Exploring the minimum noise figure of a common source MOSFET + 10nH inductive load in the drain, spectre reports a NFmin < 0dB and for some frequencies also a NF<0dB!! (with Gmin very close to 1). I use a BSIM3 model with a wide MOSFET. The gate is driven by an (unterminated) port and the drain-node is loaded by a port, so that s-parameters can be used to evaluate NF and NFmin. Terminating the input port with a noiseless resistor still renders NFmin<0dB. The "old fashioned" AC noise analysis seems to give more reasonable results, something like 0.4dB Noise Figure (which is higher then NF).
Changing accuracy options doesn't seem to have any effect ....

Can anyone think of a reason why Spectre can ever find a noise factor <0dB???

EricK