lunren wrote on Mar 24th, 2010, 11:24am:While with 500fF cap to replace the switches, the noise become 500uV.
You seem to have very large switches if the equivalent load is 500 fF!
Yes, it is a group of switches, but it should not generate that much high noise.
Quote:I have also tried to sim the AC noise of a simple NMOS connected to a cap (the other side connected to a fix bias voltage). It turns out that even the NMOS is turned off, it still generate a lot noise and the noise depends heavily on the bias level at the other side of the NMOS.
If simulating just a simple switch + cap combination with standard noise analysis gives you a lot of noise, then it is an indication that your switches are not sufficiently OFF in the OFF state. So although the off-resistance of the switches is high, it is not high enough to prevent the noise from entering the frequency range of interest. Incidentally, if you consider a large enough frequency range (from low-f to high-f), then you should see a fixed amount of total integrated noise in this simple setup, which should only depend on the cap and not on the switch anymore.
Look at the noise plot in this simple case across frequency.
You may also want to see how to get the switches to really turn OFF, maybe using other transistors which leak less, or considering pumping the gate to a more negative voltage to turn your switches OFF. Try a simple experiment, where you sweep the gate voltage negative and check the point at which the noise spectrum of the switch disappears from the frequency range you are interested in.
Vivek
Actually if the gate was connected to much negative voltage, the integrated noise is only 7.8uV (much less than sqrt(KT/C)=20uV) in the simple experiment setup. If you replace the NMOS with ideal resistor from analogLib, you will find that if you set the resistance very high like 10G (the other side just connect to ideal voltage source, which means there is no other noise source except the resistor), the integrated noise is only 0.7uV (where C=10pF). If the resistor is 1M, we could get 21uV which matches KT/C. Any one who can give this some comment?