The ever-increasing amounts of data and video traffic are placing a tremendous strain on the world's information and storage technologies and communications infrastructures (wireless and wired).
Generally speaking, computational tasks can be split into two categories -- control/decision-making tasks and algorithmic/data-processing tasks. Also, generally speaking, there are two main ways to perform computations in silicon. One is to use one or more von-Neumann-type processors, which are inherently serial in nature; the other is to operate in a massively parallel fashion.
General-purpose processors are ideally suited to control/decision-making tasks, but they are horrendously inefficient with regard algorithmic/data-processing tasks, which are best performed using ASICs or FPGAs.
ASICs boast the highest performance, the lowest power consumption, and the lowest unit cost (assuming extremely high production runs). In return, the construction of these devices requires all the layers to be custom, they have high NRE/tool costs, they have long development times (typically around 18 months), and they pose the highest risks.
Click here to read more ...