Re-reading the thread, I'd have to agree with Ken's comment--making an analog setup stable on a wafer prober can take some doing. One thing you can do to test for oscillation is wind a few turns of magnet wire (maybe 1cm diameter) and solder it to a connector so you can plumb it into the input of a spectrum analyzer. Use this as a sniffer to check your circuit for oscillation without having to probe it directly (you often stop the oscillation due to probe loading).

Oscillation caused by probe parasitics is sounding lots more likely than burst or popcorn noise.

You could also try packaging a bunch of chips and sorting for yield in packaged form. If it is the probe parasitics causing problems, it could take a lot of time on the machine to fix.

Steve

sos wrote on Nov 29^th, 2009, 3:41pm:


That's the coil that's 1cm in diameter, not the wire...

aaron_do wrote on Nov 29^th, 2009, 5:04pm:


I can tell you up front that if you hook a network analyzer up to all that wiring on the prober, you will get weird results. What do they mean--who knows? I favor a different approach. I wasn't kidding about the tweezers. Hold the tweezers and lean the same hand on some grounded surface. If you touch the supply pins and get changes in your noise output, this is pretty good evidence that your supply bypassing is suspect (maybe not 100% of the problem but significant).

The frequency where bypass capacitors become ineffective is pretty much where they go self resonant (or resonant with the external lead length on the prober). Electrolytic caps have higher internal inductance than tantalums. We typically use 4-10uF tantalums plus 0.1uF ceramic or film caps in parallel, on each supply to ground.  The simple minded idea is that when the big cap quits working due to resonance, the little cap takes over (which has a much higher self-resonance), but this is a simplification. Actually, the resonance on the big cap is still there and can cause problems in some cases. I'd go with tantalum plus 0.1uF in parallel, as close as you can get to the chip (I realize there are probe pins in the way). Keep lead lengths (supply and ground side) as short as possible. You can't get rid of the probe card, but no meandering traces!

If you still have problems, sometimes you can add resistors to damp out resonance. One place is to add a series R-C between the supply pin and ground--like maybe 25 Ohms in series with 0.1uF. You could also try supply to supply if you have split supplies (ground on your board may not have any relation to what the chip is seeing as ground). The other place you could put resistors is in series with probe pins. There are obvious problems with this idea--you're very limited how much resistance you could put in a current carrying line, like the supply. You can put them in series with high impedance pins like inputs, but be careful you don't build a filter and ruin what you're trying to measure.

aaron_do wrote on Nov 30^th, 2009, 4:44pm:


The tweezers do *not* short anything, sounds like you have the wrong idea. All I do with the tweezers is add parasitics to the  supply. The parasitics consist of me--or my hand, depending what I'm leaning on. The tweezers only contact the circuit at a single point--you could use a piece of stiff wire instead, but tweezers have a nice grip. The only purpose is a quick check to see what nodes are sensitive to parasitic loading. Obviously, you don't want to do this when there is enough voltage around to be exciting.

Just to be clear--no short circuit or bridging is allowed. The tweezers are just a means to connect your hand to the circuit as a quick and dirty, mostly medium impedance R, ~C load. Signal nodes are clearly going to be pretty sensitive above 1MHz, but the other nodes should not be.

Steve

Looking at the spectral content I do agree with the group that it is NOT a transistor noise problem, thats for sure.

Welcome to the "noise onion" problem. But at the lower frequencies, it may not be an EMI type coupling issue.

This could be either a low level oscillation within the system, or something coupling in from the outside.

Key thing is you need to selectively divide things up, separate and isolate things until you determine the cause.

First - If it is on a probe station, get the part diced and packaged. Getting things quiet and under control on a probe station is not a realistic place to test things. Get it soldered down onto a test PCB (no socket!) that is well decoupled and well filtered. Right there you may find that solves the problem.

Probe stations are a nightmare of cross coupling, random inductance and other issues not easily controlled. Use them for DC measurements only, and even there, be cautious about resistive contacts and other things.

Once you are on a PCB with a packaged part, get everything going into the part quiet and selectively filtered. If that doesn't isolate things, then its time to put it inside a metal box.

Let us know what happens!
good luck
- jerry

rf-design wrote on Dec 14^th, 2009, 7:41pm:


I can't say one way or another, but I do know if you have a lot of base resistance 1/f noise will ruin your day. High sheet poly also gets pretty 1/ffy. But I think Aaron was more concerned with the peaks than the low freq noise.
rg