Long string of fets model distributed nature of RC better than just on long fet. This might give you a better representation of start time, etc. Most foundries only support channel lengths upto 10um or 20um. If you need longer then that it is good to model it with multiple fet in series.

It is also better for matching, assuming you are trying to get ratios by ratioing lengths instead of widths.

et: You need to be a bit more specific. There are two basic families of FETs: MOS and JFET. MOSFETs are usually enhancement devices, meaning that the gate needs to be taken positive (for N channel) of the source for the FET to conduct. Conversely, JFETs conduct heavily with the gate at zero volts with respect to the source. Most common power FETs are MOS, except for some RF power FETs which are JFETs. If the top FET in the string has its gate tied to the source, the FETs are probably JFETs, as the top one will set the bias current for the string. Voltage drop across the string should also be equalized in some way. Could you be more specific? Bernie Katz.www.technicaldevelop.com. <a href="http://www.technicaldevelop.com">Technical Develop</a>

Hi Bernie,I was talking MOSFET, I haven't seen a JFET since school :-)

I'm guessing your talking about using a long channel MOSFET as a resistor, eg from supply to generate a current for a current mirror. I prefer to use a long length rather than multiple transistors until you hit the model boundary around 20um, but often it is dictated by what's convenient for layout. The only time I use MOS resistors for bias currens are for bandgap startup circuits. The rest of the time it is with poly resistors are usually small enough, and have lower process variations. I used to stack up a few MOSfets of identical size with short channel lengths to generate cascode voltages in an attempt to track process variations. I don't do that so much anymore, but it is OK to use.