Amar,

  In general, my feeling is that simulating a VCO is not a place to
consider cost-performance. They are difficult circuits to analyze
and are sensitive to numerical noise because the their gain, kvco,
tends to be high. In addition, VCO tend to be relatively small so the
penalty of simulating with high accuracy is not as high. Having said
that, I would try one of the following options when using Spectre:

1) moderate
2) conservative
3) conservative w/ method=traponly
4) moderate w/ relref=alllocal or pointlocal

In general, using the trapezoidal integration method is better for
oscillators, there is a detailed discussion of the topic in Ken Kundert
SPECTRE book. If you are seeing differences between moderate
and conservative, then options # 3 & 4 may help explain the
difference.

                                                    Best Regards,

                                                       Sheldon