The advance in fabrication technology hugely increases the number of available transistors

on a single chip. It allows the industry to build the entire system on a single chip which was

only realizable on a board in the past. On-chip System not only reduces the computer physical

size, but also increases the computation performance because modules/cores/intellectual properties (IPs) are packed closely together. When simply increasing the clock frequency to increase

the computer performance becomes harder because of the wire delay, putting more computation units on a single chip becomes a good alternative for improving computer performance.

Building more cores on a chip in the future is expected.

With many IPs on a chip, traditional bus is no longer able to provide enough bandwidth to

support the communication between IPs. Providing a high performance on-chip network infrastructure for the IP communication becomes a key to high performance on-chip computation.

This thesis focuses on an on-chip network supporting on-chip system.

This thesis is composed of two main parts. In the first part, a high performance deadlock

free dual-coded on-chip router using adaptive multicast routing is built. Compared with the

traditional deterministic XY unicast router, this router can reduce both packet latency and

energy consumption.

In the second part, a co-processor placement algorithm for an on-chip system built from

FPGAs with an on-chip network is proposed. The algorithm aims to place the communicating

modules as close as possible. In addition, an algorithm for sharing a FPGA by multiple co-processors and an algorithm for supporting polymorphic co-processor are proposed to increase

on-chip FPGA system throughput.