Opamp and LDO decoupling capacitors, which are necessary?

Question

I am planning to use the the LT3032-5 LDO to power an OPA657 on a single PCB. The OPA657 datasheet recommends 0.1uF and 2.2uF capacitors on the supply pins, and the LT3032-5 recommends 0.01uF and 2.2uF capacitors on its output pins. The high frequency 0.1uF/0.01uF capacitors should be as close to their respective pins as possible, but can't the larger 2.2uF be somewhat farther from their pins, and possibly shared between the devices?

Are (4) 2.2uF capacitors necessary here? The supply line traces from LDO to opamp will be no more than 1cm in length on the PCB.

When can devices "share" decoupling capacitors?

Answer 1

The LDO needs local bypassing for improved stability. The op-amp wants local bypassing to shunt local noise, as well as filter incoming noise.

That word ‘local’ is intentional: the connection to the device and its caps should be low impedance for the caps to be their most effective. If you try to ‘share’ them you’re adding impedance and you’re reducing the recommended capacitance.

Whether or not that added impedance and reduced bulk capacitance has a material effect on your system is hard to say without modeling it using your exact physical layout with a power integrity tool, using your expected signals. Which, to be honest, hardly anyone does unless you’re building something high volume or very high cost, because the tool and person-hours to drive it are very costly. In your case, all to save two 2.2uF caps, which might cost 5 to ten cents each.

Now, let’s talk about the devices you’re using.

The LT3032 data sheet goes into extensive (really, a lot) detail on its output and bypass capacitors, with some specifics on values that ensure stability. They recommend something like 4.7uF total; you’re proposing to shave that to just 2.2uF for both the op-amp and the regulator. That’s risky right there.

Also bear in mind that ceramic caps have bias effect: they will be lower capacity depending on voltage (5V in your case.) This effect varies depending on the material in the cap (X7R, Z5U, etc.)

Finally, with a slew rate of 700V/us and a GBW of 1.6GHz, the OPA657 is a pretty darn fast op amp. Its bypassing is not nothing: it needs to perform well.

Let me put it this way: hardly anyone regrets having more bypassing than is absolutely needed; they certainly regret having not enough. The ‘extra’ (but not really) caps are cheap insurance.

This Murata doc gives extensive insight into bypassing design: https://www.murata.com/-/media/webrenewal/support/library/catalog/products/emc/emifil/c39e.ashx

Answer 2

From the OPA657 datasheet:

If necessary, place the larger capacitors somewhat farther from the device and share these capacitors among several devices in the same area of the PCB.

So per the datasheet you do not need all 4. You can share the higher value capacitors as long they're in the same general area of the PCB.

Answer 3

It may be the case that your intuition is correct and that these devices can, in fact, share the decoupling capacitors.

Unfortunately, an evaluation that's comprehensive enough to confirm or deny your suspicions would be very difficult to carry out. You'd need to acquire detailed models of your LDO and op-amp from the chip manufacturers (good luck getting that unless you're gonna buy $1M worth of chips), plug those models into a sophisticated and hard-to-use piece of simulation software, and accurately model your range of expected input signals.

Much better to just include the recommended decoupling caps unless you have a compelling reason for not doing so. In your case, you'll probably be able to get away with it, but - as indicated in another answer - different devices may need their own bulk decoupling capacitance. Just because it works for you in this design with these specific devices doesn't mean that it would be reliable in another similar situation.