How far can you trust ceramic capacitors?

Article By : Steve Hageman

Here's a look at some of the common pitfalls when using ceramic capacitors.

There is a general trend of making electronics smaller and with more functionality. This translates into: “Cram it in there,” so naturally designers look to find smaller and smaller parts.

What was on paper at least a 0.1 µF, 1206 sized capacitor, today can be bought in a 0402 size. But is it really the equivalent capacitor? We will take a look at some of the issues here.

As engineers, the best learning that we ever get is when we do something that does not work out properly and then we discover the root cause. Hopefully, this happens long before we get a product into production. Much of what is discussed here is a long sad story of engineering “on the job training,” hopefully, if you aren’t aware of all of these issues, you will be forewarned and know what to look out for.

Ceramic Capacitor Specifications

A lot of us thought that the common ceramic capacitor nomenclature of C0G, X7R, B5X, and Z5U, et al., was a physical dielectric specification. It didn’t help that some manufacturers would say things like “X7R dielectric.” But these three letter designations are not a physical dielectric, they are a performance rating system, and the manufacturer can use any dielectric formulation that they choose as long as they meet the three letter performance specification. Ceramic capacitors of type designation C0G or NP0 (it is a “zero,” not an “Oh” in the designation) are of type Class 1, there are very stable capacitors having very low temperature drift of less than 30 ppm per degrees Celsius. C0G types also have the lowest volumetric efficiency.

Size is where the Class II capacitors come into play. The three-letter numbering system is detailed in Table 1.

Table 1 The three letter class II capacitor labeling system. The most commonly used types in electronics are the X7R and B5X. This designation does not call out a specific dielectric, but rather it is a specification. The manufacturer can use any construction that they wish as long as they meet the specification. For instance, an “X7R” has the characteristics of: A -55 oC to +125oC operating temperature range and a +/-10% capacitance change maximum.

Life (and our circuits) would be so much better if we just all used C0G specified capacitors everywhere. The trouble is they are too big for use everywhere in most modern applications. Anything other than a C0G type is where “capacitors go bad” as we will see below.

When ceramic capacitors go bad—DC bias

My “bad experience” with a ceramic capacitor was when I first decided to go with a largely ceramic power distribution bus in a new software-defined radio. The radio worked just fine but the power bus, which was a 5.5 V bulk supply feeding a large network of 5 V low dropout regulators (LDOs), oscillated. The supposed 0.22 µF, 0402 capacitors that I had used, when biased to 5.4 V simply lost a large portion of their capacitance and the bus did not have enough bulk capacitance to buffer out a low-frequency interaction that developed because of the LDOs. Oscillation ensued!

Luckily, this was easy to fix. I just piggybacked 2.2 µF capacitors on the existing capacitors and carried on with testing the first board iteration. This got me looking into the root cause and carefully studying the capacitor data sheets.

Before this product required an extreme size reduction, I had always used 0603 and 0805 sized capacitors usually rated at 50 V with at least one tantalum capacitor for the power supply rails. Since my power supply rails were 5 V or below 80% of the time, the capacitors exhibited no noticeable bad behavior.

As was expertly detailed by Mark Fortunato [1] and Istvan Novak [2] many of us have had similar experiences. What worked for years suddenly didn’t work with a size reduction.

As shown in Figure 1, the well-known reduction in capacitance versus bias for several capacitors.

Figure 1 Typical curves showing the capacitance reduction of two 0.01 µF, 50 V, X7R capacitors compared to a C0G capacitor versus DC bias (A), and two X7R capacitors versus AC excitation Level (B). There can be large differences between seemingly similar manufacturers of X7R capacitors.

Even though I used X7R specification capacitors for my bypassing, I saw a huge reduction in the capacitance value because I also reduced the voltage rating of the capacitors simultaneously, getting a “double whammy,” which is an effect I call: “whammy squared.”

Novak stated in his paper: “To help the users, all major vendors today supply at least DC bias information with their MLCC parts,” [2] but lately I find this is no longer true. In the last several years, this information has gone missing from a large number of data sheets, and now you have to go specifically looking for it if you can find it at all.

