[ExI] Gordon Moore Quote, Extropy and Panspermia

[Eugen Leitl]

On Thu, Mar 15, 2012 at 10:31:36PM -0600, Kelly Anderson wrote:

> While referring to the increase in the exponent itself .... he said,
> "What amazes me more than anything is this acceleration... I think we
> have our own form of dark energy. I don't understand it in cosmology
> and it's just about as hard to understand as it fits in here..."
>   - Gordon Moore

He must be getting mystical in his old age. There's nothing particularly
difficult about scaling laws in 2D, driven by a lot of money --
consider the costs of plants for each new node. And Moore doesn't
say anything about performance, only the number of affordable
transistors on a piece of a particular semiconductor.
 
> I am coming more and more to the opinion that when Kurzweil talks
> about the Law of Accelerating Returns, he is tapping into something

The Kurzweil's (Morave's actually) Law of Accelerated returns isn't much of a law.
Anyone who understands the basics of benchmarking would be very, very
careful to put a single scalar even on conventional systems. 
Speaking of meaningless benchmarks (which don't lie; liars do
benchmarks) how do you even run Linpack on a Jacquard loom?
Are you streaming, or is this random-access? Why?

If you're willing to cherry-pick your measurements, you can fit
any curve. 

Consider the following curves http://www.gotw.ca/publications/concurrency-ddj.htm
(the essay is from 2004, data updated in 2009, and he
doesn't mention the real problem: needing not just concurrency,
but shared-nothing asynchronous concurrency)

And what does the law forecast, actually? That the linear
semilog will go on forever? Exponential processes run very soon
into the limitations of this universe, and stop being that.
Buh-bye.

It's pretty obvious that there will be a bump in the Moravec's (because
it should be called Moravec's Law of Acclerated Returns, as he came up
with it first) performance curve around 2020. Just as the clocks stopped doubling
in 2003, quite suddenly, we'll be suddenly out of shrinkage space
and TSV stacking is off-Moore. Does this mean we can kill this
kooky pseudolaw, or will there more elaborate benchmark contortions
to plaster over the the hockey stick part curve, until we get real 3D integration going,
which should be good for another four orders of magnitude, or so?

> bigger than even he is letting on. I think that the emergence of
> complexity may be a law of thermodynamics. I think it may be a

If you study the history of the Solar System, it's more like
the law of being really, really lucky. Consider all the other
systems which all don't look at all lucky, because the losers
of this particular Russian roulette can't.

> fundamental aspect of physics... extropy may be more than a mailing
> list, it may be a law of the universe. If it is in fact a law of the
> universe, then that gives a logical mathematical basis for believing
> in galactic panspermia... 100,000,000 years is a very short time to

There is even no plausible mechanism for interstellar panspermia.
The only way I could *imagine* it happening is by crosscontamination
through Oort clouds during close flybys, which implies every icy
body out there is lousy with life which doesn't strike me as very
probable. There's not that many liquid pockets around, and they don't
stay that for very long.

> come up with single cells at the rate things were changing 4 billion
> to 3.9 billion years ago... Just some food for thought. This isn't
> Moore's idea, it's mine... so don't saddle him with my crazy idea.
> 
> Other highlights of the talk were that they are doing
> photo-lithography (this might be a few year old talk) on the scale of
> a third of a wavelength of the light they were using to do the
> lithography. That's just too cool. Also, that a constant for Intel

I don't think this is cool at all. Intel is already at 11 nm in
prototyping stage and there's really no easy way to tell the roadmap
to 6 nm and 4 nm is viable. We will only know for sure when they're
shipping. The buck definitely stops at 1 nm.

> over the years is that chips have always sold on the order of a
> billion dollars per acre. That's a fun way to think of it. He was also

This is another issue, as growing Si monocrystals isn't infitely
scalable either, and 450 mm wafers might be the last size we'll
be getting. So if you run out of real estate and shrink potential,
you're screwed. Photolitho is fundamentally a subtractive manufacturing
technology, and attempt to add layers will degrade underlying layers.

So you need a completely new manufacturing technology, which deposits
layers on top of layers in mild conditions, or actually goes real
3D with autoassembly from solution.

If such a technology were to smoothly take over in 2020 or thereabouts,
we would have it already working in the lab a decade ago. 

> talking about how they were getting quantum leaking through the
> substrate because it was so very thin. Two dimensional Moore's Law

Leakage is one thing, quantum effects another. Recent nanowire
work shows we can stay classical almost all the way down to 1 nm.

> will run out of steam around 2020, apparently, due to physical
> limitations related to the size of atoms.. but that there is always
> the third dimension to play with. Chips will undoubtedly have to start

Have you tried playing with the 3rd dimension? It's not easy.

> getting thicker and more complex in a third dimension if we are to
> continue the curve. Even new flat technologies based on something
> besides silicon dioxide won't help because atoms are atoms... the

We haven't been using SiO2 for a while, see high-k dielectric.

> third dimension is the only way out unless you can figure out how to
> use subatomic particles. 3D silicon seems easier than harnessing
> subatomic particles to me... but what do I know?

The only way you're going to touch the nucleus is via the spin.
I suggest that electron spintronics is much, much easier. See nitrogen
vacancy work in diamond, which is actually compatible with qubits
in solid state.