Hi Vivek,


I'm not sure how well it would work with on-chip inductors, but have you tried a simple multi-stage matching network? i.e. cascaded L sections. What kind of bandwidth are you looking at and is it your signal bandwidth or the system bandwidth?

Hi Aaron,

I am using multiple sections, i.e. coupled inductors etc. in order to get a good match which seems to be no issue using the textbook-based techniques. However, I need to provide a little bit of gain at half-baud to overcome higher channel and insertion losses at this frequency. I could not tell you the exact frequency but that is not really important.
What I see is that the group delay is poor using simple inductor sections with no equalization.

Hence, I used the scheme shown in this paper (link in first post), but I see that this achieves the peaking not by raising gain at half-baud but by suppressing it at lower frequencies. I am not sure whether there are good ways of achieving gain peaking without ruining the group delay in a reasonable manner. If you could suggest some ideas, then that would be great.

By the way, I found the use of the equalizer idea also in another paper (JSSC Dec. 2007, by Dally, Poulton et al.) They also let the low-frequency gain drop in order to get peaking. This implies that they need to get more gain somewhere later in their signal path. Noise is not critical I think for these chip-chip links described in their paper.

Regards,

Vivek

cheers,
Aaron