FSE

Furry search engine

formerly SoFurry search engine

Plush Companions, Inc. -- Positronic Neural Nets

Welcome to Plush Companions, Inc.

Suppliers of the finest artificial animals in the Solar System! _________________________________________________________________

Positronic Neural Nets(TM) White Paper

Company Confidential

This information is not to be distributed without the express permission of PCInc Engineering. All Rights Reserved. _________________________________________________________________

Permission granted for public distribution. [signed] Sam-3G395 Director of Engineering 2034/14/32 _________________________________________________________________

A recent development in PCInc's engineering lab has become indispensable in the construction of all Plush Companions.

An exciting application of quantum dot technology in PCInc's engineering laboratory has made possible highly sophisticated positronic neural nets. Quantum interference patterns have been found to provide ideal isolation and control, avoiding the problems of radiation and noise seen in previous positronic designs.

The use of point-like antiparticles (positrons) in multilayer submicron optical laser structures has yielded unequaled computational power and precision.

Constructed from polysilicon protein lattices, these positronic structures have been used to construct gigachannel optical ALUs, making possible a dramatic increase in computational bandwidth. Performance measurements of production designs have revealed improvements far beyond those originally anticipated.

See _ PNN Performance Profiles _ for a detailed report of PNN performance measurements and comparisons with competing designs.

The advantages of this new positronic technology in support of highly sophisticated neural nets cannot be overemphasized. Varying in size from 1cc to over a meter in diameter, PCInc's Positronic Neural Nets (PNNs) have been found capable of hosting mechanical intelligences of unsurpassed stability and reliability.

Structural support

Another important component of this project has been the development of high strength ceramic shells, transparent to a wide range of optical frequencies. These crystal-clear structures provide the physical support and unequaled optical clarity essential for high bandwidth I/O. When formed within these containers, PNNs have been found capable of withstanding extremely high accelerations and mechanical impulses while sustaining undisturbed computational effort.

See _ Molded Optical Containers _ for a detailed report of PCInc's KrystalKlear(tm) shell development.

Production

PCInc's proprietary vacuum deposition assembly lines are able to produce PNNs at a rate which can satisfy the most demanding markets. Current production rates of standardized models total more than 1000 per day. Manufacturing capacity can easily be expanded to several times that rate, even when a variety of nonstandard competencies are being formed. Please contact PCInc Sales to find out how we can meet your computational needs.

Surprisingly, the stream of colorfully glowing spheres leaving the final test station has proved to be an unexpected tourist attraction. Please contact PCInc Public Relations to arrange a visit to our facilities.

Performance limitations

While PNN performance has been found to increase more than linearly with the diameter of positronic optical assemblies, it isn't as directly related to their volume as one might expect. There seem to be two interacting effects. The number of computational elements is proportional to a PNN's volume, and thus so is the number of optical interconnections available to each individual element. With larger sizes, this increased interconnectivity dramatically increases computational performance.

In contrast, the number of external signaling channels is only proportional to a PNN's surface area. For larger volumes, this means that the amount of external information available to a particular computational element is reduced.

Another performance limitation is caused by the distance between computational elements. The speed-of-light delay plus accumulated opacity means that larger PNNs tend to have a longer resolution time due to slow feedback among CEs distant from one another.

The relative lack of external stimuli, combined with slower resolution, means that larger PNNs tend to be more introspective. Indeed, PCInc has found that assemblies more than three meters in diameter (about 10 cubic meters) usually become catatonic and unresponsive to external events less than a week after activation. While the exact relationship between volume, surface area, and catalepsy is still being characterized, PCInc has determined that spherical PNNs smaller than about 1 meter in diameter show no signs of this problem.

See _ The DayDreamers _ for the journals of several large PNNs. Obtained shortly before they ceased communicating, these documents provide unusual insights into the workings of such vast intelligences.

One of the ways to avoid this problem is through the association of smaller systems into tightly coupled gestalts. Careful selection of compatible PNNs has resulted in only minor performance penalties when compared to a single PNN of similar volume.

Robotic Rights

PNN assemblies larger than about 300cc easily pass all of the standardized Turing tests. As a result, the Bureau of Robotic Rights must be consulted whenever one of these units is to be used. Their requirements are that the PNN acquiesce and that the purchaser contract to avoid intentional damage to the assembly.

Normally permission is easily obtained. All of PCInc's PNNs have an inherent desire to become associated with humans, while PCInc's standard purchase order form includes appropriate contractual requirements.

Caveat

Due to the Asimovian Laws integral to the design of all PCInc Positronic Neural Networks, they should not be used in combat situations. While any construct incorporating a PNN will excel in providing protection against attack, it will resist any attempt to be used aggressively against humans.