AVX, Murata and KEMET to name a few, have websites that show all sorts of parameters and performance curves for nearly all their capacitors [3], but sadly this information is not generally transferable to another manufacturer’s part. For instance, a 0.1 µF, 10 V, X7R from one manufacturer will lose -4% of its capacitance with a 5 VDC bias, and another can be -35% lower capacitance with the same conditions. This shows that you simply cannot assume that one manufacturer’s capacitor performance is equivalent to any other manufacturer’s capacitor.

Additionally, you may recall that in 2017 there was a worldwide shortage of ceramic capacitors of all types. At that time, manufacturers scrambled to get enough parts to sell. I ran into several production-related issues with not only end users making untested parts changes, but manufacturers also making changes to their parts that caused them to have different DC bias characteristic curves.

Looking around for present-day capacitance versus DC bias curves and you will find manufacturer information that is decades old, and I begin to wonder if it even closely matches reality anymore. It simply isn’t being updated, and more recently seems to be removed instead of updated.

When Ceramic Capacitors go bad—distortion

Another issue I see is distortion. That capacitance change with DC bias may cause an issue with any use of capacitors in the analog signal path. I have seen far too many engineers with a space crunch, simply pick capacitors based on the size and temperature rating. This gets them to design all sorts of what amount to “digital only” capacitors, with disastrous results to their analog signal processing.

To show and measure this distortion effect, I used part of my BlasterAmp project [4], namely the audio output of a Sound Blaster USB dongle, and software for creating very low distortion audio tones, a custom-built 18 bit, FFT analyzer to measure distortion [5], along with the circuit of Figure 2.

Figure 2 NE5534 op-amp circuit for testing distortion. The test capacitor is where I solder my various capacitors for testing. This circuit, when connected to a Sound Blaster USB Dongle and some custom software, produces a distortion free signal to the 16 bit level (> 95 dBc distortion).

The setup in Figure 2 is limited to the 16-bit distortion of the DAC in the Sound Blaster. I measured all sorts of 0.01 µF, C0G ceramic, and stacked film capacitors as expected. They added no measurable distortion to the output.

The fun part came when I measured the X7R capacitors. Normally, I use X7R capacitors in bypassing circuits only, but I have let one or two slip into the signal processing path during my carrier for sure. Fortunately for me, they caused no issues, because they were almost always 50 V rated and that was well below the signal swings being used.

Measurements of two, seeming identical 0.01 µF, 50 V, 0603, X7R types with a 20 volt peak-to-peak signal swing are shown in Figure 3. As can be seen, these capacitors have very different distortion signatures on the FFT plot. Table 2 is a better comparison of the distortion products. One of the ‘seemingly identical’ capacitors has a 2:1 better distortion than the other!

Figure 3 The FFT distortion signature for two seemingly identical 50 V, X7R, 0603 sized capacitors. As can be seen one has significantly worse distortion characteristics.

Table 2 Tabular data for the different capacitors measured in Figure 3. One of the capacitors has twice the percentage total harmonic distortion of the other. Neither has performance even at the 8 bit level (-49 dBc) for the third harmonic!

I also measured some X7R, 0805, 50 V capacitors, and even a 0402, 10 V rated capacitor with similar distortion to the above. The 0402 should have been much worse because of the greater swing relative to its rated maximum working voltage, but it wasn’t. This is what makes me think that the datasheet curves on a lot of these parts are very old and do not match reality anymore. I also biased the 0402 capacitor up to 50 V with no measurable increase in leakage current, so perhaps it is built on a 50 V capacitor process after all? I don’t know, but based on the classic drop in capacitance versus working voltage curve, it should have been much worse than it was.

When rummaging around in my parts for capacitors to measure, I also ran across a 45-year-old Z5U disk capacitor. I thought it would exhibit very bad distortion, but it turned out to be not so bad at all, about the same as a modern X7R! (See Figure 4).

Figure 4 A 50-year-old Z5U capacitor measured just for fun. Its distortion wasn’t as bad as expected.

I took one of the capacitors from Figure 3 and ran the peak-to-peak voltage across it down until the distortion products were in the noise floor and then plotted that data in Figure 5.

