Re: Hardware Progess: $165/Gflop

From: Dani Eder (
Date: Tue Jan 06 2004 - 08:44:52 MST

> In fact (though I can't find it now) I've seen one
> chart for the next
> Semiconductor Industry Road Map that shows
> traditional increases slowing down.
The 2003 version of the technology roadmap for
semiconductors is at:

If you read through it, they posit a new chip
generation every three years, as has been the
trend. Chip generation is defined as feature size
scales by 0.7, or feature area scales by 0.5.

Then they go into detail describing what the
corresponding other changes in semiconductor
design and production methods will have to
achieve. They also show where the known methods
of improvement reach their limits.

As smart engineers work on the various problems,
the limits get pushed back. The question is
whether the limits get pushed back fast enough
to allow progress to continue at the present rate.
So far it has, and in fact the purpose of the
ITRS roadmap is to help focus R&D investments
so approaching roadblocks get addressed in a
timely fashion.

> per action potential). With those parameters the
> internal brain
> bandwidth is ~6000 TB/sec. You can juggle these
> numbers a bit but you
> aren't going to change them by many orders of
> magnitude.

If you are modelling the brain, you would only need
this much bandwidth if neurons were randomly
In reality you would assign each processing node a
brain slice of neurons that are locally connected,
and interprocessor communication then needs to only
carry traffic crossing the boundary of the slice.

> Its already to a large extent been worked out.
> For a general overview see:
> Kinematic Self-Replicating Machines by Freitas and
> Merkle

According to that website, that book is unreleased,
so it's kind of hard to review it.
> and for the first serious study see:
> Advanced Automation for Space Missions

I've got the paper version sitting on my desk
(NASA CP-2255). The paper I've been working
on is a followup to that study, which is 23 years

> The key point is [automating] *all* the pieces
>-- we are a long way away from that.
> Robert Bradbury

The point of tracking manufacturing productivity
data is to get an estimate of how many years
'long way away' actually is. Then you can
speculate on whether full automation or AI induced
singularity comes first.

If fully automated industrial systems, whether
single factory or over the whole industrial base,
can achieve doubling times of the order of 1 year,
then you can ask: how will the price of manufactured
products change, including computers? What will
be the scarce commodity in such an environment?


Do you Yahoo!?
Yahoo! Hotjobs: Enter the "Signing Bonus" Sweepstakes

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:43 MDT