On 12/2/2010 3:37 AM, Invisible wrote:

> "[T]here is a limitation to how small, fast and compact silicon computer
> chips can be. DNA computers show promise because they do not have the
> limitations of silicon-based chips."
>
> O RLY?

Of course there's a limit. And there's likely a limit on how small/fast 
a DNA-based machine could be as well. The higher the clock frequency, 
the more susceptible it is to external influences such as EMI. The 
faster the clock, the more limitations on the actual circuit design 
there are, but you knew that.

> "For one, DNA based chip manufacturers will always have an ample supply
> of raw materials as DNA exists in all living things; this means
> generally lower overhead costs."
>
> And you understand that silicon chips are MADE OF SAND, right? You know,
> as in "worthless as sand"? Given planet Earth's gross elemental
> composition (60.2% silica, 15.2% alumina, >5% everything else), I
> suspect that silicon is rather more abundant than DNA. And let us not
> even get into the fact that DNA for computers would be utterly different
> in sequence to DNA from living organisms.

The advantages of living on a planet with a primarily silicate crust: 
Lots and lots of silicon to go around. Not sure what the ratio is of 
silicon to carbon, but I imagine its not hard to find a rock containing 
some sort of silicate.

>
> "Secondly, the DNA chip manufacture does not produce toxic by-products."
>
> Riiight. So because the end product is DNA, a molecule that already
> exists in nature, therefore you can produce it with no toxic by-products?
>
> And the DNA itself wouldn't be toxic, no?

Not unless it codes for something that could cause illness or kill you, 
but who's paying attention anyway. Organic compounds are some of the 
most toxic to us, because they are the most likely to interact. To be 
sure, there are a lot of extremely toxic inorganics as well.

> "Last but not the least, DNA computers will be much smaller than
> silicon-based computers as one pound of DNA chips can hold all the
> information stored in all the computers in the world."

Storage is not the same as computation. No mention as to how fragile 
that pound of DNA is. UV light or any form of ionizing radiation? 
consider your data hopelessly corrupt. so much for DNA computers in the 
space program. Other chemical and even physical processes could degrade 
it. If it happens to get contaminated by bacteria, they'd certainly 
enjoy the amino-acids that used to compose your photos from your latest 
family vacation to Hawaii. Not stable in the least!

> Current computers are much, much larger than strictly necessary mainly
> due to issues of heat dissipation. You can already make RAM chips that
> hold absurd quantities of information; it's just that they tend to melt
> when you switch them on.

I believe the bigger limitation on the amount of information that RAM 
can ultimately hold is more about the lower limit on the size of a 
transistor, rather than heat. What generates heat is the act of 
switching. If a latch is held, then there isn't much switching going on, 
though I'm sure there is some power consumption leading to some heat. 
Processors get burning hot because they constantly switch. RAM gets hot 
because of the way its state is maintained and how its accessed, at very 
rapid rates nowadays. Flash memory is obscenely high density, yet 
generates very little heat, unless it's accessed at its full bandwidth 
for an extended period of time, and even then its not nearly as much as 
a processor. Its also a few orders of magnitude slower than RAM.

>
> Besides, just because a strand of DNA can /store/ a lot of information,
> it does not necessarily follow that you can build a working
> /computational device/ which is only slightly larger.
>
> "a DNA computer the size of a teardrop will be more powerful than
> today's most powerful supercomputer."
>
> Possibly. But if you want it to do something /useful/, the teardrop by
> itself won't be much help. You still need I/O devices, for example.
>
> "The capacity to perform parallel calculations, much more trillions of
> parallel calculations, is something silicon-based computers are not able
> to do."
>
> I beg to differ. It would be more accurate to say that nobody has come
> up with a way of structuring computer programs as trillions of
> independent steps. We could totally build really parallel silicon chips.
> For example, recent GPU designs involve executing several hundred
> computations in parallel. There's no particular reason why you can't
> scale that up to thousands or millions - it's just that the extra R&D
> work probably wouldn't pay off in extra sales, because the software to
> utilise that much parallelism is lacking.

You'd need the pipelines to do it in the chip die. You'd need to build 
very small computational units to get that massively parallel.

I don't get how DNA can compute anything. DNA is essentially a coding 
for proteins. What would your end result of a computation be? A glob of 
proteins that mean some sort of result?

> "In the current technology of logic gates, binary codes from the silicon
> transistors are converted into instructions that can be carried out by
> the computer."
>
> This is a highly questionable and very muddled statement. It's hard to
> read something like this and continue to believe that the writer has any
> clue what they're talking about.

The statement doesn't seem to make a lot of sense. The instructions, 
encoded in binary do tell the computer what to do, yes? But it's 
essentially flipping a bank of switches to route the data from the input 
of an ALU to one unit or another....

> "though it may be very fast in providing possible answers, narrowing
> these answers down still takes days."
>
> This rather suggests that the operation of a DNA computer is
> non-deterministic (and hence, applicable to a much smaller set of
> problems than a Turing-complete machine).

Huh? Computers can very quickly give exact answers to a wide class of 
problems. Modeling complex systems, however, is a different story. I 
don't see DNA helping with that. Often performing calculations much 
quicker than our organic brains can. That cluster of neurons may be 
massively parallel, but it's still not very fast at doing things like, 
say, computing a Fourier transform. In fact, before the silicon-based 
machines of today, the Fourier transform was regarded as largely useless 
because it was so difficult to compute. And still, until FFT was 
discovered it was regarded as simply a novelty.

> I won't hold my breath for this happening any time soon. :-P

If you do that, you may cause a catastrophic failure of your DNA-based 
system.

I haven't read the article, just adding my commentary to your 
selections, but based on what I've seen this sounds like massive 
speculation and nothing more. What is needed, really, is more 
improvements on existing algorithms to allow them to operate in parallel 
on simpler computational units. Once a high degree of parallelism is 
met, then we'll see some huge jumps in how fast these silicon machines 
can really work.

-- 
~Mike

Post a reply to this message