Hello, guys:
I would like to say thanks all of you, specially to Ivan_ and toonafishy. I have solved the problem with the help of Ivan_. And I think toonafishy's suggestions are also correct. Now I think the conclusions to reduce the memory required include:
1. Use the symmetrical plane boundary if your structure is sysmmetrical.
2, Config thin filmconductor by the altenative sheet with finite conductivity boundary.
3, Choose a litter bigger convergence restriction.
4, Using PML to substitute radiation boundary.
I have not poven the 4th approach. But I think if you have the similar question, you can try one or all of these methods.
For reducing airbox to decrease the memory consumed, I can't say. I choose the minimun required distance from the radiation component is a quarter wavelength. I have no idear if I keep slander the size of the air box. Hope someone can comment this point.
Thanks again for all of you!