🚀 go-pugleaf

This is a posting from the sci.nanotech newsgroup.  This closely
parallels a rumor that was reported in Robert X. Cringely's
"Notes From The Field" column in _Infoworld_ a month or two back
(which I posted to the *xtrop* lists).  Real-time RSA video
encryption, anyone?  (Too bad about the funding source, though!)
					dk

------- Forwarded Message

Received: from planchet.rutgers.edu by rand.org; Wed, 17 Jun 92 19:42:51 -0700
Received: by planchet.rutgers.edu (5.59/SMI4.0/RU1.4/3.08)
	id AA11668; Wed, 17 Jun 92 22:39:27 EDT
Date: Wed, 17 Jun 92 22:39:27 EDT
Message-Id: <9206180239.AA11668@planchet.rutgers.edu>
To: nanotech@cs.rutgers.edu
Path: nanotech
Newsgroups: sci.nanotech
From: mgj@cup.portal.com
Subject: High tech info exchange
Approved: nanotech@aramis.rutgers.edu

12345678901234567890123456789012345678901234567890123456789012345678901234567

                                   OFFER:

           TECHNOLOGY ON *TERA-OP* PROCESSING SPEED ON A DESKTOP

              in exchange for your assistance in determining a
          sufficiently challenging application for this technology.



A new DARPA sponsored technology has created unprecedented opportunities in
2-D signal processing. It is now feasible to produce unheard of speeds in
compact, rad-hardened, low power systems. The invention is based on a unique
integrated circuit producing a total of 92 Billion multiply and add
operations per second!  With this technology correlations, convolutions,
complex coefficient filtering and non-linear operations (including neural
networks) are performed with blinding speed.

My collegue and I have recently completed a SBIR Phase 1 project that
successfully demonstrated how this technology could be used to process 2-D
signal data at Tera-Op speed. For example, suppose each frame of input data
consists of a 1000 x 1000 pixel image and you wish to detect small image
patterns that could be anywhere on each frame.  You decide to use template
matching that requires correlating the frame with many different templates,
each 100 x 100 pixels in size.  A single correlation of one template against
the frame would require about 10 billion multiply-accumulate operations. A
set of 100 templates would require 1 trillion operations. This would take
about 16 minutes on a typical Cray supercomputer; however, 22 chips utilizing
this special technology would perform the same process in only 1 second,
almost 1000 times faster! A black box containing 22 chips would be about
the size of a desktop laser printer, cost about $25,000, and have a price-
performance ratio of only 2.5 cents per Mega-Op!

So, what s the point of this announcement?

We have been asked by the U.S. Government to consider designing and building
any practical black box that uses this technology on a *real world* 2-D
signal processing task.  Consequently, we are looking for a business,
university or research center that needs extremely fast 2-D processing for a
real application.  You show us the processing algorithms you would like
applied, and if it is suited to this technology, we will create a working
system, completely funded by the U.S. Government.  This involves no cost or
risk on your part.  When the project is completed, we will show you what we
did.  (Your firm saves about $300,000 in R&D costs!)

The catch?  To satisfy the U.S. Government, our project needs a convincing
real world application, one that could be extended now or later for military
use (eg, image pattern recognition, image stabilization, speech recognition).

WHAT COULD YOU DO WITH TERA-OP PROCESSING SPEED ?

Convolve 2-D video frames 1000 X faster than real time.
Match over 1,000,000 2-D templates (128x128)  in less than 1 sec.
Correlate data 1000 X faster than a Cray supercomputer.
Create tera-op multi-layer neural networks.
Eclipse your competition in I.R., radar, lidar & sonar signature analysis!

This technology offers other benefits too: gallium arsenide radiation
hardness, compactness, low power needs, and flexible modularity.

After we discuss with you the modular nature of this technology, you submit
to us a white paper on how it would solve one of your outstanding technical
challenges.  If we select your idea, you re in.

For more information, contact...

Mark Jurik
PO 2379
Aptos, CA  95001     (408-688-5893)

[I'm posting this on the chance that there's anyone with a molecular
 modelling or measurement task that's interested...
 --JoSH]

------- End of Forwarded Message