adding ceramics across power pins

Leon

That doesn't answer my question. Why does the theory fail in practice? Why doesn't a capacitance add?

I know that caps have a non-capacitive impedence but surely its no that bad?

What I'd like to see is the frequency response of a tantalum cap with and without a ceramic to see how it actually works... it's nice to know that it should be done but I want to actually know how useful it is(so far for all my projects I have gotten away with just tant's... of course I don't do anything about 40mhz)

ps

All real components have some amount of inductance. The best you can do on inductance is limited by the mechanical size of the part. To have a very low impedance at very high frequencies, you need the inductance to be low and hence want a mechanically small part.

At lower frequencies, you need a lot of capacitance to make the impedance low. This favors a large mechanical size.

It is hard to make a component that is both large and small at the same time.

cs

Most of those sorts of recommendations IME are based on waving a dead chicken over a circuit and having it work. Sort of like the little boy who snapped his fingers to keep polar bears away. I've yet to see anyone present data showing supply hash or EMI before and after diking out the ceramics. That would make an interesting article.

Cheers,

Phil Hobbs

ps

In my - moderately humble - opinion, it's all about self-resonant frequencies. Tantalum's have quite high equivalent series resistances (ESR), and are perfectly useless at high frequencies, but the ESR is high enough to damp the much sharper self-resonance of a 100nF ceramic disk.

$file/TechTopics%20Vol4No5%20Sep94.PDF

A 10nF "microswave" capacitor can look like a capacitor up to even higher frequencies.

-- Bill Sloman, Nijmegen

One doesn't "have to." How much bypassing you actually need depends on the logic family, the pcb layout, and circuit details.

On multilayer boards with power and ground planes, and HC-type logic, one or even 0 caps per board will usually work. Opamps circuits are usually happy with a ceramic cap here and there.

Faster stuff, or crosstalk-sensitive stuff, may need a few more caps.

We use a scattering of 0.33 uF 0603 ceramic caps on most things. Tantalums tend to explode, and aluminum 'lytics crap out at low temperature.

Most boards have too many bypasses, often absurdly too many.

John

I know a guy who doesn't believe in using bypass caps at all, and his stuff works too.

I occasionally design SMA connectors into real pcb's to TDR/TDT the power planes and later measure actual operating noise. I conclude that few multilayer boards need more than a few ceramic caps per power pour, even if they include FPGAs and uPs and ECL. We use 2-3 ceramic caps per power voltage on big FPGAs; some people use hundreds.

John

What are the functions of bypass capacitors?

To prevent local instability To reduce coupling via supply lines between different parts of the circuitry To reduce noise To absorb static discharges Small ceramic caps are good at that. Using them might save your circuitry from transient spikes on the supply line, and when fitted on input circuitry will enable your design to (survive) pass EMC static discharge tests.

John

It's really a question of how critical the parts are, how fast the design has to work, and "How lucky do you feel?"

As a counter-example, some years ago I bought a Boca Ethernet card based on the AMD Lance/PCNet Ethernet chip (a chip which has an excellent reputation).

The card didn't work well at all. It would freeze up and start dropping certain packets - an FTP transfer of a large binary file would freeze (packets dropped with a frame check sequence error) while a TELNET session at the same time worked fine.

Turned out that Boca had ignored AMD's Appendix B recommendations about the location and placement of the bypass caps needed on the board... they had only about half the necessary number of ceramics, skipped the tantalums, and had the ceramics separated from the chip by an inch or more of PC-board trace rather than right at the pin.

As a result, the chip was suffering from severe power-supply/ground bouncing, which was rather sensitive to the specific bit patterns in the incoming packets. Transmit a packet with the right pattern of 1- and 0-bits, and the rail/ground oscillations grew severe enough to cause one bit to be mis-read... FCS error, packet rejected by the chip. When the TCP layer retransmitted the packet, the same thing would happen, indefinitely.

I ended up having to replace the board with a different brand... Boca never managed to correct its mis-design.

As to why one tends to use both ceramics and tantalums on the same power supply trace... as I understand it, it's due to the fact that all components have unwanted parasitic characteristics.

