Half a croissant, on a plate, with a sign in front of it saying '50c'
h a l f b a k e r y

idea: add, search, annotate, link, view, overview, recent, by name, random

meta: news, help, about, links, report a problem

account: browse anonymously, or get an account and write.



Computer: Hybrid

Hybrid digital analog computer
(+4, -4)
  [vote for,

An advanced compiler would provide blueprint for the creation of an analog circuit of a given function. So for the function x^1/2, the analog curcuit would have an input of say 49v and an output of 7v.

From the blueprint, the circuit is fabricated and plugged into the motherboard of the classical machine.

Whenever this function is needed the classical program would obtain the result by a (single) hardware call, saving machine cycles to calculate this digitally.


1) A simple program.. private function squareroot(x as integer) squareroot = x ^ 1/2 end function

sub main dim a as integer = 49 a = squareroot(a) print a end sub

2) The blueprint for the function "squareroot" is sent to myFunctionFabriquePlantinHongkong.com on the internet.

3) myFunctionFabriquePlantinHongkong.com builds the board and sends it to the requestor. The board is not a digital sub processor but an analog circuit, perhaps with a DA a the front and AD at the back..

4) The requestor plugs in the board and runs the program.

5) The runtime environment would generate the interrupt required to obtain the output value from the hardware whenever the function was called.

peterpeter, May 31 2004

Josephson transistor http://ieeexplore.i...sp?arNumber=1288218
[kbecker, Oct 04 2004]

Field Programmable Analog Array http://en.wikipedia...mmable_analog_array
Brief description [BunsenHoneydew, Jul 09 2009]

What is a field-programmable analog array (FPAA)? http://www.ee.ualbe...audet/fpaa/faq.html
Longer description and FAQ [BunsenHoneydew, Jul 09 2009]

FPAA research http://opencircuitd...arch/fpaa/fpaa.html
[BunsenHoneydew, Jul 09 2009]


       Completely ridiculous. (+)
xrayTed, May 31 2004

       I don't know much about this shit, but I read an article in the Smithsonian about a guy who was working with hybrid digital/analog robots. He had little tiny insect robots whose actions were comepletely random--tough stuff for a robot. And he had a digital/analog with video cameras that was about as intelligent as your average labbador but creepier. Why not a computer?
Eugene, May 31 2004

       So far I like the idea. I have a stupid question - If a=16 does the compiler need to order a new circuit with input=16v and output=4v? Or is the circuit capable of computing any int as input?
1st2know, May 31 2004

       Next time my exit input call is to provide Y as input over a range selector carrying an N validator, just so I gain control over the escape key, I'll donate the board and there will be a ready-made one function board before anyone even asks for it.
dpsyplc, May 31 2004

       [1st2know] the idea is that any value input would have the function performed on it. So in the case of sqaureroot (above) any input voltage would be processed by the card to give an output voltage equal to the square root of the input voltage...ie. an infinate analog variety like a volume knob on a sterio.   

       Except that the above function specifies integers so the compiler would include components to force the step function of integer voltages only.
peterpeter, May 31 2004

       If you do the anlog/digital conversion fast enough this should work. Many DSPs these days have "programable comparators." They are a step in the same direction.   

       The comparator is a true analog circuit. It gets a reference voltage from a slow DAC in the DSP and compares it to the analog input voltage on a pin. The DSP only reads a single bit that shows if the input voltage is higher or lower than the DAC voltage.
kbecker, Jun 01 2004

       This idea does have some merit, but you would have to be talking about much more complicated computations than you are before it would start to pay dividends. Analog circuitry might be best used to analyse continuous signals.   

