My continuing saga with power supply bypassing

(snip)

(snip)

If his advice includes going with the cheapest, then it includes crappy dielectrics like Z5U and Y5V and horrible tolerances like +20% -80%. So just pick the nearest value that is available.

One of the things any real world engineer knows (but text book ones often do not) is there is a very big difference between what is in the data sheets and what you can actually get. This is ESPECIALLY true for caps.

Sounds like your boy doesn't get out into the real world enough. Esp since they sent you to Digikey and not Arrow, TTI, Avnet, or some such. I spend more time in selecting parts in seeing if they are actually available then finding the best part for the job.

Anyway .018uF does exist for 0201s but in Y5V dialectic and if you are having bypass issues i am sure you don't want Y5V. Does your boy understand dialectic types or has he been behind a text book and into Math-CAD and Spice Simulations to long to know any real world realities?

What are you tring to do? Is the bypass problem bypassing what is coming from the ASIC back to the power or the voltage going to the ASIC? If it is to the ASIC and you are having to do this then you are not fixing the problem, only masking a power supply design or PCB layout issue. The problem needs to be fixed there. Adding a pile of inductance from a pile of caps isn't going to really fix the issue. My guess is you have a PCB layout issue, not a cap issue.

Hawker

On 4/27/2007 12:12 PM, The digits of snipped-for-privacy@netzero.com's hands composed the following:

I can't imagine the brand-name capacitors from Digikey are any different than those from the big distributors... and for prototyping purposes (small quantity orders), Digikey beats the big distributors hands-down in parts availability and search engine power.

We told you to fire this guy, and you didn't.

John

LOL, John. Look, I like this guy. I just honestly don't understand what he's cooking up. People are allowed to make mistakes, you know? But I honestly freaked out when I found out the capacitor values don't even exist. Looks like he just plugged in a "pure" capacitance in Spice, plugged in the partial inductance of a 0201 footprint, and ran with that. I am asking stuff here first so when I do ask him what's cooking, I'll be in a better position to evaluate the response, you know? I'm also genuinely curious how he expects us to measure -70dBm power levels on power planes, we don't have a spectrum analyzer, and how do you adapt the impedances to get a good measure of the power???

Am I wrong in thinking that a real capacitor on a board has a partial inductance associated with the way it's built (including the footprint size and the construction of the dielectric), a partial inductance associated with the layout ( vias close? vias far?) and that the eventual resonant frequency will be determined by an L that is difficult to figure out even in theory since I don't really know how the power is routed to the actual chip?

I mean the complete circuit the cap is in includes vias, planes, traces, solder joints, ultrasonically bonded wires, silicon and back again as the circuit is completed. Am I right?

On 4/27/2007 2:24 PM, The digits of Joel Kolstad's hands composed the following:

True but if you have a relationship with the big disties you tend to use them first and then Digikey only if the big boys are out of what you want. I don't pay for small quantities of small parts. I just tell Future, Avenet etc what I am designing in and they sample me whatever I need. I assume most folks have similar relationships if they are doing any production.

Take TDK P/N C0603X5R0J223K, in stock at Digikey, 15670 of them, just a few pennies a pop. 0603 in there seems to be the metric size (arrrgh!) which would be 0201 in "real life". 0.018uF would be a boutique part, I wouldn't use that. Purchasing wouldn't be happy. And they must be kept happy or they won't let you dig into the Christmas bonbons the distributors bring around in December ;-)

The medical folks would politely say "that is sub-optimal". When I design stuff, like right now, there is always a 2nd PC running with the Digikey site on the browser. If it ain't in stock it ain't going into the circuit. And with stock I mean some sizeable quantity. It can still get you, as it did here last weekend. Saturday there were thousands of a particular ferrite choke. Monday the client took my list and whoopdidou, only 30 left and that wasn't enough. So it can happen.

Who's Chtulu? Anyhow, staggering caps is a good idea but with SMT you don't have to go to extremes anymore. Something like a 0.1uF, a 4700pF and maybe a 100pF should take care of the higher frequency scrubbing. Sprinkle a few 1uF across the planes (hoping it's planes...) plus here and there a nice low ESR 100uF should really keep things quiet.

Watch them aphids. They got ours almost every year.

On 4/27/2007 4:07 PM, The digits of snipped-for-privacy@netzero.com's hands composed the following:

Exactly. My experience is the PCB inductance alone swamps the cap inductance, never mind the rest of the sources you mention. This is why stacking them up is dangorous. The way you have to "break out" to lay out that many caps can introduce tons of capacitance.

