Dear all.

Suppose I have a classical CMOS bandgap in a large mixed signal environment where the ground is bouncing. The ground bounce is a few milivolts, at a certain beat frequency which is not very high. The current solution is to have the bandgap voltage directly connected to a pad where some large capacitor decouples it. This capacitor is sufficiently large, and keeps the voltage steady in this environment.

Now the issue is that we see 1/f noise popping up when we operate the chip on long time intervals (say miliseconds to tens of seconds). What to do?

Obviously, enlarging components does not suffice. The components are already quite large. I would like to use chopping to remove the 1/f noise. But since the capacitor is in the loop, the bandwidth of the bandgap circuitry is very low. So in that case the chopping frequency should also be low (correct me if I'm wrong), to allow the circuit to settle. But as the chopping frequency becomes lower, the  improvement in terms of noise is also reduced. =(

Anyone an idea to solve this trade off between the filtering of ground bounce and removing 1/f noise?

Thanks for your reply.

So suppose I need a certain chopping frequency to get my noise spec.
This will limit the size of the capacitor I can put at the output node.
The capacitance will determine how resistant the circuit is against ground bounce.
But suppose the capacitance is too low to get rid of the ground bounce. Then, what can I do to obtain both specs?

Thanks for the idea Tosei.

I'll have to verify what the largest 1/f contributors are for the topology you propose, but I think you are correct. =)

Hello RobG

Your observation about burst noise is quite correct. In fact, what I describe as ground bounce, you could also name burst noise (correct me if I'm wrong). The burst noise present on the ground/substrate is due to the fact that the power consumption of our digital processing block is signal dependent. Now, for multiple channel application, the effect is multiplied. Suppose the block works with a series of high input value and then low values and then again high values. In that case the ground bounce is large, and becomes visible in our reference voltages.

The 1/f noise is something we observed orthogonally to the ground bounce. I know that high resistive poly can have some 1/f, but this is quite poorly modelled at the foundry. I will try to incorporate some numbers from other foundry to see whether it is significant.

By the way: do you think I need to fear the combination burst noise - chopping? Is it possible that they multiply and downconvert?

If I apply zero input signal, and if I sweep over time, at the digital output higher noise is seen as function of time. If I plot the variance over time, it really gives me the 1/f characteristic.

Your remark about the rectifying capability of the bandgap is an interesting one! The only coupling can occur via substrate/ground or supply, as other disturbing signals are >50 um away. Most of the wires are shielded with ground. So if I understand you correctly, this ground bouncing (basically a 5mv pulse of a few us every 100us), could show up as as low frequency noise at the output of the BG? Interesting! I will check this effect with a simulation on the extracted version.

I will have a range of chopping frequencies available, so I can always select another frequency externally if some noise source appears to be on the chopping frequency.