To help understand polymer capacitors better, I made an episode of The Learning Circuit where I compared measurements. While creating the episode, I captured over 150 screenshots and about 50 reference waveforms. The final video used less than 20 of them! While they are not all here, I thought these additional pictures would be interesting.
There were two devices under test (DUTs). The first is a 1.2 volt 2 Amp DC-DC converter module from TI. It is an evaluation board for the TPS62097. My goal was to simulate replacing an MLCC with a polymer capacitor, in an existing design. I picked this board because it has built measurement points and selectable frequency output. Additionally, it used an 0805 package for its output filter capacitor. An 0805 SMD is relatively easy to solder.
A Commodore 64 was my other victim, I mean, DUT. Its onboard linear regulators use Aluminum Electrolytics. Click here for the C64 blog post with pictures comparing traditional wet aluminum electrolytics to polymer aluminum capacitors.
In the video, I showed the typical application schematic for the TPS62097. I fully expect some people to point out that the schematic shows a 1.8 volt output while I say "1.2 volts." Oh well. I didn't have time to re-draw it, and the schematic of the physical board is too detailed for a video. The evaluation board comes configured for 1.2 volts out. However, by changing the feedback resistors, it can output anything up to 3 volts? 5 volts? I don't remember now.
Please note that in most cases the volts per division setting changes from picture to picture. My goal was to get the best vertical accuracy for the Vpp measurement.
Ceramic 22 µF No Load and Load
These were both shown in the video, but not side by side. You can see how noisy the switcher gets once there is a load on the regulator.
22µF Polymer Tantalum No Load and Load
Similar to the above measurement, but with the polymer tantalum replacing the ceramic. The left unloaded picture shows the higher ESR ripple from the T529. With the load turned on, the switching spikes are now riding on that ripple. The surprises here is that the total peak-to-peak noise, with the transients, is significantly reduced.
Loaded 22µF Ceramic compared to 22µF Polymer Tantalum
In the video, I did not show this conclusion shot. The left is the ceramic, and the right is the polymer tantalum. It is comparing their loaded performance. Please take note that the Polymer Tantalum has an 11 mV/div scaling. Which is almost 2X more sensitive than the ceramic's 20 mV/div scaling. Pay special attention to the automated Vpp measurements.
The ESR of the T529 increases ripple voltage a little bit. However, as I said in the video, I think it also serves to dampen the transient spikes, which is why the overall ripple appears to be lower. It is questionable how vital these spikes are in this measurement. Depending on the trace design, they may get filtered out. That said, let's compare the impedance performance of a ceramic and polymer.
22 µF Ceramic and 22 µF Polymer Tantalum ESR Plots
I wish I had shown this comparison in the video. Here I am using KEMET's KSIM to compare capacitor types. The T529 is the Polymer Tantalum I used in the video's measurements. The ceramic shown in this graph, however, is not. The one on the TI board is from TDK. Yet, I don't otherwise have a simple way to compare the two ESR plots (without generating them on my own.) But the key is that the T529 has two orders of magnitude more ESR than the ceramic.
Keep in mind; this is a tradeoff that has to be considered an existing design where you are trying to replace a ceramic capacitor.
22 µF Ceramic and Polymer Tantalum with a 100 nF Ceramic Filter
Those transient spikes bothered me. Fortunately, the TI evaluation board came with some extra capacitor pads for precisely this reason. I was able to add a 100 nF capacitor to see how it affected the performance. You might argue I'm adding one more capacitor than the original design. However, I think these results show that the 100 nF filter capacitor should have already been there.
Just based on my measurements, it appears the absolute peak to peak ripple of the polymer is lower than the original ceramic. However, I would also point out that the harmonic noise of the polymer is going to be worse than the ceramic. The ceramic's ripple is composed of relatively simple sine waves. Picking a value other than 100nF could remove or dampen those wave components. However, the polymer's ripple is a bit more complicated.
One more thing... well 2.
Knowing that the 100 nF filter capacitor helped with transient spikes, I started thinking about what else could reduce the overall ripple. The ESR of the Polymer Tantalum is the next point to tackle. So as an experiment, I replaced the single 22 µF with two 10 µF Polymer Tantalums. These individual replacements had the same ESR as the larger one. So by putting them in parallel, I effectively dropped the ESR in half. The result? Even less ripple. (This result includes the 100 nF filter capacitor.)
As I said in the video, the takeaway here should be that it is possible to replace an MLCC with a Polymer capacitor. However. You need to take care when doing so because polymers are not a one-for-one or drop-in ceramic capacitor replacement.
Click here for the C64 blog post comparing traditional wet aluminum electrolytics to polymer aluminum capacitors. There is also a post where I go into detail about a part of the C64's power supply design.