Process variation compensation in analog circuit design

[hrkhari]

Process Variation

Hi Guys:

A freind of mine had notified me that it is not crucial in performing a process variation compensation in Analog circuirt design. How true is the claim?. In the context of high frequency circuit design, is it feasible in neglecting the process variation effect in Analog front end design?.


Rgds

 

[daredge]

Re: Process Variation

Analog designers have to make sure their circuits work across the different process corners.

Sometime, for best performance, process compensation circuit is used to allow the circuit work in its optimal operating point.

 

[nxing]

Process Variation

Considering process variation is very critical in Analog design, especially for high frequency circuit.

 

[Teddy]

Re: Process Variation

process variation is important because you are trying to get the yield as high as possible.
In case of univeristy environment it is probably not.
But in any design - rf, analog, digital you have to account for process variation, temperature and power supply - or you are just bad designer

 

[mark_nctu]

Re: Process Variation

You can reduce device mismatch by layout compensation and
don't use accurate resistance and capacitance in your design.

If you need more accurate device in you IC, it wiil be sensitive
to process variation.

 

[ambreesh]

Re: Process Variation

forget analog it is highly sensitive to process variation digital liabraries are designed to meet specs in their worst corner

 

[selvaraja]

Re: Process Variation

The 4-corner models (ws,bs,typ..etc) to predict the fastest or slowest speed of digital ICs, but the grouping simply doesn't apply to analog IC`s. In fact, no grouping is possible; a parameter's influence differs from design to design, die to die and wafer -to wafer. You don't need to vary every parameter of a device, only the major ones. For example, the threshold voltage, the transconductance and the capacitances in MOS transistors.Runnning Monte Carlo analysis varies the device parameters in a random fashion much feasible then running 4 corner analysis, so that every combination of variations is covered. Conclusion is you must compensate the process variation but not neceessary to satisfy four all the corners for analog ICs if you can satisfy for all the 4 corner then you are good analog IC designer .... .... but you must aware of technology limitation too!.

 

[hrkhari]

Re: Process Variation

Hi Guys:

Pls correct me if I'm wrong, according to Mr.Selvaraja, corner parameter compensation is not neccessary, whereas it is substituted by certain percentage variation of other parameter such as threshold voltage, transconductance gm and MOS capacitance in a typical process.

This is in contrast with the claim of Mr.Teddy, Mr.nxing in their respective idea on the importance of corner compliance. I would appreciate if someone could shed some light on it. Thanks in advance

Rgds

 

[mark_nctu]

Re: Process Variation

It is difficult to reduce process variation effect on analog cicuit.

So we need evaluate which parameter dominate circuit performance.

 

[no_mad]

Process Variation

Hi hrkhari,

In my experience, I did involved in the Analog IC simulation with process variation.

We varied the threshold voltage of Nmos and Pmos for example +20%, -20% and typical value. Also with temperature variation from -50 to 125 celsius. It is very important information for us to make sure that our ICs still can work at this condition. As a designer, all these parameters u need to keep it in mind when u r design an IC.

In silicon, it is very dificult to control these variations (threshold, size).

As for digital design, it is true u will have a set of corner models. Worst case, best case and typical model. In digital design (RTL), u dont have the parameters like analog design (which is a lot of parameter u need to consider).

Thus by using these models, u can determine the design speed.

Hope it helps

 

[ambreesh]

Re: Process Variation

We have following corners
1. Supply corner (2)
2. Process Corner (4)
3. Temperature Corner (2)
4. If passive devices are used e.g. resistor (2 )

Have all the permutation and combination for all the analysis to be performed. See that the circuit performs th desired functionality.
The design may never face such harsh conditions, but make your design robust.
All these would show the worst behaviour so u know range of performance of your circuit.

Hope it helps

 

[Vamsi Mocherla]

Re: Process Variation

Actually Mr. Selvaraja has a point. In a circuit where process variations are existing, the digital cells will fail if there are variations and the cell falls in that part of wafer. In analog design and layout, the layout of a transistor or any circuit is much bigger than a digital cell. Hence, the variations in analog are subjected to a distribution(normal or gaussian). Hence we can run the monte carlo analysis to more accurately model all the variations. One thing we should remember is that the values due to monte carlo reside in the limits set by the corners.

 

[bg21359]

Re: Process Variation

does anyone use statistical analysis instead of corner based analysis?

 

[Vamsi Mocherla]

Re: Process Variation

I have used them earlier. TI follows this flow. They give more realistic results.

 

[avanti!]

Re: Process Variation

We have following corners
1. Supply corner (2)
2. Process Corner (4)
Hope it helps

I run simulation for three corners, such as ff, ss and typ. what's your fourth corner?

 

[ambreesh]

Re: Process Variation

Dear Avanti,
Correct me if I am wrong, are u asking what are 4 process cornes.
They are follows
1. Fast N Fast P
2. fastN Slow P
3. Slow N Fast P
4. Slow N Slow P
I would not concider Typical as a corner. Typical would form the center of gravity for all the four corners
Hope I am clear

 

[avanti!]

Re: Process Variation

Dear Avanti,
Correct me if I am wrong, are u asking what are 4 process cornes.
They are follows
1. Fast N Fast P
2. fastN Slow P
3. Slow N Fast P
4. Slow N Slow P
I would not concider Typical as a corner. Typical would form the center of gravity for all the four corners
Hope I am clear

Dear Ambreesh,
You are right.
Thanks.