3 ms·

Tantalum caps apparently don't suffer from DC bias capacitance decrease (or at least not as much as ceramics). So you don't have to use caps rated for like 10x

by mordae 3y ago

Tantalum caps apparently don't suffer from DC bias capacitance decrease (or at least not as much as ceramics). So you don't have to use caps rated for like 10x your working voltage to get anywhere near the rated capacitance when decoupling. Higher-rated caps usually come in larger packages. One can frequently see an 6.3V 22uF to 100uF SMD tantalum capacitor being used for the bulk capacitance right after the DC-DC converter. Closest MLCC you can buy is 25V 22uF X7R, which costs slightly more and provides less capacitance (even when comparing 22uF nominal) with further loss due to aging. I have yet to see 100uF MLCC. Not all designs strictly require them, though. When for example dropping 5V to 3.3V using an LDO, having all components decoupled well enough locally, you might get by without bulk capacitance. Some boards are large enough for electrolytic caps that can be combined with ceramics to get you close.

exmadscientist 3y ago
> Tantalum caps apparently don't suffer from DC bias capacitance decrease Correct! This, plus their CV product, is one of the reasons they get used. > So you don't have to use caps rated for like 10x your working voltage to get anywhere near the rated capacitance when decoupling. You still need to derate them appropriately, to 50-70%. They have a rail voltage and category voltage in addition to their nominal voltage. It can get kind of complex, but at least it's well documented in the manufacturer literature. (These days.) Ceramic DC bias derating is not related to voltage derating. It's really most correlated with package size. (Which used to track voltage rating pretty well, but doesn't these days.) There are also more and less expensive dielectric formulations, even within X5R or X7R classes, and guess what? The more expensive ones tend to lose less capacitance with increasing voltage. Go figure. Using the manufacturer tools is critical here. > I have yet to see 100uF MLCC. They're all over the place. You can even get 330uF 1210s from Murata. And probably bigger from someone else. These guys are popular among certain classes of design (mostly size or height constrained). Not my favorite choices though... because as SimSurfing will show you, they're sure not good for that 330uF: you get about at 240uF at 1.8V or 130uF at 3.3V. Which is still pretty great for 1210! But tantalum (AVX TAJ series, your boring basic tantalums) will give you 150-220uF in nearly the same space (1411/B case) depending on how hard you feel like pushing them. (Ceramics can beat tantalum's CV product for the small cases as there's some fixed overhead to just getting the tantalum pellet hooked up.) But then if you use MLCCs you have to deal with the underdamped PDN you just created somehow. (Which isn't hard, unless you forget about it.) > Not all designs strictly require them, though. When for example dropping 5V to 3.3V using an LDO, having all components decoupled well enough locally, you might get by without bulk capacitance. Please don't do this. (Unless you're trying to keep me employed.) Yes, you can engineer the bulk caps out of your design, and that's a viable strategy. But leave the footprints in your first spin so you can make it work, then Muntz them out after you know everything's OK. You will save everyone a lot of grief this way. > Some boards are large enough for electrolytic caps that can be combined with ceramics to get you close. This is actually one of the best overall strategies if the form factor allows it. Ceramics in parallel with tantalums or aluminum electrolytics is a fantastic combination and gives you the benefits of everything and drawbacks of... well, the electrolytics are still big.

by mordae 3y ago

exmadscientist 3y ago

> Tantalum caps apparently don't suffer from DC bias capacitance decrease Correct! This, plus their CV product, is one of the reasons they get used. > So you don't have to use caps rated for like 10x your working voltage to get anywhere near the rated capacitance when decoupling. You still need to derate them appropriately, to 50-70%. They have a rail voltage and category voltage in addition to their nominal voltage. It can get kind of complex, but at least it's well documented in the manufacturer literature. (These days.) Ceramic DC bias derating is not related to voltage derating. It's really most correlated with package size. (Which used to track voltage rating pretty well, but doesn't these days.) There are also more and less expensive dielectric formulations, even within X5R or X7R classes, and guess what? The more expensive ones tend to lose less capacitance with increasing voltage. Go figure. Using the manufacturer tools is critical here. > I have yet to see 100uF MLCC. They're all over the place. You can even get 330uF 1210s from Murata. And probably bigger from someone else. These guys are popular among certain classes of design (mostly size or height constrained). Not my favorite choices though... because as SimSurfing will show you, they're sure not good for that 330uF: you get about at 240uF at 1.8V or 130uF at 3.3V. Which is still pretty great for 1210! But tantalum (AVX TAJ series, your boring basic tantalums) will give you 150-220uF in nearly the same space (1411/B case) depending on how hard you feel like pushing them. (Ceramics can beat tantalum's CV product for the small cases as there's some fixed overhead to just getting the tantalum pellet hooked up.) But then if you use MLCCs you have to deal with the underdamped PDN you just created somehow. (Which isn't hard, unless you forget about it.) > Not all designs strictly require them, though. When for example dropping 5V to 3.3V using an LDO, having all components decoupled well enough locally, you might get by without bulk capacitance. Please don't do this. (Unless you're trying to keep me employed.) Yes, you can engineer the bulk caps out of your design, and that's a viable strategy. But leave the footprints in your first spin so you can make it work, then Muntz them out after you know everything's OK. You will save everyone a lot of grief this way. > Some boards are large enough for electrolytic caps that can be combined with ceramics to get you close. This is actually one of the best overall strategies if the form factor allows it. Ceramics in parallel with tantalums or aluminum electrolytics is a fantastic combination and gives you the benefits of everything and drawbacks of... well, the electrolytics are still big.