Abstract:
As semiconductor technology advances, a highly complex system can be put into an integrated
circuit (IC) called System-on-Chip (SoC). Nowadays, a single integrated circuit can contain many
heterogeneous processors to form a SoC called multi-processor system-on-chip (MPSoC) in which
each processor connects to the others through multi-interconnection called a Network-on-Chip
(NoC). The Network-on-Chip design is an important part of MPSoC efficiency because
interconnects become dominant part of the chip. There arc 2 types of interconnection: serial and
parallel. Although serial interconnect is preferable because it needs only a single data line, it cannot
be used to support cases with high data transfer rates. However, parallel interconnections require
high number of lines which increase power consumption and complexity. This thesis proposes to
use data compression before sending data through the NoC so that the number of lines is reduced,
which leads to reduction of power consumption. The application methodology, case studies and
model of the point-to-point parallel transmission system are proposed using a data compression
called the Unused Significant Bit Removal (USBR). The verified model was implemented on a
Field Programmable Gate Array (FPGA) chip. The verification results show that data compression
application possibility and efficiency on Network-on-Chips parallel transmission depend on power
consumption and compression ratio of the compression algorithm.