Interconnect noise analysis and optimization for high-performance designs in nanometer technologies

Interconnect has become the dominating factor in determining system performance. In this context, we study layout analysis and optimization for interconnect. Crosstalk noise, as well as the impact of coupling on aggressor delay is analyzed. The pulse width of the crosstalk noise, which is of similar importance for circuit performance as the peak amplitude, is also measured. We consider parameters like spacing between wires, wire length, coupling length, load capacitance, rise time of the inputs, place of overlap (near driver or receiver side), frequency, shields, direction of the signals, and wire width for both the aggressors and the victim wires. Also, we consider parameters like driver strength as several recent studies considered the simultaneous device and interconnect sizing. The most important result of this analysis was determining the most important interconnect parameters that can affect the interconnect noise. These parameters were found to be the number of shields and spacing between wires.; We eliminated the inductive noise problem by power lines (shields) insertion. For multiple coupled nets, we formulated and efficiently solved the min-area shield insertion problem to satisfy given noise bounds. The noise model used can handle different wire widths, different spacing among wires, and different wire lengths that are coupled to more than one segment. The proposed algorithm is very fast as it consumes only 10% of the reported CPU time for a reported benchmark. After getting an inductive noise free layout, we minimized the maximum crosstalk capacitive noise among adjacent wires. We present a re-spacing framework to reduce this crosstalk noise. We explain the noise model used and how it is extended to handle multiple segments that have simultaneous multiple crosscoupling. The algorithm and the model used can handle the effect of re-spacing among signals in one layer and the other connected segments in another layer. The results show that we are able to decrease the maximum crosstalk noise by up to 20% for some benchmarks.