I already had a utilization of 65% with >400k standard cells (w/o the macros) and it fit about half the logic I wanted, so PnR improvements won't get me there. Proprietary synthesis would probably compress the logic quite a bit, but working with proprietary tools is not an option for this project. Except for larger slots I currently don't really see a solution. If it helps, I think we would be able to pay a reasonable premium for a larger slot and would not necessarily need as many samples.