Designing a circuit is a lot easier on paper, where components have well-defined values, or lacking that, at least well-defined tolerances. Unfortunately, even keeping percentage tolerances in mind isn’t always enough to make sure that circuits work correctly in the real world, as [Tahmid] demonstrates by diagnosing a buck converter with an oddly strong voltage ripple in the output.
Some voltage ripple is an inherent feature of the buck converter design, but it’s inversely proportional to output capacitance, so most designs include a few smoothing capacitors on the output side. However, at 10 V and a 50% duty cycle, [Tahmit]’s converter had a ripple of 0.75 V, significantly above the predicted variation of 0.45 V. The discrepancy was even greater at 20 V.
The culprit was the effect of higher voltages on the ceramic smoothing capacitors: as the voltage increases, the dielectric barrier in the capacitors becomes less permittive, reducing their capacitance. Fortunately, unlike in the case of electrolytic capacitors, the degradation of ceramic capacitors performance with increasing voltage is usually described in specification sheets, and doesn’t have to be manually measured. After finding the reduced capacitance of his capacitors at 10 V, [Tahmid] calculated a new voltage ripple that was only 14.5% off from the true value.
Anyone who’s had much experience with electronics will have already learned that passive components – particularly capacitors – aren’t as simple as the diagrams make them seem. On the bright side, they are constantly improving.
Coping With Disappearing Capacitance In A Buck Converter
Class 2 ceramic capacitors are probably the most popular way to start learning about Standard Datasheet Lies, the nice-sounding specs that are technically not quite lies but are applicable to approximately 0% of real-world use cases. Like the “75 A” surface-mount MOSFETs that can only handle 15 A unless soldered to a copper brick immersed in ice water, or the “85°C” electrolytic caps that have a lifespan of a few days anywhere near that temperature.
It was way worse before companies started to have easy to browse simulation results (like Murata’s SimSurfing). Now it’s a rite of passage but you’re not tearing your hair out.
Lies, damned lies, and product specifications. “Rated for 2000 Watts… [under breath] for approximately two milliseconds, at which point it becomes a rapidly-expanding cloud of incandescent vapor.”
“No root mean square figures, sorry.”
That’s why you don’t read the “Features” part of datasheet. You check SOA chart for transistors. And yes, it can handle 75A, at 10us pulse width. And learn about thermal characteristics, and how to read them first.
Also when it comes to capacitors, unless your design is under control of accountants, double or triple all filter capacitors in capacity. Also SMD MLCC capacitors are fragile and can crack. Also for any converter, you can’t go wrong with Pi LC filter.
Datasheets don’t lie, they are designed to advertise the product fist, then they include all necessary information. For example a uC datasheet might tell you “High performance” followed by “Low power”, and it’s a perfectly true pair of statements. But you need to get to DC characteristics table to figure out it’s either high performance or low power. And then calculate, if it’s better to run program at low clock for longer, or run it ASAP and then go to ultra low power sleep mode. Turns out, paradoxically that second option might be more efficient…
It’s a serious problem for EE’s. MLCC voltage-coefficient of capacitance, and aging give you much much less capacitance than you expect. It varies amongst different manufacturers. Some dielectrics age badly within the first 48 hrs!
Continental Automotive parts maker had big problems with this causing a TI LDO to oscillate. Not sure if it was a safety system like ABS. They lobbied the capacitor industry to add aging data and you’ll now see it in some datasheets. They also train their EE’s on how to select and rate MLCC’s. See PCNS 2019:
https://passive-components.eu/wp-content/uploads/2019/10/HIGH-CV-MLCC-DCAC-BIAS-AGEING-CAPACITANCE-LOSS-EXPLAINED.pdf
You should see it in most aecq qualified parts the thing might be on giggidy key or whatnot is you only see the datasheet common to the family of parts … only when you are in the market for millions of units for automotive or above then you start seeing different sheets
I had that issue last year with a ti chip the public datasheet didn’t cover what the engineer was setting things up. Once I talked with our rep a new confidental datasheet showed up and it was twice as many pages