> On 9/30/2011 8:47 AM, Stephen wrote:
>>> Regardless, have you ever actually seen an analogue computer with
>>> independently addressable memory cells? I haven't heard of such a thing.
>>>
>> What to answer first? I don't know, the only actual working valve
>> computer I've ever been in the presence of was at Glasgow University,
>> over 40 years ago. I've never had hands on experience working with them.
>> But you would not use addressable memory as you would in digital
>> computers. Remember that we are talking about voltage levels
>> representing numbers. Also back in the day, what was being asked of then
>> was much simpler and less complex so a lot of memory would not have been
>> required.
>>
> Just use a large, covered, lake as the "storage". You get a certain
> level of error, and possible leaks, but you could compute a number as
> large as... what ever the lake could contain in what ever unit of water
> your system dealt with. lol Seriously though, I would say that the
> distinction between digital and analog in this sense is a) how long you
> can hold the value, and that assumption that the value you can hold has
> a set number of limits, in DNA, assuming you where using it to encode
> numbers, that is like what 4 values (ATGC), in some indefinite set of
> combinations, 3 states, for like the "quantum" systems they are trying
> to develop, or 2, on/off, for current systems. In all of these cases,
> the "values" are discrete, hence digital. You can't, in the case of
> something like DNA "get" any other values (unless you just completely
> change the proteins involved), I am not sure what you get with quantum
> effects, but generally they seem to be discrete states, but with binary
> we *intentionally* ignore any difference of state, other than below X
> level, or above it. So, if your "analog" circuit had these values 0.1,
> 0.3, 0.6, 0.12, and 0.7, "binary" simply enforces the rule that this is
> actually 00101.
>
> In principle, your "analog" computer just throws out that assumption.
> The result is less reliable, which is the main reason we stopped trying
> to use it. But, in theory, if someone had wanted to, they maybe could
> have made a base 10 computer, by treating each "range", 0-0.1, 0.11-0.2,
> etc. as a different "state". It probably wouldn't have been at all
> feasible, in that the fail rate on circuits that didn't produce the
> correct result, or match specifications, would have been much, much,
> larger. Like, if now we threw out one in every 1,000 processors, you
> might see a fail rate of like 1:50 for such a base 10 system, or worse.
> Because, of you are dealing with on and off, 0.6-1.0 is "acceptable" for
> the "on" state, and 0-0.4 might be for the "off", with only the absolute
> middle range of values being too ambiguous.
>

There is another consideration: Power consumption and heat dicipation.

With a binary gate, you consume almost no power on a zero or a one. You 
do consume a lot more when you switch value and are at an intermediate 
state.

With a decimal processor using 10 ranges, 8 times out of 10, you are at 
a steady intermediate state that forces you to dicipate much more heat.

Things get even worst if you want to create an analog logic switch.

Post a reply to this message