Trinary Code

[MyndFyre]

Yes, but even if his was actually bigger in binary bits, I'm having trouble with his measure of "more efficient". If how good something is is measured in the number of possible combinations, a single trinary bit is 50% more efficient than a single binary bit, a trinary byte is 2563% more efficient than a binary, and a trinary word would be even many times more more efficient than a binary. Just like a 1 byte larger hard drive is 256 times more efficient since you can store 256 times as many combinations....

But it would take more space to store, thereby negating the added "efficiency."

[Topaz]

Doubt it. With more speed, you can make it more efficient affordably without losing usability.

[Arta]

Explain?

[Topaz]

The whole issue with Trinary code is that you have to change everything, plus more storage space because of 50% increase of voltage.

If you were to convert the current amount of space/equipment from binary -> trinary, you would still gain a 25% increase in efficiency, without losing anything in particular.

It's kind of hard to explain =/

[Arta]

What is your measure of efficiency?

I think trinary would be much less efficient. The hardware would be much more complicated and probably slower.

[Adron]

The whole issue with Trinary code is that you have to change everything, plus more storage space because of 50% increase of voltage.

If you were to convert the current amount of space/equipment from binary -> trinary, you would still gain a 25% increase in efficiency, without losing anything in particular.

It's kind of hard to explain =/

The whole issue with Trinary code is that a basic binary gate can be made with two transistors. The transistors can be made very small and inaccurate because they only have to be all-on or all-off. When you build something analog, something where transistors can be other than either full open or all closed, you need to design them with more precision. This makes them bigger.

You further cannot use the same circuit that a binary gate uses. The basic building block of a binary code computer is an inverting amplifier with "infinite" amplification. That's a "not gate" or "inverter". Inverters are used all over the place in digital circuits to amplify the signal and remove noise.

An inverter consists of one n-mos and one p-mos transistor. Depending on the input signal applied, either the n-mos will be open and the p-mos closed, or the p-mos will be open and the n-mos closed. Between clock signals (i.e. when the input isn't changing), no current will flow.


When getting technical, it's easier to think about it in terms of an analog circuit where your analog ground is at a 2.5V level. This turns binary 1 into +2.5V and binary 0 into -2.5V. The basic inverter, "infinite" amplification inverting amplifier might in practise have an amplification of say 100 times. An input of -2.5V will make the inverter want to output -250V, but the clipping of the supply voltage causes it to become just -2.5V. An input of -1.5V will generate an output of -150V which is also clipped to just -2.5V, meaning that a whole Volt of noise has just been removed.

If you were to use a trinary signal, there would have to be a middle state where the voltage was right between Vdd and Vss, i.e. 0V in the analog scale. The inverter as described above would not remove noise for that. If you have 0.1V of noise on that signal, you'll be inputting 0.1V to the inverter, which will generate 10V on the output, which will be clipped to 2.5V, which is now a logic high. The value of the bit has been destroyed.

To accurately reproduce and de-noise such bits, you need to design a much more complicated circuit, consisting of many more transistors. I haven't seen a minimal one, so I can't say for sure how big, but perhaps 5-10 transistors could do it if you were clever. Here's an example of an analog amplifier, you can see the circuit diagram in one of the first pages.


In addition to a trinary logic gate being bigger than a binary one, power consumption becomes a big problem. To generate the intermediate voltage, you need to have a bias current flowing through the amplifier at all times. Outputting the mid level voltage means that both of the output transistors would be conducting current at the same time. This would increase the power consumption of the circuit extremely much.


Another problem, connected to the power consumption one, is speed. To limit the power consumption of an analog amplifier, you limit the current flowing through the two output transistors when a mid-level voltage is being output. This limit will also limit the maximum output current of the amplifier.

The binary inverter is designed to drive a lot of current through its transistors. That is not a problem because the two output transistors will never be conducting current at the same time more than momentarily. The more current you can generate from the output of the gate, the faster the input of the next gate can be charged (stray capacitances).

To make the chip survive the heat generated, you'll have to limit the bias current through the output transistors. But then you'll be limiting the output current and the speed at which the input of the next gate can be charged, i.e. increasing the gate delay.


Conclusion: Trinary logic may be possible, and you'll have more information processed in comparison to the number of logic gates used. However, each logic gate will be at least three times as big, and your power consumption will be up perhaps 10-100 times, as will your gate delay.

Maybe you could turn a Pentium IV from a 32-bit binary processor to a 32-bit trinary processor (50-bit binary equivalent) in a 5x5 inch package, with a power consumption of 3 kW and a clock speed of 25 MHz. Would you want to?

[Topaz]

Technology improves over time, and the efficiency should more than make up for its space and expense.

My measure of efficiency is the amount of processing power and speed it can manage without burning out.

I think trinary processors are only worth it when using high end applications and/or server administration/applications.

[Arta]

Did you actually even read Adron's post? Sure, technology improves, but you're missing the point: even if electronics was sufficiently 'efficient' (whatever that means) binary would still be more 'efficient' than trinary. High end applications are exactly what you wouldn't want to run on a trinary processor.

[Adron]

Technology improves over time, and the efficiency should more than make up for its space and expense.

My measure of efficiency is the amount of processing power and speed it can manage without burning out.

I think trinary processors are only worth it when using high end applications and/or server administration/applications.

Read my post carefully. I took the time to write it, the least you can do is read it and try to understand it.

Respond to the three problems brought up in my post:

* Circuit design and complexity for analog vs digital circuits
* The requirement of idle current in analog circuits
* Gate delay in logic gates with non-saturated transistors.


After rereading my post, find the part about speed going down 10-100 times and power usage going up by 10-100 times (i.e. processing power going down 10-100 times).

[R.a.B.B.i.T]

Technology improves over time, and the efficiency should more than make up for its space and expense.

My measure of efficiency is the amount of processing power and speed it can manage without burning out.

I think trinary processors are only worth it when using high end applications and/or server administration/applications.

By the time practical trinary computers become available (which won't happen, because nobody is investing in their developement), quantum computers will already be in a position to 'retire' binary, so it doesn't really matter.

[Arta]

No, you're missing the point too. Trinary is not better than binary. It is not preferable. It would not be an improvement. It is not a beneficial thing to develop or use.

[R.a.B.B.i.T]

That's why I said "practical"
I was stating that by the time effecient trinary proccessors would have become avaiable quantum computers would already be making a move into the market.

[Topaz]

And there ends the great discussion 

[Adron]

And there ends the great discussion 

By everyone agreeing trinary computing is no advantage over binary?

[R.a.B.B.i.T]

It's advantage is that when converted to an integer its value is greater than that of 'binary'.  That's about all, though.