X5R vs X7R MLCC

I'd guess that a majority of EEs, and most recent grads, are unaware of the nonlinearity of ceramic caps. I have to teach them about it.

Even without DC bias, the nonlinearity can cause distortion in signal circuits, filters and tanks and such.

What do you mean by salvaged parts?

I use 1 uF 50V 0805 caps by default, so we can buy them in bulk and always have a reel loaded on the pick-and-place.

It's an interesting problem to pick the cap in stock that gives the most C at a given voltage. But too much work to actually optimize, usually.

yes - surely it's the sort of thing degree courses should be teaching.

Yep. There must be some good use for this property, I've not thought of one yet. Ceramic cap capacity loss has been discussed a lot on here though, I'd be surprised if any regular contributor didn't know of it.

NT

was that when you can't, the options are especially wide for decoupling ca ps. Even more so if you've added a few extra pads so aren't relying on any one cap to provide enough capacity every time. I use/specify salvaged parts routinely to keep BOM rock bottom.

You know you shouldn't ask that. :) Re-used parts, ex-scrap. Not your cup o f tea, but that's where I operate. When specs can be vague at best it's the ultimate way to minimise BOM. They're big business in China, key to keepin g consumer product costs down.

EE time isn't cheap.

NT

I think EE students should concentrate on theory, Signal and Systems and things like that. Thet's harder to learn in the field.

But a few lectures about real parts would be good too.

Somebody sells a varicap-like part that's used to tune RF circuits, but based on nonlinear ceramics. I recall that it's 3-terminal, ground/rf/dc tune. But the bad (good for this case) dielectrics mostly come in higher values.

I think one could build a parametric power amp based on ceramic caps.

You could build a tunable oscillator or filter with bad caps.

I have seen NLTLs, shock lines, that used discrete ceramic caps, or a slab of the material, as the nonlinear element.

I bought a reel of 3.3 pF NTC caps, custom brewed. I ordered N4700 and they were actually about N5600. I wonder if the high NTC is associated with a high voltage coefficient; I could measure one.

Are they tested? I'd worry about delams and such on recycled parts. Troubleshooting is so expensive that we are considering just throwing away some boards that fail in test.

Right, it's not worth an hour of engineering for us to save a few cents on a board that will be built in the hundreds.

There is a widespread theory (one that I don't subscribe to) that degreed engineers should be taught more theory and not so much practical. The idea that college shouldn't be a trade school. I'm glad that my school had a reputation more on the "trade school" side, though. Opened doors my whole career.

VCO?

There is room for a mix. EEs who don't know which end of a soldering iron to pick up are handicapped their entire career. They never catch up. Lectures for theory and labs for implementation (practical) should work. Labs seem to be a thing of the past, though.

Hands-on

No advantage for us, though I do use multiple parts when needed. If there is some free space in the area, I'll add an extra pad or two, so I can increase the capacitance, if needed.

Optimize, no, but I know which ones in our usual list are better than the others and cheaper. I rarely use anything above 10uF, and mostly below 4.7uF.

Yikes! How do you get quantity? Reels? Rework isn't cheap.

Laptops, especially when purchased by the students, are cheaper.

I think I have always over-bypassed things, especially digital logic. So I keep using fewer. I suppose some day I will get into trouble.

FPGAs and uPs seem to have substantial on-chip bypassing these days. And multilayer board planes are the first level of bypassing. If you TDR a power plane, it typically looks like a few nF way up into the GHz, and if you add caps anywhere on the plane, it just looks like a bigger cap. Most of the advice online is overkill or wrong. The idea of staggering values to spread SRFs seems to be silly hearsay.

I thought that I'd under-done it last week. An LPC3250 ARM chip on a new board was intermittently clocking at half the expected rate, very erratic on temperature and touching the board and all sorts of things. I had only two caps bypassing the 1.2V core and two on 3.3V i/o, none super close, so with great difficulty we kluged in more, on vias flip side of the BGA. Didn't help. Playing with the various numerators and denominators in the clock PLL fixed it... don't know why.

We now have some 47uF 6.3V ceramic caps in stock. They are good at the outputs of switchers making low voltages, around 1 volt, for uP and FPGA core voltages.

That's why (LT or other) SPICE is so universally believed. Nevermind that the board-level models aren't worth squat. OTOH, nobody actually does board-level designs anymore, right?

I was thinking more of power supply input and output caps.

Have you looked at the planes with a network analyzer? I generally add a few caps around such parts with an SRF close to the part's clock. Another good reason to use lotsa caps is because that's what those in the design reviews want to see. I certainly don't use as much on SOCs as the manufacturers recommend, though. The passives markets are really nutty now, and will be for a couple of years, so using more than necessary is double-dumb.

Even if you go cheap on decoupling, leave the pads. It's a lot easier than kludging stuff and costs nothing.

Those large caps have a horrific CV curve. If I need that much capacitance I use Aluminums (or more lately hybrids). I use ceramics to cover my ripple calculations (derated for voltage) and electrolytics for load regulation if the needs get above 10uF, or so. That's one of the nice things about the high switching frequency regulators - the passives get small and cheap.

test:

uld

he X7R

down only

e rating

ied

sins,

voltage

by C(V)

s, if they're ceramic, have enough capacity, which can be under 1/10th rate d, are adequately voltage rated & not unsealed MLCCs - once those are satis fied, what's the problem? One really could use salvaged caps of any unknown ceramic dielectric & be ok. (I'm not suggesting actually doing so.)

re

Only some ceramic caps are significantly non-linear. It depends on the diel ectric. COG caps are fine, but only offer relatively low capacitance.

It's usually only a problem large value capacitors - X5R and X7R are notori ous for it. It's like charge soak in plastic film capacitors.

You have to know quite a bit to know that you can need to worry about it.

"Can" isn't all that useful.

are

her

Perhaps. But there's a lot of other basic knowledge that first degree cours e need to impart. My experience was that my undergraduate chemistry course never taught me enough to let me carry anything but the simplest procedure. As soon as I wanted to do anything even slightly demanding I had to get ba ck to the library and read up about it - and my undergraduate course had tr ained me to do that.

ne yet.

Parametric amplifiers exploit it

I looked into using the idea once, and decided that it wasn't going to be w orth the trouble in that particular application.

surprised if any regular contributor didn't know of it.

John Larkin doesn't seem to hire the sort of people who would contribute he re, perhaps because the people who contribute here would be aware of his ne ed for constant flattery.

I don't have a VNA, or much sine wave gear in general. I've done a lot of TDR analysis of PCBs, traces and planes.

I generally

Once the pads are on the board, it's essentially free to shoot down all the caps. So why not?

We have put bypass cap pads on the bottom of the board for insurance, and not stuffed them. Stuffing would require a second pass through the line.

The CV curve isn't scary at 1 volt. The small size, low cost, and low ESR/ESL of the ceramics are nice.

Aluminum polymers are good at switcher outputs at higher voltages.

Anybody who has found their decoupling network to be resonant at some frequency has a different take on that. Big capacitors typically have enough series resistance to damp the kind of resonances you can get with lead inductance.

The three capacitor decoupling networks that Burr-Brown prescribed for their faster op amps always struck me as using the series resistance of the tantalum component purely as a damping resistor.

It would be interesting to see the same board in both the frequency and time domains.

At least the larger capacitors aren't free and they're getting to be a PITA to get. We've been told that we're simply not going to get some devices. Others have lead-times of over a year.

You don't already have stuff there? No BGAs? Our reverse side is usually as packed as the top.

That's great for ripple but doesn't matter so much for bulk capacitance.

Aluminums don't like ripple.