The ceramic caps are usually fairly high in Q (low series resistance), and if mounted close to the chip with neglibly-short leads there's a low parasitic inductance involved. This makes them good for bypassing away short-term (high-frequency) noise and current-drain changes.

However, their capacitance and thus their energy storage is limited. If the chip's current demand increases sharply, the ceramic 100n (or whatever) cap can't supply enough stored energy to hold the voltage stable for very long. Hence, you need another (larger) cap nearby to help take up the load.

There's another disadvantage to the ceramic, too - its high Q can actually cause destabilization of the voltage supply rails. If all you have on your Vcc trace is high-Q ceramics, then the capacitance of these ceramics can interact with the inductance of the PC board traces, creating an LC resonant circuit. This circuit can ring (badly) if excited by load fluctuations or noise near its resonant frequency, resulting in worse voltage fluctuations than if the cap wasn't there.

For this reason, it is helpful to add some additional "bulk" capacitance to the power supply wiring/traces - higher capacitance, in series with a modest amount of resistance (the inherent resistance of a tantalum or aluminum electrolytic, for example). This bulk-bypass capacitor, with its relatively low Q, has the effect of "swamping out" the unwanted resonances of the parasitic LC circuit, and thus keeps the voltage stable and prevents the rise in noise (and lack of stability) at the parasitic LC resonance frequencies.

Why not use just the bigger tantalum or aluminum caps? Well, they're big and thus hard to place close to the IC pins, they're more expensive, and by themselves they may not provide adequate bypassing at the highest frequencies.

So, designers tend to use a hybrid approach... ceramics as close as practical to the pins, and bulk 'lytics (tantalum or aluminum) scattered around the board.

--
Dave Platt                                    AE6EO
Friends of Jade Warrior home page:  http://www.radagast.org/jade-warrior
  I do _not_ wish to receive unsolicited commercial email, and I will
     boycott any company which has the gall to send me such ads!

Aluminums basically open up at -10C or so. Tantalums have a tendency to detonate when used on supply rails. Polymer alums are about the only way to get hundreds of uF on a bus without those problems.

But if power distribution is via copper pours adjacent to a ground plane, a scattering of 0.33 uF 0603 ceramics is fine; ignore any theoretical whining about resonances. If chips have low-frequency current steps, and the local voltage regs and ceramic caps can't handle that case, you will need more bulk C. But often a voltage reg + several uF of ceramics will be pretty stiff wideband (if it doesn't oscillate!)

We commonly use 22 uF 1206 ceramics; a couple or few of them can usually eliminate electrolytics.

John

oy

=A0 =A0 =A0 =A0 =A0 =A0 =A0 AE6EO

r

I think it would be difficult to get significant inductance on the VCC line unless you did poorly designed two layer PCB.

r caps

. It

n

ramics

y

Typically applications engineering has the final say on what goes in a datasheet. The more capacitance they specify, the less phone calls or email they get.

A case in point would be LDOs. All that capacitance the apps guys like to slap on the input really doesn't make the part work much better. Capacitors on the output is another story. The overkill on bypassing the input to LDOs goes back to poorly designed ICs where the reference might act funky if the input voltage wiggled. Later generation LDOs aren't that finicky, well assuming the design is good.

Much depends on the application. For example:

Example [1]:

When I designed a ground support test rig for military cargo aircraft actuators, I was faced with a project that had no weight constraints (the fixture already weighed several tons) no size constraints (I had a full-height 19 inch rack for the electronucs and the bought gear used up less than half of it) no power constraints 15A at 120V, and I could have gotten 100A

3-phase if I had asked for it) no real per-unit cost constraints (we made up 10 copies of each board, built and tested three boxes, and put one into use with the rest stored as spares).

The real constraints were minimum downtime, repeatable results, getting the fixture into service as fast as posible, and ease of use.

For that application I put a ceramic cap on each device as close to the power pins as possible, used power and ground planes, and sprinkled handful of tantalum caps about. I could have gotten by with far fewer, but I didn't care.

Example [2]: When I designed electronic toys for a major US toy manufacturer, the top three priorities were low unit cost, low unit cost, and low unit cost. At a production rate of over 100,000 units per hour for that one toy, a one cent reduction in unit cost equals a pure profit of $1000 per hour / $168,000 per week. That one got zero bypass capacitors.

