Two promising new technologies for parallel processing that will have a far-reaching impact on parallel professing are:
- Neural Networks and
- Optical Computing
Technologies for parallel processing
Neural Networks consist of many artificial neurons, or processors, that have densely parallel interconnections. The processors communicate across the connections in terms of “activation” and “inhibition” signals that excite or inhibit responses by connected processors.
Neural networks differ from one another in the way the nodes are interconnected and in the rules they use to produce a given output signal of different sets of input signals.
Some networks are fully interconnected gride, with each processor directly connected to every other processor; others embody tree-like or layered architectures.
Neural networks represent a new approach to computing and artificial intelligence that shows great promise for solving problems that are clearly not suited for conventional computers and currently established techniques.
The main features of optical computers are parallelism and global connections. Optically interconnecting a large number of computing elements in a structure creates any independent channels that can be exploited to increase the computational power by decreasing the communications overhead.
Optical signals can propagate through separate channels with essentially no interaction and through parallel channels without interference.
A neural network computer built from optical elements consists of two main components. One is a two-dimensional array of optical switching elements of the array that can be interconnected to all other elements by light beams.
The second component consists of interconnections among the elements. Such interconnections consist of interconnections among the elements. Such interconnections constitute the memory of the neural computer and must be modifiable in a general propose optical neural computer.
A hologram whose volume in one cubic centimeter can specify more than a trillion connections. This is enough for all possible interconnections of one million processing elements.
It is not surprising that optical neural computers combining the technologies of neural networks and optical computing form an exciting new area of parallel processing.
You may like also