You might want to add one of these components to your decoupling network
In a previous article, we looked at the role of decoupling capacitors, as well as the difference between decoupling and bypassing. A decoupling capacitor sometimes called an RF decoupling capacitor, provides the same functions as a bypass capacitor, but it also provides another important function in that it compensates for changes in the ground potential as an IC switch.
There is another important point involved in designing your PDN to ensure power integrity. This is the role of inductance when designing your PDN. In high-speed designs (which is every design nowadays), the decoupling circuit is generally purely capacitive, until you start looking at sufficiently high frequencies. Now there is the inductance that can produce a large transient response in the PDN. This brings up two questions:
The answer to #1 is "Yes," but as we'll see, going the route of #2 is more practical and is standard practice in the industry. As we'll see, attempting #1 gives us a chance to learn plenty about real capacitors, inductance in a PDN, and what is decoupling.
Designing a decoupling network is not a simple task. With lower frequency circuits, using an RF decoupling capacitor was sufficient for decoupling. The self-resonance frequency of many smaller capacitors was still somewhat higher than the knee frequency for many logic families, thus it would be difficult to drive a power bus to resonance during switching. Furthermore, decoupling capacitors would also act as a bypass capacitor to compensate for potential changes as ICs switched.
With faster logic families, knee frequencies can now coincide with the self-resonance frequency of the equivalent circuit formed by the bypass/decoupling capacitor, the power supply decoupling bus, any nearby bypass/decoupling capacitors, conductors that connect components, and the components themselves. This creates the potential for ringing in the power bus with high-speed circuits as logic gates switch. Under repeated switching, this would cause a driven resonant oscillation in the power bus with high amplitude. Just as was the case with ground bounce, a single switching output on an IC may not have much of an effect, but many components switching simultaneously can produce significant ringing in the power bus and large changes in the voltage level seen across an IC's power pins.
For this reason, inductance in a PDN is seen as a bad thing: it creates higher impedance throughout the PDN's impedance spectrum beyond a particular frequency limit. High broadband impedance is bad for broadband digital signals as these signals will transform the transient current into a larger voltage throughout the signal bandwidth. At the high current draw, ringing in a power bus can exceed tolerances on core voltage levels near one of the resonance frequencies in the PDN. Some guidelines suggest adding a decoupling inductor, capacitor, and sometimes a PCB resistor to bring ringing within limits. It's worth looking at exactly how inductance affects the power bus and ringing, and what a "critically damped" PDN might look like.
As discussed in the previous article, the equivalent RLC model for the RF decoupling capacitor may be underdamped, and you can try to bring this circuit as close to the critically damped case as possible. However, you will need to consider the entire equivalent circuit for the decoupling capacitor and the rest of the system.
Ideally, you want to suppress ringing in a few ways:
#1 and #2 might be okay for an analog PDN as you should only care about what happens within a very narrow bandwidth. #3 is more important for digital components, which have broad bandwidth.
These three methods are somewhat mutually exclusive. Adding a decoupling inductor in series between the RF decoupling capacitor and an IC will increase the impedance seen by any high-frequency signals (including a ringing signal) propagating towards the load, but it will also decrease the resonance frequency. Additionally, it will decrease the damping constant by a greater level since the resonance frequency is only inversely proportional to the square root of inductance. Therefore, if the response from the decoupling circuit is already overdamped, adding a series PCB inductor between the decoupling capacitor and the load decoupling can bring the response closer to critical damping.
If the response seen on the power rail is already underdamped, then you need to increase the damping constant and decrease the ringing amplitude. One simple way is to use a capacitor with a larger equivalent series resistance (ESR). Note that electrolytic capacitors tend to have larger ESR values. The other option is to add a decoupling resistor and decoupling inductor before the relevant IC, as shown in the circuit below:
Full decoupling circuit with a bypass capacitor
Note that L in the above model is equal to the inductance of the conductor (e.g., power plane inductance) leading to the load plus the value of the decoupling inductor. The damping constant in the equivalent RLC network formed by the load, decoupling capacitor, L, and R is equal to the usual value for an RLC series circuit. Adding the inductor decreases the natural resonance frequency while adding a small resistor R can increase damping in the circuit. When R is equal to the critical value shown above, then the transient response in this circuit can be critically damped.
PCB resistors are great for adding damping. Unfortunately, you lose power, so a decoupling resistor is only good when it has a low value so that it does not drop too much voltage. An alternative way to look at damping would take out the PCB resistor and just consider the decoupling/bypass capacitance with any inductance between these and the load.
The network shown above will increase the DC voltage drop throughout the PDN, thus there is an alternative decoupling network that approaches critical damping:
Alternative decoupling network with a bypass capacitor
These equations tell you what the limits are on the bypass and decoupling capacitances for given ESR, ESL, and L values, which will give you critical damping. Note that L may not necessarily be a real inductor; we could be looking at the inductance of the power rail, although in such a case we would have R approaching zero and ESR controlled below a certain value.
In this decoupling circuit, the critical resistance is the same as that shown in the earlier network. However, there is also a restriction on the values of the decoupling and bypass capacitors (shown above). Increasing the damping resistance between the limits shown above will cause the response to move into the overdamped regime, thus slowing down the overall response from the RF decoupling capacitor.
It's important to remember the role of inductance in any PDN, whether it's a parasitic element or intentionally placed. The circuit viewpoint states that a bypass capacitor placed between the power and ground pins on the load will provide a low impedance path to ground for high frequencies, basically lowering the overall PDN impedance below the capacitor's self-resonance frequency and making the PDN look like a low-pass filter. Inductance is counterproductive and eventually turns the impedance purely inductive.
This should illustrate the point of placing RF decoupling capacitors on the PDN along with bypass capacitors near large ICs. The decoupling capacitors provide a set of low-impedance elements in parallel, to create an overall low impedance in the PDN.
When designing a PDN for your PCB, you’ll need the layout and simulation tools in Altium Designer to ensure your board is free from power integrity and signal integrity problems. Using circuit simulations will help you qualify your component selection and layout, as well as allow you to visualize electrical behavior in the PDN during a transient event.
Contact us or download a free trial if you’re interested in learning more about Altium Designer. You’ll have access to the industry’s best routing, layout, and simulation tools. Talk to an Altium expert today to learn more about PCB resistors or capacitors.