In "Know the sometimes-surprising interactions in modelling a capacitor-bypass network" (abbreviated to "Know the..." when referred to here), Tamara Schmitz of Intersil and I provided some simulation background behind the interaction between multiple decoupling capacitors used in parallel. The series inductances inherent in the capacitors cause several resonant peaks and dips in the impedance response, sometimes at frequencies that might be critical for circuit operation.
We advised that, if you intend to use decoupling capacitors of different values in parallel (there's almost always more than one capacitor attached to the power rail), you'd better be sure that the location of the resonant peak won't cause trouble in your circuit. But we didn't say what kind of trouble that might be.
So this series of articles is an attempt to quantify what's happening on the supply line of a representative analog circuit with regulated supply rails and decoupling capacitors. We'll do this entirely in simulation - mainly to show that this is both possible and revealing - and the path to interesting results will prove a somewhat bumpy one. Over the course of the six articles we will:
- look at the overall supply impedance seen by a component when we include the decoupling caps, the voltage regulator and the board traces providing the component's power. In passing, we'll discover just how supply voltage affects the value of some ceramic capacitors;
- see how such an impedance responds with a characteristic voltage transient when you 'ping' it with a small test current step such as an IC might demand from the power rail. Choice of capacitor dielectric turns out to have a significant effect;
- see how these supply variations punch through to the output of an op-amp running on these supplies;
- discover that many op-amp simulation models, used for this purpose, are so inaccurate that they can produce seriously misleading and even physically impossible results;
- run a signal through the attached op-amp, drive a load, and see how the actual load current interacts with the modelled supply, and how this affects the amplifier output;
- see that simulations provide an objective approach to selecting decoupling capacitors in order to alleviate a previously poorly documented effect on the precision of op-amp circuits. And discover another way in which many opamp simulation models are so inaccurate they are useless for this work!
Paralleled decoupling capacitors - the story so far
If you didn't catch them first time round, why not first review the article linked to above. The graphs in that article show the dramatic impedance excursions possible with paralleled decoupling capacitors of different values. A valid criticism of these plots is that they just refer to the capacitance that the device sees. They don't include the impedance actually already present on the power supply rail, determined by the voltage regulator (which usually also has its own output capacitor), and the tracks or power planes that connect them all up. They don't reflect any loss mechanisms that might lessen the severity of these effects.
First step, therefore, is to rectify these omissions. Now, instead of just looking at just the theoretical impedance of the decoupling capacitors themselves, we'll connect up a 'real' regulator IC to 'real' capacitors and connect all of this to a 'real' amplifier which is handling a 'real' signal. All in our virtual world, of course.
For all the simulation work in these articles I've used SwitcherCad 3, which uses their LTspice simulation engine, and is available free from the Linear Technology website. If anyone is interested in experimenting further with the material presented here, I'll be uploading the test files to the LTSpice Users Group, so that you can reproduce this work and adapt the fixtures to your own needs.
I had loaded SwitcherCad up as part of another project to fearlessly evaluate, on your behalf, some of the free software out there for circuit simulation, and for filter design (something close to my heart). The LTspice engine seems to be admirably robust, accurate and speedy, and SwitcherCad comes with a large library of Linear Technology op-amps, regulators and other devices, which I used here as typical of the parts that would be used on a quality, precision analog design.
After some browsing through the SwitcherCad library I chose the LT1761-5 and LT19645 as the +5V and -5V regulators for the dual supplies on this virtual project, on the assumption that it will be handling bipolar analog signals. These are low-dropout regulators (LDOs); this type of regulator is commonly used for local regulation of power on analog subsystems.