| In very deep sub-micron design, the coupling capacitance of interconnects on a VLSI chip has shown dominant impact on the system energy and performance. The bus energy and delay become more data dependent because of the crosstalk effects among neighboring wires. Bus energy dissipation is determined by the overall crosstalk effects across the bus lines, while the bus delay is determined by the worst-case local crosstalk effect. The goal of this research is to design and implement novel bus encoding algorithms with low energy and hardware overheads for data buses to achieve lowest energy dissipation and transmission delay. The proposed bus encoding algorithms focus on reducing the bus energy by exploiting the probabilistic characteristics of data as well as minimizing crosstalk transitions. We first introduce the Weighted Code Mapping (WCM) technique that leads to one-to-one mapping from data to the code. Two algorithms are then developed improving the WCM scheme to optimize bus delay and energy dissipation simultaneously. One algorithm minimizes the crosstalk transitions by inserting temporal redundancy and achieves optimal energy, while the other aggressively achieves optimal bus delay by mapping the original data to low-energy opposite-transition-forbidden codes. An adaptive encoding method is used to further improve the energy and delay by adjusting the code mapping based on the detection of significant data statistical changes. The other major contribution of this research is the development of theories and techniques that the computation of the energy matrix can be significantly simplified by an analytical model with four component matrices. Based on this finding, we developed a Fast Transition Pattern Coding (FTPC) algorithm, whose major advantage is that in hardware implementation, the area overhead is a linear function of the bus width and the delay overhead remains constant. Experimental results show that FTPC-based algorithm works best for random distributed data; while WCM-based algorithms work best for data with high temporal and spatial correlations. They all outperform the existing bus encoding algorithms in energy, delay and implementation overhead. |