I tested the pilot run (a thousand toys) for any loss of function and I crawled through the electronics of three samples with a scope and DMM looking for any problem areas that might require a bypass cap, and would have added one if needed. The microcontrollers we use in toys nend to be designed to work well with just a battery supply and no bypass caps.

--
Guy Macon

caps

ceramics

Everybody thinks their part is the center of the universe.

Most regulators, lin and switchmode, are picky about output cap ESR, and many specify their requirements poorly or not at all. The all-ceramics movement is new (ie, 22-100 uF ceramics are now affordable) but lots of old data sheets haven't caught up. It's bad for jitter performance when your FPGA's core supply is being driven by a relaxation oscillator.

John

boy

Tendency? The only tantalums I have seen explode are those that have been abused - reverse voltage or over-voltage.

They do make a pretty show when htey go, though. I have ssen one take out three boards in a dense rack, and another send a program manager heading for the lab door.

e:

ger caps

ce. It

can

ceramics

boy

I'm not a power supply expert, so I just obey the app notes.

Thank you for not bothering to specify the type of circuit. That allows me to supply a vague and generalized answer targeting my area of expertise (RF).

The reasons are frequency, ESR, cost, and whether you're trying to prevent noise from entering or exiting the device being bypassed.

Frequency effects include self-resonance and changes in impedance (ESR) over the operating frequency range. You usually want the device power supply "pins" to represent the lowest possible impedance across the entire operating frequency range. If the power supply were insufficiently decoupled (bypassed) and presents a finite impedance to the device being decoupled, then that impedance could easily become part of a tuned circuit, tank circuit, load, or whatever. In extreme cases, this uncontrolled impedance could produce instabilities and oscillation. Bypassing each device individually also prevents coupling between devices through the power supply leads.

Devices that must operated at both low frequencies and RF frequencies need to be bypassed at all frequencies in their operating range. Each capacitor family and range of available values are self resonant at some frequency. The idea is to select a range of bypasses that are low impedance across the entire frequency range. Super-caps for very low frequencies. Tantalum and aluminum electrolytics for Why don't they make special "power" caps that combine tantalum and ceramics

The low cost of capacitors is due to the huge quantities that are produced. Whatever conglomeration of parts might be used to combine such parts, the number of possible combinations increases by the product of available values of tanatalums and ceramics. Since the demand will be less, so will the quantities produced. That makes the conglomeration considerably more expensive.

Light reading:

--
Jeff Liebermann     jeffl@cruzio.com
150 Felker St #D    http://www.LearnByDestroying.com
Santa Cruz CA 95060 http://802.11junk.com
Skype: JeffLiebermann     AE6KS    831-336-2558

They blow up erratically when on power busses that have moderate or high dV/dT which pushes high peak currents into the caps. Tantalum is the fuel and MnO2 is the oxidizer, and some tiny granule, pumped by a current spike, is the detonator. We did an RMA analysis and found that half our field failures were tantalums, so we avoid them now. Some companies don't allow MnO2 tants at all.

See section 5. We derate voltage 3:1 when we really need a tantalum on a power buss.

Yup, there's an exothermic chemical reaction, not just a short.

Wet-slug and polymer tants are OK.

John

I spent a lot of time trying to figure out the real root cause of our persistently-exploding smt tantalums. We used a 2X voltage derating.

At first, it appeared that it was a dV/dt issue (i.e. high inrush and/or ripple current). This indeed may be a factor, but not in our case.

After proving that our dV/dt (and associated inrush and ripple current) was WELL under the spec, more research showed that it may be due to absorbed moisture and subsequent damage due to high dT/dt during smt reflow. So, we asked AVX if they had moisture-resistant packaging (similar to what is commonly done with ICs). It turns out that AVX already offered this option but they told us that this option was rarely purchased by their customers. We started using them -- even though they were hard to get. Low moisture tantalums didn't seem to help, however.

More research seemed to suggest that it is merely the nature of high density smt tantalums. The higher the capacitance and the smaller the package, the more likely it was that they would explode violently.

So, what did I learn from months of trouble and investigation? DON'T USE HIGH DENSITY SMT TANTALUMS.

Bob

--
== All google group posts are automatically deleted due to spam ==