Figure 5 Third harmonic distortion versus peak-to-peak voltage applied on a 0.1 µF, 50 V, X7R, 0603 from Figure 3. The measured distortion followed a fitted logarithmic trend line as expected.

A trend line plotted on the data in Figure 5 shows a good fit to a logarithmic curve, this suggests that even if you must use an X7R capacitor in the signal path, if your signal level is low enough you may not notice any distortion issues.

No manufacturer lists distortion as a specification and, as shown above, the change of capacitance versus DC bias data on seeming similar capacitors varies wildly.

All you can do is stay away from anything other than a C0G type or use a film capacitor where distortion may be an issue. Even testing carefully may not ensure success, as you never know when the design or construction of the part may change causing production issues. Yes, that means that size may become a problem, but design tradeoffs sometimes have to be made.

When ceramic capacitors go bad—piezoelectric effects

I learned of this when I worked with some knowledgeable phase lock loop (PLL) designers; they told me that anything other than C0G or X7R capacitors would be an issue. That “issue” is that any dielectric other than what is used to make C0G capacitors use a material that is naturally piezoelectric and, when deformed, will cause a voltage to develop across the part. I think the PLL designers first found this issue when a design showed RF sidebands at the cooling fan rotation frequency. The fan vibrated the PCB, and this vibration caused the capacitors in question to generate sufficient piezoelectric voltage to modulate the oscillator tune line of the PLL, causing the sidebands. Changing the capacitor to a C0G type made the issue go away.

The capacitor industry knows about this, and they call it: “singing capacitors,” as most people learn about this piezoelectric phenomenon in a kind of reverse way to my experience. If an AC voltage is impressed on one of these capacitors, then they flex themselves and, if the frequency, voltage, and mounting are just right, it turns the PCB into a speaker, producing audible noise.

Reading the reviews of many laptop computers sometimes describes an audible whine that can be heard under certain load conditions on certain laptops. They usually describe this as “coil whine,” but it is probably really a “singing capacitor.”

Several manufacturers have modified the design of their capacitors to mitigate the issue and lower the possible acoustic noise [6].

There has been some good work to demonstrate and measure these issues in the past and there is no point in me repeating them here, see Reference [7].

As I mentioned based on those other designers’ experiences, I have stayed away from anything other than C0G and X7R capacitors in my analog designs, only using the higher density capacitors where you have to, like between the power pads of an FPGA, etc., or in strictly digital designs. Although, if you use one of these very piezoelectric capacitors in a clock line, who is to say that some piezoelectric potential cannot cause a switching threshold jitter thereby causing unwanted clock jitter farther down the line? Be careful!

The assumption that I have always had is that X7R capacitors are safe from the piezoelectric issue, which is simply not true as X7R capacitors use dielectric materials that are inherently piezoelectric also. It has just been that to date, the level of piezoelectric charge has been too low to cause me any issues, your mileage may vary as they say. This situation may change at any moment, as my experiences with the 2017 great capacitor shortage showed, so it is best to be very cautious. Or use one of the specially designed low acoustic noise capacitors where appropriate.

When ceramic capacitors go bad—cracking issues

Ceramic capacitors are very fragile. Who hasn’t cracked one or had the end caps fall off? This fragility can be exasperated by using large ceramic capacitors on a thin PCB where the flexing can cause many capacitors to crack (Figure 6). My experience has been that I get scared using anything larger than a 1206 sized part on a standard 0.032- or 0.062-inch-thick PCB. I have even gone so far as mounting the capacitors upright in an I-beam fashion to reduce the possible stresses. Many large ceramic capacitor arrays are even mounted in stress-relieving frames to reduce the possible cracking stress.

Figure 6 If force is applied to any completed PCB assembly, causing any bending (red lines). Then any parts mounted along the bend will experience force at their mounting points (black lines). Ceramic capacitors, being very brittle, usually suffer from the bending force(s) first and crack at the mounting points.

It does matter to some degree what the capacitor construction is, but all ceramic capacitors are liable to be cracked by flex stresses. It is good to keep this in mind and to use an appropriate thickness PCB for the size of the components to reduce the possible flexing. A 0.090- or a 0.120-inch-thick PCB is much stiffer than the standard 0.032- or 0.062-inch-thick material and may be enough to solve any potential issues.

