[extropy-chat] self replicating machine ....

[Dan Clemmensen]

Adrian Tymes wrote:

>--- Jan Gruber <bchjg at nus.edu.sg> wrote:
>  
>
>>I thought this bit of BBC tech news might be of interest - not sure
>>how the electronic components are supposed to be replicated though. I
>>seem to remember that this was discussed here some time ago - anybody
>>know what is the state of the art with this ?
>>    
>>
>
>The electronic components - specifically, the microprocessors - are
>indeed the stumbling points.  Simple mechanics can self-replicate, if
>guided appropriately, but so far no one has been able to make the
>information carrying bits - again, the microprocessors -
>self-replicate. 
>
Adrian pointed out that the real issue is the definition of "raw material."

It is now theoretically possible to build a complex electronic circuit using
using an ink jet printer. I do not know if you can build the circuitry 
needed by an
ink jet printer using an ink jet printer, but my gut feeling is that you 
can.

Today even the most trivial process control microprocessor is massively more
capable than it needs to be for most simple processes. We need to perform
a gedankenexperiment: What is the minimal circuit that can control an 
ink jet printer,
and what is the biggest circuit that an ink jet printer can print? If 
these numbers are
reasonably close, it should be possible for an ink jet printer to 
replicate its circuitry, and
to replicate the circuitry for the rest of a self-replicating system 
that includes an
ink jet printer.

For this exersize, we must consider the paper (or other substrate) and 
the rather exotic inks
as raw materials. At the system level, a small production plant for the 
paper and a small production
plant for each of the inks is likely to be a whole lot easier to build 
than even the smallest conceivable
infrastructure based on silicon wafers.

The only problem here is the ink jet print head. The use of an ink jet 
to print a dense circuit depends
critically on the precision of the print head, and today's print heads 
are lithographic. An ink jet printer
has a resolution of 600dpi (about 100micrometer) or better. If we relax 
the resolution, we lose circuit
density. My gut feeling is that a printer using non-lithographic, purely 
mechanical techniques, can
achieve a precision of at best 100 micrometer. This is an areal density 
of one percent of the density
achieved by an ink jet printer.

But let;'s go for it.
Assume a mechanical plotter with 100micrometer resolution.
Assume a substrate (page)  with a 32cm.32cm printable size
Assume a 1-bit-per pixel ROM technology: we can store and replicate 
10Mb/page.
If a transistor plus routing overhead cost 10 pixels, we have 1 million 
transistors per page.

Note that this technology is incredibly SLOW by comparison with modern 
semiconductors.
who cares? It is fast enough to control a self-replicator. Recall that 
computers in 1980
had 1Mhz clocks. A system with a 100micrometer resolution should be able 
to operate
with a 1Mhz clock.

Let's arbitrarily assume that the self-replicator can be described in 
100MB=800Mb. This is
80 pages of plotting. Further assume that the electronics can be 
instantiated in an additional
80 pages of electronic circuitry. This is roughly equivalent to two 
reams of office-quality paper,
or about 1Kg to store the information and electronics.

I think I need to write a paper on this subject instead of an e-mail. 
Can someone point me to
references?