why binary system is used in software-hardware?

could not be ternary, or 4-ary or 16-ary?

is it a technical impediment?

even if it would be phisycaly possible it would be dificult to work with?

what would be the benefits of such system?

faster? biger capacity?

or it would be totally impractical?



please give some ideas...thanks.

Designing the hardware is not easy.

We already have MLC NAND FLASH which is effectively 4-ary.

Plus see here for an organic ternary memory: http://www.physorg.com/news190451302.html

What would be the Boolean equivalent operations in a ternary logic system?

in the early ages of computing, there were analog computers, based on operational amplifier.

instead of beaing Nary, it was analogic, then, with a thearical infinite resolution of the datas.

the problem was the storage of theses analog datas.

maybe one day, there will be a reborn of analog CPU... with some DAC/ADC as R/W datas on binary ram.

but ADC are teribly slow, or low res. for video, it is a little fast, but low res, only 8__bits resolution.
for audio, it is high res (up to 32bits) but slow.

i have an idea about a high res, fast ADC, but never tested with success...

but binary is not dead, because of the ON/OFF feature, the simple switch or interrupter, that can be combined in order to build a computer from only electromechanical relays, if a end of world occurs, the future humans will be forced to reuse this technique.

Very simply because that's how electric current works. Either it's on or it's off. Either the capacitor (e.g. every cell in a DRAM) is charged or it's not.

Well that is how electric current works in binary. It is still valid for ternary, we just add a third state: no-charge, half-charge, full-charge. No change needed in the capacitor. The main problem is to make reliable detectors. Ternary is not impossible, just not as practical as binary.

Whats wrong with binary? I quite like binary

revolution,

Balanced ternary looks pretty practical. Inherent signedness and unambiguous rounding probably are the most pleasant properties. And capacitor can be in three states too: almost discharged and charged in two distinct polarities.

They use switches anyway: 0 and 1 ie. "no-charge" and "charge".
Also their multiplication looks difficult, whereas in binary we do like in decimal:

Yeah, okay, good luck with designing a circuit to handle that nicely. MOSFETs get quite upset with negative voltages. Knuth is looking at it from a programming perspective and says it is no problem. But people like AMD and Intel will looks at from a design perspective and run away crying.

Are they? N-channel depletion-mode MOSFET even requires negative Vgs to turn it off.

Yep, but what happens if the gate goes positive wrt D & S? If you give it positive voltage, i.e. the negative voltage of what it normally wants, then things go all haywire. Anyhow, it is not impossible, it can be done with complex wiring. Although it is very very inconvenient, and it needs lots of transistors to implement even a simple function. The only really clear advantage would be in storage, i.e. memory, where the sensors are common to all cells. There, the overhead to make the multivoltage circuitry is paid back by the increase in storage capability. MLC NAND already does this. But it comes at great cost. The extra storage per unit area also has the drawback of lower write/erase endurance and more read errors.

By the way, there was at least one working mass produced computer based on ternary logic:
http://en.wikipedia.org/wiki/Setun
And, it was cheaper than the binary equivalents of the time.
My computer history professor talked a lot about it, but i can't remember any details by now.

anyone pointed out that a ternary digit will be named TIT, it is interresting anyway.

for example, a 2 TITs computer will have 3^2 combinaisons, then, 9 combinaisons.
cool. looks like sexual feature...

i want a 2Tits computer for my own use.

more seriouslly, a Tit is interresting because the 3 states can be easy defined.
negative, off, positive. and is near the human logic. frequentlly, we (humans) answer in at least 3 ways for a elementary question.

do you believe in freedom, for example, will have 3 answers:
1/ Yes,
0/ I don't know,
-1/ No.
it is ternary logic.

an other example, more political, is:

can you make something for the unemployement of young people?

1/ Yes we can,
0/ Maybe, give me more time,
-1/ No it's impossible.


maybe ternary is the better way to compute.

a simple way is to see how many N^M combinaisons possibles for N and M binom.

N=2, M=3 => 2^3 = 8
N=3, M=2 => 3^2 = 9

N=5; M=3 => 5^3 = 125
N=3; M=5 => 3^5 = 243

ternary is the best compromise to build a computer. but now, it will be very hard to create all the modern software for such a system, from scratch.

and how to represent datas in ternary? 0,1,2?

for example, a simple ternary value on 3 Tits:
012t= 3+2 = 5
222t= 18+6+2 =26

and the representation of ternadecimal? leading to 27 combinaisons for a "nibble"

Trit, not tit. Yeah yeah, I know, you were just trying to be funny.

ok, all my effort to be funny are dead. i'm not funny i know...

That is a false trichotomy.

revolution:
what do you think of such a circuit?
do you have the equipment and the time to test it?



theorically, it can be a very fast Analog to Digital converter.
but in practice?

edfed: Search for successive approximation (SA) converters, well proven, reliable, low power and fast. Your circuit is much the same except you have no intermediate latches. And your transistor count is very large because you duplicate all the sense circuitry multiple times. It would require a lot of time to calibrate properly, each conversion block will have to be very well matched to the others else you will get conversion errors (missing codes and/or duplicate codes).

[edit]If you need a very fast conversion, then perhaps the flash converters will better suit your needs. They can run at 100s of MHz, but they are power hungry and have low bit counts. Usually used for video capture circuits.

i still know all the ADC types, and all have some defects. flash converter use a lot more transistor than this idea.

SA converter use a cycle based conversion, one clock = one bit.

in my idea, only the propagation delay across Nelements is considered.
with high speed transistors, it can be very fast to convert on 32 bits. more fast than a SA, delat sigma, or else converter.

it is just an old idea, but i ask myself the way to do it... and test it of course.

Well it might be fast but it won't be accurate. How will you calibrate it? How will it respond to temperature changes? How will you know when the conversion has settled? How sensitive will it be to power supply variations?