raja.cedt wrote on Dec 7th, 2011, 4:26am:but i doubt about the amount of degeneration because of it's huge value. any how i would like to ask now a days is any one using gm=1/r ckt for biasing?, because in this lower technology it may not work as we expect in theory due to lack of square law.
Thanks,
raj.
I feel like I'm part of a conversation where everyone but me can see another poster's responses as I can't figure out what you folks are talking about
.
Raj - what sort of bias
should they use? If we assume Larry's circuit is a constant gm bias (M5 is a multiple placement of M7) then it should properly bias a PMOS diff pair operating at the average current density of M5 and M7. Under these conditions the gm will remain constant over temp (well it would if a zero-tc resistor was used). I think this will be true from subthreshold to velocity sat, and to different devices such as bipolar.
Regarding the "positive" feedback path, the signal path is counter-clockwise. Assuming M5 is multiple placements of M7, at
low currents the path is positive feedback. This is what you want at startup since this keeps increasing the current in the devices until the gain from M7 to M5 is reduced to 1 by the degeneration of M6. At this point the feedback isn't really positive or negative.
rg