Remember that flex stress not only happens when you physically deform the board. Temperature cycling the completed assembly can also cause enough stress to crack ceramic capacitors.

Some manufacturers produce capacitors with special flexible or soft terminations that allow the solder joint to actually flex some, which can greatly help mitigate the problem.

Another real issue with any component smaller than 0603 is when you handle the board or try to place the assembled board into a stiff/faraday-shielded metallized anti-static bag. These types of metallized bags are quite stiff and can shear small parts off the board very easily. Placing the assembly into a more compliant or padded anti-static bag before placing the assembly in the stiff Faraday shielded bag can prevent a lot of damage issues.

You don’t know what you don’t know

All these capacitor issues are well known and documented, but still not widely recognized by the rank-and-file engineering community. I still see designs today that try to use the smallest possible capacitor in an analog signal path. These designers are sadly about to learn about the drop in capacitance, distortion, and piezoelectric issues firsthand. That is unfortunate, as it is much less painful to be guided around the pitfalls as I have been, so I haven’t suffered as much as I would have if I had to stumble on all these problems personally.

Recently, after the great capacitor shortage, I have re-evaluated how much extrapolation I can make on the piezoelectric rule of thumb that I have assumed, namely that X7R capacitors are somehow immune to the issue. I am not assuming that blindly anymore and am much more cautious. As the TDK application note [7] states:

“The engineer cannot make general assumptions based on … {out of circuit} … measurements alone.”

Carefully test the part you want to use in circuit, but don’t try to extrapolate to other similar parts. They may be totally different, and even worse, they may change next week when the next part shortage arrives.

Bonus: Capacitors aren’t the only possible problem

When Linear Technology started to produce 18- and 20-bit ADCs some time back, they discovered that even the smaller SMT resistors can cause distortion [8]. It turns out that the 12- and 14-bit resolution was pretty easy, and anything greater than 16 bits today takes a lot of careful design where every single part needs to be scrutinized for nonlinearities. Including “tapping” on the completed board to look for piezoelectric effects!

References:

[1] Mark Fortunato, Maxim Semiconductor, 2012 https://pdfserv.maximintegrated.com/en/an/TUT5527.pdf

[2] Istvan Novak, et.al., Oracle-America Inc. http://electrical-integrity.com/Paper_download_files/DCE11_200.pdf

[3] AVX, Murata and Kemit Online Selector / Simulation Tools,

[4] Hageman, Steve, “Simplify testing of embedded analog-to-digital converters”, EDN, June 23, 2022 https://www.edn.com/simplify-testing-of-embedded-analog-to-digital-converters/

[5] DMT9000, 18 Bit, FFT Analyzer will be detailed in a future article.

[6] TDK Corporation, “Singing Capacitors (Piezoelectric Effect)”, December 2006.

[7] Cadwell, John, Texas Instruments, “Stress-induced outbursts: Microphonics in ceramic capacitors” December 2014, Parts 1 and 2,

[8] Hutchison, Tyler, “Matched Resistor Networks for Precision Amplifier Applications”, Linear Technology Design Note 502 https://www.analog.com/media/en/reference-design-documentation/design-notes/dn502f.pdf

This article was originally published on EDN.

Steve Hageman has been a confirmed “Analog-Crazy” since about the fifth grade. He has had the pleasure of designing op-amps, switched-mode power supplies, gigahertz-sampling oscilloscopes, lock-in amplifiers, radio receivers, RF circuits up to 50 GHz, and test equipment for digital wireless products. He knows that all modern designs can’t be done with Rs, Ls, and Cs, so he dabbles with programming PCs and embedded systems just enough to get the job done.

 

Expo booth:

New products & solutions, whitepaper downloads, reference designs, videos

Conference sessions:
  • Internet of Things (IoT)
  • Supply Chain
  • Automotive Electronics
  • Wave of Wireless
3 Rounds Lucky Draw:

Register, join the conference, and visit the booths for a chance to win great prizes.

Leave a comment