Evaluating Techniques for Wireless Interconnected 3D Processor Arrays

In this thesis the viability of a wireless interconnect network for a highly parallel computer is investigated. The main theme of this thesis is to project the performance of a wireless network used to connect the processors in a parallel machine of such design. This thesis is going to investigate new design opportunities a wireless interconnect network can offer for parallel computing.
A simulation environment is designed and implemented to carry out the tests. The results have shown that if the available radio spectrum is shared effectively between building blocks of the parallel machine, there are substantial chances to achieve high processor utilisation. The results show that some factors play a major role in the performance of such a machine. The size of the machine, the size of the problem and the communication and computation capabilities of each element of the machine are among those factors. The results show these factors set a limit on the number of nodes engaged in some classes of tasks. They have shown promising potential for further expansion and evolution of our idea to new architectural opportunities, which is discussed by the end of this thesis.
To build a real machine of this type the architects would need to solve a number of challenging problems including heat dissipation, delivering electric power and Chip/board design; however, these issues are not part of this thesis and will be tackled in future.