Try running the simulation for many cycles to see if they two single-ended waveforms of the differential signal get more balanced with time. Use the recommended resistor value for this experiment.
Thank you for your suggest, I simulated for several cycles and the behavior is the same along.
My questions are so :
I disconnected one memory to avoid reflexions problems and simulate with only one memory.
For that I put the resistor near the memory and run simulations for 100ohm and 500ohm.
The picture below shows the results and the crossover point is valid for the 500ohm case.
RT = 200ohm , crossover Ko >1.125V f = 333MHz
RT = 500ohm , crossover ok =1V f = 333MHz
Is these some bad configuration ?