CMOS Inverter design

Thread Starter

kdillinger

Joined Jul 26, 2009
141
I am trying to simulate a basic CMOS inverter. I am trying to target a 2.5V for my midpoint voltage but it is shifted about 400mV.
I have attached my circuit and calculations for kp, kn, W/L, Vtn, Vtp, and Vmid. It all works out on paper, just not in the simulation.
There are other parameters in the Level 1 Spice models that I cannot zero out because I get floating point errors if I do.
Any ideas?
 

Attachments

SgtWookie

Joined Jul 17, 2007
22,230
Looks like you have a voltage controlled voltage divider with no feedback, and no way to determine what the Vgs threshold is for the MOSFETs.

Without a feedback loop, you have no way to control the output voltage.

Without switched capacitors or switched inductors, you have no control over power dissipation in the FETs.

Go fish!
 

SgtWookie

Joined Jul 17, 2007
22,230
Ahh, OK.

In that case, the output will be indeterminate if you are in the "forbidden zone"; ie: in between a logic 1 and a logic zero.

However, it's easier for electrons to flow than holes. So, if you use a P-channel FET that's complementary to an N-channel FET in the arrangement shown, you're likely to be showing a lower output voltage than you would expect.

Use positive feedback to create a Schmitt trigger. That way you'll avoid the indeterminate zone.
 

vustudent

Joined Mar 11, 2009
38
I am trying to simulate a basic CMOS inverter. I am trying to target a 2.5V for my midpoint voltage but it is shifted about 400mV.
I have attached my circuit and calculations for kp, kn, W/L, Vtn, Vtp, and Vmid. It all works out on paper, just not in the simulation.
There are other parameters in the Level 1 Spice models that I cannot zero out because I get floating point errors if I do.
Any ideas?


Did you make your pmos 2 times wider than your nmos?

Since holes have lower mobility than electrons, so you will have to compensate by making your pmos wider, in order to have balance on and off for your inverter.
 

Ghar

Joined Mar 8, 2010
655
His calculations show that he wants the PMOS 2.5 times larger than the NMOS.
The transfer curve implies the PMOS is actually too large compared to the NMOS...
 

Thread Starter

kdillinger

Joined Jul 26, 2009
141
Did you make your pmos 2 times wider than your nmos?

Since holes have lower mobility than electrons, so you will have to compensate by making your pmos wider, in order to have balance on and off for your inverter.
2.5 times wider to be exact. Wn is 4um and Wp is 10um. They both have the same length.

This is a cookbook example for a symmetrical inverter. Known variables are Vdd, Vtn, Vtp, Wn, Ln, Lp, k'n, and k'p.

Kn, kp, and Wp need to be determined before the midpoint voltage, Vm, can be calculated as shown in my Calc.pdf.

So when I edit the Level 1 spice model there are several parameters that are populated but are not involved in the above calculations, so I do not know if they are affecting the midpoint voltage. If I reduce them to 0 then I receive a 'divide by 0' error.
 

Thread Starter

kdillinger

Joined Jul 26, 2009
141
I don't know how to export the models from TINA. They are existing Level 1 models with all sorts of variables to tweak. So I have to do it the old fashioned way:

pMOS:
vto:-1
kp:80u
phi:600m
is:10f
N:1
pb:800m
mj:500m
tox:100n
uo:600
tpg:1
af:1
l:2u
w:10u

nMOS:
vto:1
kp:80u
phi:600m
is:10f
N:1
pb:800m
mj:500m
tox:100n
uo:600
tpg:1
af:1
l:2u
w:4u

From TINA's help file and other sources around the web:

IS: bulk p–n saturation current
N: bulk p–n emission coefficient
Mj: bulk p–n grading coefficient

af: Flicker noise exponen
t
PB: Bulk p-n potential

TPG: Gate material type


I am no device physicist so I have no idea what these parameters control.
 

Thread Starter

kdillinger

Joined Jul 26, 2009
141
According to Digital Integrated Circuits by DeMassa kp of the pMOS should be equal to kn of the nMOS in a symmetrical inverter.

Attached are the new values you asked me to try and the shift is even worse.
 

Attachments

Ghar

Joined Mar 8, 2010
655
There are two varieties that are often used... you're even using them in your calculation, kn and kn'.

One is just the other multiplied by W/L.
You make kn' W/L = kp' W/L
If kn' = kp' then you don't need to change the W/L
I believe that in this case the model parameter called kp is your kp'

Are you sure you put the correct values in, regarding NMOS vs PMOS?
PMOS should be 16u and NMOS should be 40u.
That curve suggests PMOS is even stronger than before.

This is what I think happened originally...

kp = 10/2 * 80u = 400u
kn = 4/2 * 80u = 160u
Vth = 2.84 V

And your change:
kp = 10/2 * 40u = 200u
kn = 4/2 * 16u = 32u
Vth = 3.14

Those fit fairly well with your plots...

Alternatively leave what you had original (kp 80u for both) but make both have W = 10u and see what happens. I suspect you'd have a symmetric inverter then but it wouldn't fit your calculations.
 

Thread Starter

kdillinger

Joined Jul 26, 2009
141
No, that didn't work. I doubled checked the calculations and my entries in the models.

I think it has to do with the Vt's of the device, but I am not sure. There are several parameters that go into the definition of Vt, but it does not look like they are involved in the Level 1 model so it is hard to say why the transfer graph is shifted.

I changed the Vt of the pMOS and nMOS to be 2.5V which is the same as the calculated VM of the inverter. See attached.

Couldn't ask for a more text book transfer function than that! But because it is ideal, I still wonder what I missed.
 

Attachments

Top