opamp by gm/Id approach

[amitjagtap]

ugb model of opamp

hi all,
I'm trying to design a low power opamp using gm/id approach.
When i calculated gm/Id according to desired specs. I was arround 380. But normally gm/Id lies between 0 to 29 for a transistor.
I don'n know what to do with this.
Is there any method which convert(by some way) this high value of Gm/Id = 380 to some value which will be included in Current Density Plot of the transistor, so that corrosponding normalized current can be calculated from the graph.
Do u have any comment.........
Thanks...........:?

 

[qslazio]

can you provide more infomation?
It's may because many reason:
1) subthreshold SPICE modeling accuracy
2) different scaling between Id and Gm by simulator
3) wrong scaling option setup in netlist
...

 

[sutapanaki]

I don't know what you are doing but gm/Id=380 is certainly wrong. Because you can calculate 2*(1/gm/Id) which is informative for Vdsat.

 

[amitjagtap]

I selected approx specs as---- Av=60 db,
CL=10pF, UGB=600kHz
UGB=gm1/2*pi*CL
which gives gm1= 37.699uA/V
seleceted branch current ID1=100nA
Which gives Gm/ID=37.699/100 K = 376.99.

Qslazio : I'm using TSMC 0.35 tech level 53
Else i don't think any other mistake i did in graph ploting of Gm/ID Vs ID/W.
Sutapanaki: what u r saying is correct. But does this mean that these specs cannot acheived with selected model file.
What i have in many papers that gm/Id lies in bet 0 to 28.
If during designing, people get gm/iD lets say arround 50, waht the do with this.........Need ur suggestion.

 

[ee_ykhab]

actually your current is extremely small. 1uA is normal in typical design. remember : gm / Id = 2/vdssat. When you increase gm, you need to increase Id 2.

Thanks

 

[amitjagtap]

hi ee_ykhab
Whatever ur saying, i agree with this.
But i'm targeting for low power opamp inwhich current is the main constraint. Thus i have selected Iss=200nA. Normally i subthreshold we can achive high gain as compared to opamp operating in saturation. What i observed that, for Vgs < Vth, Gm value is arround 2 to 4uA/V, off course for some fixed w/L(L=1.4um & W=2.8um)

 

[steve_guo]

Current is too small,
You can simulate a simple case with diode connected MOSFET and try to get it's gm/ID vs ID curve
280 is too high and I think you should reconsider the accuracy of the model.
From my experience,Model has dicontinuity in extrmely low current condition

 

[amitjagtap]

hi steve_guo
Why do u think that 200nAcurrent is too less for low power opamp.
I think it is well above the device leackage current. In one paper that i have seen, they have used current of 50nA in the amplifier.
What do u think.......... plz let me know.

 

[sutapanaki]

Obviously you are trying to put too many things in the same basket for this technology.
1. Based on your Av and if you have gm*ro plots you can choose the L of your transistors.
2. I don't know exactly your topology but let's say it is just a regular diff pair and you need a gm1~40uA/v. Don't fix the current, let's see what current will be necessary. If you assume that gm1/Id~2/Vdsat and you choose Vdsat=100mV-200mV (to have good current efficiency) for the value of gm1 from above you need a current per transistor of 2uA - 4uA. Ok, 1uA if you aim at Vdsat=50mV and weak inversion. But thats about what you can get.

I selected approx specs as---- Av=60 db,
CL=10pF, UGB=600kHz
UGB=gm1/2*pi*CL
which gives gm1= 37.699uA/V
seleceted branch current ID1=100nA
Which gives Gm/ID=37.699/100 K = 376.99.

Qslazio : I'm using TSMC 0.35 tech level 53
Else i don't think any other mistake i did in graph ploting of Gm/ID Vs ID/W.
Sutapanaki: what u r saying is correct. But does this mean that these specs cannot acheived with selected model file.
What i have in many papers that gm/Id lies in bet 0 to 28.
If during designing, people get gm/iD lets say arround 50, waht the do with this.........Need ur suggestion.

 

[amitjagtap]

hi Sutapanaki,
I'm using simple conventional two stage opamp configaration.
I think ur correct. I should not fix the current. Up till now i was fixing current and thus selected very crucial values to get gain and UGB but still UGB is value is low as compared to selected one.
I will try by ur way, I think this will increase some power dissipation of my circuit.
Thanking you......

 

[DenisMark]

The main mistake that you done is selecting current before gm/id value.
You should choose gm/id value at the first (from Gm/ID vs ID/(W/L) curve). For example 22 [1/V] (weak inversion region, Vov=Vgs-Vth<50mV). Than you are able to calculate required current I=1.713uA per branch.
There are no way how make transconductance efficiency higher than process limit.
UGB=gm1/2*pi*CL - valid for single stage amplifier.
For two stage:
UGB=gm1/2*pi*CM=2*gmL/2*pi*CL (for phase margin 60deg). Choose Cm firstly, than compute gm1 and gmL, Choose Gm/ID for both parameters and calculate currents (for gm1 it's better high values ~22, for gmL - lower ~4-6). Don't forget about RHP zero e.g. use zero nulling resistor.

 

[amitjagtap]

hi Denis mark,
First time when i calculated Gm/Id it was arround 380.
After that i selected a maximum value of Gm/Id from Gm/Id Vs Id/W plot such that the transistor will be in subthreshold. I did exactly what u said, because i found the only suitable way to deal with this problem.
For two stage u have written UGB=gm1/2*pi*CM=2*gmL/2*pi*CL
in this i know UGB=gm1/2*pi*CM,
but what is gmL?
can u plz tell me.......

 

[DenisMark]

Easily, transconductance of output transistor (second stage).
More correct condition UGF=gm1/2*pi*CM=2*gmL/2*pi*(CM+CL) if Cm and CL have compatible values.