       For simple functions like square root, the D/A and A/D conversion times would probably mean that you would be quicker with a digital computation. There may also be accuracy problems in some cases, with possible errors occuring in the conversion or computation stages.
Thod, Jun 01 2004

       [thod] I was hopeing that it would be faster for large numbers, or that the dividends on the immediate calc would outweigh the overhead of the da/ad for very large numbers. If one could avoid the errors typical to analog systems and could compute large squares, it would be a step closer to finding factors of large numbers more quickly.
peterpeter, Jun 01 2004

       If you're talking fuzzy logic, where data is represented by non-integer values between zero and one up until final output, this might work. A full analog circuit path for each logic branch, with only the final result being subject to a conversion to binary might actually function rather well, especially if the analog circuit can be reconfigured on the fly. (+)
Freefall, Jun 01 2004

       //Now all you have to do is create a new fast acting transistor with a variable NPN or PNP junction.// There was actually some work done to build an ultrafast ADC using tuned Josephson junction transistors (sorry not published, I think the lab in the link did it). A single transistor would be a fixed threshold comparator. Using 1024 transistors (plus a magnitude comparator) would give you a 10 bit ADC. Estimated conversion time 0.3ns or 3Ghz frequency.
kbecker, Jun 01 2004

       //- you'd need to deal with a voltage difference of 1,000,000 between the input and the output//
Awesome! It's a computer AND an arc welder! [+]
1st2know, Jun 01 2004

       Yes for very large numbers I think there is a problem of scale, too many volts or too little precision. Perhaps the circuit could segmentize the output into manageble chunks and get reassembled by the AD. Not very pretty. To factor, you'd need the precision. Or perhaps the circuit could just get you in the ball park...and you do the rest in digital..
peterpeter, Jun 02 2004

       A Field Programmable Analog Array (FPAA) [link] could probably do this, while remaining physically attached to the computer. No need to fab up a new device for each equation - the array can be reconfigured in situ, as many times as required.
BunsenHoneydew, Jul 09 2009

       like ee doc smith, realized.   

       The lensperson series was about duplicating neural amplifiers. This idea goes so far beyond that.   

       I almost voted to promote this then I realized that if it ran an AI, no one, including the ai, could tell the ai how to reverse engineer its consciousness.   

       imagine a parallel universe where the creator paused prior to creating humans, then said Make them of that which makes them capable of knowing precisely what they are. rather than Hey look, it runs!   

       I am pretty ignorant though. I am certain the universe was created, that I am not clever enough to know how it actually works is different than sufficient cause to say, "hasty job dude, what were you thinking" although that is what I just typed.   

       we need a technology that goes beyond digital or analog which is pretty obvious when you consider the notion that digital is just a side effect of believing there are integers, wheras geometry creates autodescriptive spaces with meaningful points of function inflection outside the concept of analog.   

       think about a lissajous figure overlap of two waves can create what appears to a gradually rotating circloid yet is actually just two linear waves. The structure of perception, or unitarily worded the perceptron,makes us humans think there is a regular cyclic geometry rather than just a structured geometry. Thus if you look at the size, as well as interval of a sensor, it has intrinsic generation of shapes or geometries of recurrent recognition or "meaning", Taking that computation we create recurring waveforms that have a combination of similarity to a "clock" as well as a recurring absolute platonic form. so the new experiential information processing thing would be defined from the perceptron basic structure as much or more than the "facility of integers"(digital) while having more callable order than "everything all the time" (analogish) That is why gears or chairs are different than digital or analog.   

       Hey look it runs!
beanangel, Jun 20 2011

       It's claimed that functional programming languages are less natural than imperative ones, because the hardware is always really executing imperative code. This would put an end to that. [+]   

       //the problem of scale// Sliderules were analog computers, and encountered the same problem. They solved it with logarithms. (Well, sort of; there was indeed a trade-off between scale and precision. But the same is true of floating point arithmetic.)
mouseposture, Jun 21 2011

       [+] Of course there's been more digital development than analog the last few decades; you might as well just run it totally digitally... the original idea that is, not whatever [beanie] and [mouse] are going on about.   

       Re: the idea, step 3, there would be A/D's and D/A's on the board, but there'd be quite a few of them: it's easier, cheaper and more accurate to run as much in the digital domain as possible. So as well as the converters necessary in every chip to translate from the mainboard's "language" and voltage/timing to the chip's, the chip would be "supervising" the real analog processes digitally.   

       Neat idea.
FlyingToaster, Jun 21 2011


back: main index

business  computer  culture  fashion  food  halfbakery  home  other  product  public  science  sport  vehicle