We have not even begun to talk about PCB stackup issues. The stackup and pre-preg/laminate thickness can make a big difference as well. In fact if he has not asked for a stackup and looked at that than that is a big clue as to if he is overly theory based and not going to get you home or not.

I have no experience in such high-speed designs, but you might present your questions to the people in one of the PCB design forums. I use PADS, and they have a very active forum at

Paul

Best to have an honest talk, along the lines that he should only spec in what Digikey has in stock. Or Mouser.

You need to either rent one, buy a used one on EBay or if 1GHz is high enough go for the little Hameg which is around $2k.

Pretty much.

On 4/27/2007 4:53 PM, The digits of Paul E. Schoen's hands composed the following:

I thought I recognized your name from there. Yes that would be right up that groups ally. At least they would look at the layout issues before adding caps. And way better discussion topic than Mentor Licensing bashing that is going on now.

Hawker

Actually, surface mount components can have some very low inductance. Resistors are nearly ideal. Multilayer caps are pretty good. I recall Johansen made some really good caps, but expensive.

Rather than hacking, if this guy wants results, have him rent you a network analyzer and component test set. The test set comes with calibration standards. While you are at it, make him spring for a test PCB that you can measure with the network analyzer as well. Set up the board with SMA mounts.

This is not the kind of thing you can do on the cheap. What really matters here is the cost of failure. I did a video DAC project where the lead inductance really needed to be known. I had a pure math guestimate, but project development costs are easily a few hundred thousand dollars. I used what known loosely as a jumper chip, really just silicon, metal and a vapox opening, i.e. a big ass bonding pad, and routed bonding wire to the spot, then measure everything in a test board.

Don't forget to test the caps under bias, obviously the supply voltage of the circuit in question.

I don't think this is true? I remember reading that only the area matters and not the thickness? Maybe not talking about the same thing but I was reading over some adc specs and it mentioned something like this.

On Apr 27, 4:15 pm, Hawker wrote: I don't pay for small quantities of small parts. I just tell

no, we don't accept free samples for the same reason government employees aren't allowed to accept free handouts, reciprocation is expected and usually given even when all the signs point to somewhere else, in the long run you lose.

Of course I always give away free samples of my products for the opposite effect :)

I've been guilty of some of that bashing, but much of it is deserved. Their product is very good, and their support staff excellent, but they have had a record of breaking previous functionality while adding frills, leaving bugs unfixed, and then penalizing their customers.

But the forum often provides much good advice and information. One good resource person who has helped me immensely is Robert Kondner, who has done some solid research on high speed issues, such as

Another good resource is the

Paul

On 4/27/2007 5:34 PM, The digits of Jon Slaughter's hands composed the following:

The thickness affects the impedance (as does the area) and hence that is part of the equation. Esp for ringing. In theory it also affects the capacitance, but you need fancy dialectics or a huge PCB before it does much.

On 4/27/2007 5:48 PM, The digits of Paul E. Schoen's hands composed the following:

I'm with you. I have issues to with where they are going, but think it is still the best product out there for most uses when considering all the compromises.

I was a PADS user from the PADS Work/Perform/Dos days. Had dongle #508 I think. True Type fonts, the new UI - give me a break fix the old issues first. It just that a few posters (we all know who they are) chose to bash Mentor or talk about the license issues as a reply to every posters question. It's the one track mind stuff that I have issues with.

Poor Dave getting in the middle of all that mess.

Hawker

Exactly, The power and ground plane is your first line of defense. Their spacing, size, and distance from the ICs is critical. Any "tuning" which is BS IMHO, can only be done with this information known. Select a single value bypass cap, 0.10uF is a good value, with as small an aspect ratio (L/W) as possible. A SM0805 is poor with an ratio of 8/5=1.6. There are many with a ratio of 0.50 such as the SM0204. I'm going with JL, send him home! Harry

Not the same thing at all. The cost of a purchase order can far outweigh the cost of some parts. The time it takes getting all the signatures decreases the amount of time you can spend on a project.

When you order free samples, you are defining what you want, some bloke in the shipping department sticks it in a bag, and UPS brings it to your door.

This saves everyone a great deal of time, plus the cost of a regular PO. Nobody ever checks to see if you are putting the parts in production.

Gifts from salesmen are completely different. They choose the gift based on their knowledge of you and your personal likes and dislikes. They fully expect you to reciprocate down the line by buying something from them, and they make sure you are aware of that even though they may deny it. Refusing gifts from salesmen is the only ethical choice.

Free samples from vendors speed the development process and reduce the paperwork cost for everyone. It is anonymous, everyone benefits, and there is no commitment implied.

Regards,

Mike Monett