How to know unCox and upCox when calculate gm?

[mpig09]

Hi all:

I am designing a OPA and I set the input gm = 10uA/v = upCox*W/L*(vgs-vth)=(2*upCox*W/L*Id)^(1/2),
but I found the upCox doesn't a constant value,
it changes with W/L or vgs or vds.

How to know the upCox value?


Thanks for your reply.

mpig

 

[tompham]

Hi mplg09
You are rite. UnCox (UpCox) depend on W/L or vgs or vds. In submicron (45nm or below) the quadratic formula of Ids doesnt work well. I think you should you gm/Id methology (from UC Stanford lecture). This method you dont care about process parameter

 

[mpig09]

Hi tompham:

I studied some doc of gm/id method before,
if you have used gm/id method, I have two problems of this method:

when simulation gm/id, does the vgs and vds need to a constant value?
and the value need to the same with the circuit application?

If you have any suggestion, please let me know.
mpig

 

[dick_freebird]

One problem is that u0 becomes meaningless when you
get into velocity saturation and such things (i.e. when
the long channel assumption becomes invalid).

u0 is a model fitting parameter used in the classical
device model equations. You might extract a reasonable
value by measuring a device of long dimensions, and
you might fit the geometric effects by measuring and
refitting short-L and short-W devices based on the large
device fit.

I think you want to simulate your circuit at corners of
gain-stage and output-stage Vds(es); many topologies
see large gain variation across the common mode output
range (lose plenty, when you creep into linear region
near the endpoints of output swing, and there will be
a "sweet spot" you'd prefer to operate near, by higher
level circuit biasing perhaps).