spectre iip3

I have a problem with my IIP3 simulation. In my receiver system, I have an LNA and a Mixer connected together and I want to simulate the IIP of the whole chain by using Spectre. I follow the manual from Cadence by using PSS+PAC analysis to simulate it.

The result shows that the slope of the third-order curve is not exactly 3. It varies around 3. Are there any ideas about simulating IIP3 to get the slope for the third-order curve equal to 3?

Thanks in advance

try to use conservtive, minimize simulation max step. use qpss.

Don't care about the point where input power is too small and the point where the input power is too large.
There is another way to simulate IIP3: input two frequency nearby and do the FFT with the output. formula 2.33 in "RF Microelectronics" by B. Razavi shows how to derive IIP3.