About the poly resistor for analog circuit design..

poly resistor model

In addition to Rsh, I have seen the model card which includes the voltage drop coefficient (2nd order), temperature coefficient and Rend. Length and width off set are considered as well. I wonder if the complicated poly resistor model is really necessary for analog design? It is a hard work to model the device with so much parameters(W,L,Rs,Rend...) after all.

resistor model rend

The temperature and voltage dependent changes in the resistor values might not be very important when the ratio, rather than the absolute value of the resistors is important.

But in some cases, such as CONSTANT GM BIAS CIRCUIT, RC based Filters etc, the absolute value of the resistor is important. For such cases, the use of accurate resistor model is beneficial when checking for the corners [temperature, process, VDD variations etc].

I hope the above information would be of some help.

poly resistor rend

The main aspect of polysilicon when designing as a resistor : grain size.

Polysilicon as we know is not crystalline in nature; more of semi-crystalline. Therefore, the effective resistance is not "linear". To achieve near linear resistance (ohms law), you need to have a minimum width for poly. Then use the width as 1 block size and then calculate the length "blocks" size and then you lay it out. The model becomes slightly better stabilized adn therefore, you do not really need to worry about second order effects, unless if you are going to be too precision. (In that case, follow the above and make all resistors as big linear one; no bends whatsoever; then have 3-D enclosure; the problem LVS maynot get it!!!)

However, after layout, the monte carlo analysis will help determine whether temp, voltage, or etch effects has any appreciable impact on the circuit. Monte carlo analysis are worthwhile only when you get the minimum width correctly.

Personal Experience: When designing my opamp, the min width was about 7um (grain sizeis much smaller). I worked it as 14um (too conservative but works!!). I needed about 3.8kΩ. Each sq ~= 1kΩ. Therefore my length was about 50um. Its bigger but the grain size will not affect my opamp performance. Thats what I needed. If you are not having much space,then go for 10um or even 7um. It should work. My fabricated opamp worked very close to simulation....

Finally, make sure you have enough contacts on both ends... they can depreciate your performance. Rule of thumb: if the drop across resistor end and the output of a circuit which feeds the resistor ≥ 5~10 mV; then you have a problem... Figure out contact resistance and make necessary number of contacts so that a parallel resistance is smaller.

Hope it helps..

Srivats