Pipelined Bidirectional Bus Architecture for Embedded

This paper proposes novel high-performance bus architecture for memory-intensive embedded multimedia SoCs. It has a pipelined bidirectional bus for high speed and small area. It has two separate bus called system bus and memory bus, where memory-intensive IPs are connected to memory bus so not to degrade system bus performance. To avoid starvation of low-priority masters, the proposed bus exploits probability-based arbitration policy where the arbitration probability of each master is determined in proportion to its execution time. To increase transmission bandwidth, it also exploits bus partitioning where several masters often access their slaves concurrently without multilayer structure. The proposed bus is designed, implemented, verified, and evaluated in hardware level. Simulation results show that the proposed bus improves effective bandwidth by 2.8~3.6 times and communication latency by 3.1~4.7 times when compared to AMBA bus.