Worst Case Crosstalk Noise for Nonswitching Victims in High-Speed Buses Jun Chen and Lei He

Outline Introduction Preliminaries WCN without timing constraint WCN problem with timing window constraints Conclusion

Outline Introduction Preliminaries WCN without timing constraint WCN problem with timing window constraints Conclusion

Introduction The coupling-induced crosstalk noise gains growing importance in deep-submicrometer circuits and circuits with higher clock frequency Crosstalk noise may cause variation of delay and logic failure of a victim net In multigigahertz designs, the inductive crosstalk noise can no longer be ignored

Introduction (cont.) Worst case delay (WCD): the maximum possible delay caused by crosstalk noise Worst case noise (WSN): the maximum possible crosstalk noise For the RC model, the WCN problem is formulated as finding the alignment of switching times for multiple aggressors such that WCN is reached

Introduction (cont.) With the consideration of inductance, the WCN problem becomes much more complicated: Switching patterns, alignment, and switching direction Coupling between both adjacent and nonadjacent interconnects Routing direction

Outline Introduction Preliminaries WCN without timing constraint WCN problem with timing window constraints Conclusion

Preliminaries victim types: quiet, noisy, or switching Only non-switching victims are considered Arbitrary switching patterns for aggressors Assume all drivers(receivers) have a uniform size and are cascaded inverters Assume all wires have uniform width and spacing

Preliminaries (cont.) Only coupling capacitance between adjacent wires are considered Full partial element equivalent circuit(PEEC): Self-inductance for each wire segment Mutual inductance between any pair of wire segments

Outline Introduction Preliminaries WCN without timing constraint WCN problem with timing window constraints Conclusion

WCN under RC model No resonance in the noise waveform The problem can be reduced as the alignment of aggressor switching times Simultaneous switching(SS) Superposition(SP) Aligned switching(AS)

Impact of shielding for RLC model Assume there are shield at both edges of the bus structure

Impact of switching pattern for RLC model The waveform may have resonance due to inductance under the RLC model

Impact on routing direction for RLC model Signals are routed either from left(top) to right(down) or from right(down) to left(top)

Algorithms for the quiet victim SS: All aggressors switch simultaneously in the same direction. SP: Find the maximum noise peak for each aggressor when only this aggressor switches AS: Obtain the individual noise waveform by simulating the interconnect structure with only one aggressor switching for each time PP alignment NN alignment PN alignment Simulated annealing(SA) and genetic algorithm(GA)

New algorithm SS + AS: WCN is approximated by the larger one between the results obtained by SS and AS

Time complexity

Experiments with the quiet victim Aligned bus Different routing direction Unaligned bus

Aligned bus

Different routing direction

Unaligned bus

Noisy victim SS:

Noisy victim (cont.) SP: Sum of all peak noise values and the peak value of the propagated noise AS: Treat the propagated noise as an individual noise waveform of an extra aggressor

Noisy victim (cont.)

Outline Introduction Preliminaries WCN without timing constraint WCN problem with timing window constraints Conclusion

WCN problem with timing window constraints Each aggressor has the switching timing window The victim has the sampling timing window at the far end Both timing windows should be considered in the algorithm

AS Algorithm For PN alignment:

Expansion of noise waveform

SS Algorithm Determine all the overlapped regions for the timing windows of all the aggressors For each such regions, find the simultaneous switching noise within the sampling window The largest noise among the simultaneous switching noise of all the overlapped regions is the WCN

Experiments The timing windows and routing directions are randomly generated Aligned bus structure The driver size is 100X The victim is quiet

Experimental results

Comparison of WCN under RLC and RC Aggressor switching windows size = 20(ps) Victim sampling windows size = 10(ps)

Comparison of WCN under RLC and RC Aggressor switching windows size = 30(ps) Victim sampling windows size = 15(ps)

Comparison of WCN under RLC and RC Aggressor switching windows size = 50(ps) Victim sampling windows size = 25(ps)

Outline Introduction Preliminaries WCN without timing constraint WCN problem with timing window constraints Conclusion

Conclusion Both switching time and switching logic pattern of aggressors affect the WCN under the RLC model and the routing direction also impacts WCN significantly under RLC model A new SS + AS with linear time complexity has proposed, which has an average underestimation of 3%