an alternate timeline with extremely long-lived vacuum tube circuits.
I recently read up on "zone melting" which was developed to purify germanium - but of course this same method can be used for other materials. Interesting that Hamming of Hamming codes fame gave encouragement to the developer of this materials process: https://fs.blog/great-talks/richard-hamming-your-research/ (search for Bill Pfann)
Say a CPU was made the size and shape of a plasma TV. You wouldn't have to run it all at once. You could make many CPUs or logic blocks and re-route traffic to hitherto unused blocks when a fault was detected.
Oof. I don't know much about tubes, do they always fail in ways that are reliable to detect? (E.g. I can imagine that detecting a burned out heater is easy, but I don't know what other common failure modes there are.) And even then there's a lot of effort to just make the system "switch over"... and a lot of extra real estate with technology that was already constrained by the minimum size bound of the tubes.
