Replacing a NIC with something hopefully high quality from Intel. Noticed that Intel uses a healthy mix of 0402, 0603, and 0805 components. Especially capacitors. Not sure if this is due to cost vs capacity, or if smaller, harder to place components are used sparingly as needed. From experience typical caps get cheaper as they get smaller, except for very large values like 10uf ceramics.
I need this card and am reticent to pull it apart to measure the values. Its really tempting though. Any thoughts?
BTW discovered WordPress apparently was installed on my Nexus 5, thus ushering the era of random China and workshop blogging. Hope you enjoyed it so far this week.
On the manufacturing floor at work, they can place 0402 but really prefer 0603. The problem of lost 0402 parts is compounded by the fact the ICT can’t detect if these components (typically bypass caps) are missing because they all connect to the power plane. We still use 0402 for things like DDR3 memory bypass caps because they have better impedance at high frequencies.
It really depends on the type of ceramic they need. Sometimes you have to use a class 1 or class 2 and you simply cannot get it at all in 0402 or 0603 at any price. For instance there are no NPO caps over 0.47 uF in any size (on digikey). I needed 1 uF for my audio project. At first I selected a polymer tantalum, but I don’t really like using tantalum even “conflict free” if it can be avoided because even “conflict free” indirectly causes harm by having a market for it.
I adopted X7R dielectrics on my second version as these are still quite good. Two issues with non-NPO are their non-linearity in capacitance and the fact they are ferroelectric and also with it often piezoelectric. These properties make them speakers and microphone which has an obvious negative coupling effect in audio. However, these effects are probably negligible and I hope to measure them. In all cases I stabilize the voltage on X7R capacitors with NPOs.
I try to adopt Al polymer caps for large sizes. Aside from cost and bulk they seem excellent.
I expect some of the larger caps on this board are large value NPO ceramics. Or very large value X7R/X5R caps. I am using some 100 uF X5R for another design to have what I think will be better failure lifetime than Ta or Al caps. I can’t think of any basic physics reasons for a ceramic to fail though I can imagine a ferroelectric degrading its permittivity over time due to crystal structure changes. I think the barium titanates are probably more stable than the PZT and PMN-PT I have worked with though.
It really depends on the type of ceramic they need. Sometimes you have to use a class 1 or class 2 and you simply cannot get it at all in 0402 or 0603 at any price. For instance there are no NPO caps over 0.47 uF in any size (on digikey). I needed 1 uF for my audio project. At first I selected a polymer tantalum, but I don’t really like using tantalum even “conflict free” if it can be avoided because even “conflict free” indirectly causes harm by having a market for it.
I adopted X7R dielectrics on my second version as these are still quite good. Two issues with non-NPO are their non-linearity in capacitance and the fact they are ferroelectric and also with it often piezoelectric. These properties make them speakers and microphone which has an obvious negative coupling effect in audio. However, these effects are probably negligible and I hope to measure them. In all cases I stabilize the voltage on X7R capacitors with NPOs.
I try to adopt Al polymer caps for large sizes. Aside from cost and bulk they seem excellent.
I expect some of the larger caps on this board are large value NPO ceramics. Or very large value X7R/X5R caps. I am using some 100 uF X5R for another design to have what I think will be better failure lifetime than Ta or Al caps. I can’t think of any basic physics reasons for a ceramic to fail though I can imagine a ferroelectric degrading its permittivity over time due to crystal structure changes. I think the barium titanates are probably more stable than the PZT and PMN-PT I have worked with though. (Ti oxide is far harder to reduce than Pb oxide)
Very simply, different size capicators have different impedances at different frequencies.
Have a read of this application note and enjoy.
http://www.intersil.com/content/dam/Intersil/documents/an13/an1325.pdf
The graphs on page 6 shows it simply.
Smaller capacitor = less parasitic inductance. Very important for bypass capacitors.
With a controlled-impedance track (usually 50ohms or 100ohms), choose a component that is the same width as the track.
With ESD protection devices smaller device = less capacitance, very important to avoid degrading signal integrity.
Most of the light brown MLCCs are 0.1uF or bigger. I have dismantled a few dead HDDs over the years to make them dead-er and has attempted to scavenge parts from them. MLCCs are easy to remove with an iron… None over 10uF, because >10uF are comparatively expensive even now. Measured their values with an old Escort capacitance meter to bin them, most are within -10%+10% from 0.1uF to 22uF.
Not much to re-use actually, but sometimes I do re-use old SMD caps only in low-value prototypes just for the kicks. On failure of MLCCs, there are many online docs from manufacturers about MLCC failure modes, but should be pretty reliable in normal use. In re-using, I have killed one out of maybe a dozen, probably desoldering-soldering overheating plus age. But for projects I use and expect to always work, I use new parts X5R/X7R only. They are cheap after all.
I would say that most of the capacitors on that board are with X5R or X7R dielectrica.The exception would be those in the crystal oscillator circuitry.
Some parameters that will affect the choice of components:
Available board space
EMC demands
– Frequency range? 100 nF is not very effective if the frequency is increased. Better to have several capacitors with equal value in parallel. At DC they will add capacitance and at higher frequencies they will have lower total impedance without several peaks.
– Different package sizes are good at different frequencies. 1206 is good as energy storage but quite bad at 100 MHz. Higher frequency => smaller package to maintain low parasitic inductance.
Operating temperature range
– Ceramic capacitors need voltage derating from 80’C and above. Murata’s MLCC catalogue has a derating chart on one of the first pages. How can AVX and Kemet, using the same kind of materials in the same kind of component, just happen to “miss” this? I have first hand experience of short circuit failures of 16V rated X5R MLCCs operating at 12V and close to 85’C. After the voltage derating was applied, the failures stopped. No other changes were made to the circuit.
– Packages larger than 1206 can be prone to cracking during thermo shock. MLCCs and resistor are good examples. 2512 resistors are not suitable for high reliability automotive electronics even in the driver’s compartment.
Sourcing
– Multiple sources will give you better backup and a possibility to deal about the price
“Better to have several capacitors with equal value in parallel. At DC they will add capacitance and at higher frequencies they will have lower total impedance without several peaks.”
Usually application notes state it is better to have different values, with the smallest value nearest the IC pins.
It depends on when those appnotes were written and also a bit on how the capacitors are selected. What was best practice a couple of years ago might not be the best now as the frequency content increases and the rise time is faster. Adding several capacitors in parallel can work both for you, and against you, depending on how they are selected. For lower frequencies it will most likely never cause any problems, but if your product will have to pass EMC tests at 100 MHz or up to 3 GHz and above…
If you look at the impedance versus frequency curve for a MLCC it will follow a V-shape. From DC up to the bottom peak of the V it will mostly behave as a capacitor but above that frequency the inductance will take over and increase the impedance. Take several, different capacitors, all with different V-peaks, and superimpose their impedance curves over each other. That resulting curve can get a sawtooth shape and those positive peaks can cause problems. Careful selection of dielectrica, package size and capacitance together with an optimized PCB layout can compensate for some of that.
This Card is a reference design for the ethernet-Chip, you can download the full schematics from their website. I think a bom is also available.
What are the tiled square patterns on the top copper layer for? They don’t look connected to anything.
Somebody had the very same question some time ago on Electronics StackExchange: http://electronics.stackexchange.com/questions/45567/whats-the-purpose-of-this-checkerboard-pattern
(and he apparently had the same or at least a very similar Intel NIC as Ian. What a coincidence… )
Doub: Those are for balancing the amount of copper on each layer. Also know as copper thieving.