A question about S12 of LNA

I design a LNA using two stage CS architecture, the S12 is -30dB, seems worse than cascode architecture.
Could you tell me s12(reverse gain) usually should be how much?
Are there any methods to improve it?
some reference paper may be better , thank you!

I think the main causes of S12 are the parallel feedback due to Cgd of TR and the series feedback due to imperfect ground.

To reduce the parallel feedback effect due to Cgd, use the parallel inductor between gate and drain, so at resonance frequency you can get more isolation.

To reduce the series feedback effect, minimize ground routing and use more vias.

BR.

That's a very common problem for 2-stage CS or CE amplifier operating at GHz range.