I have a vague memory of once upon a time reading a paper alleging that if you could start everything over again from the beginning, that somehow ternary was more efficient than binary for a digital architecture, something having to do with power usage. But other than that it just sounds like nonsense.

Love to see that paper; seems like nonsense to me. If someone came up with a more optimized transistor architecture than CMOS I'd love to have a look at it.

Oh! I remembered more details about it if it helps. It has to do with e being the "most efficient" numbering base. The reason e is the "most efficient" has to do with wanting to minimize both the size of the alphabet of symbols as well as the number of digits in an average computation. If you allow for a non-integer base, and a non-integer number of digits, the optimum base is e. e rounds to 3. ;-)

Now, full disclosure, I'm not entirely sure I believe in this, I'm just citing sources but check these out:

http://bit-player.org/wp-content/extras/bph-publications/AmS... https://hackaday.com/2016/12/16/building-the-first-ternary-m...

Also, Don Knuth likes balanced ternary, and, well, I guess that counts for something.

(again, personally I have no idea if I think it's nonsense or not)

Yeah, it's complete bullocks. Rest assured, if anything was better than binary for _logic_ then it would already have taken over the world. Binary have a lot of wonderful properties (not least noise immunity) and deserves a lot of credit for the success of modern technology. Note, the ENIAC wasn't binary! It was based on decimal.

However there are places where we look beyond binary; most notably in storage where data density is job one. The vast majority of FLASH is using MLC (4-values) or TLC (8-values) which need heroic circuits to recover the data. But note, these are 2^2 and 2^3, still binary based.

It is not "complete bollocks",it just not as efficient as it appears to be. The reason for the (supposedly) higher efficiency of ternary logic is so called Radix Economy (wikipedia has an article). Properly implemented ternary (with a split power supply - positive, negative and ground) would have the same noise immunity as binary.

"It is not "complete bollocks",it just not as efficient as it appears to be."

That's a contradictory statement. If it was as efficient as claimed it wouldn't be bollocks. But it isn't, thus, bollocks.

"(with a split power supply - positive, negative and ground) would have the same noise immunity as binary."

[citation needed]

I'm not a EE, but my understanding from EE friends is that isn't true.

> Properly implemented ternary (with a split power supply - positive, negative and ground) would have the same noise immunity as binary.

Only if you increase the supply voltage such that: V+tern - GNDtern == V+bin - GNDbin == GNDtern - V-tern