I would like to propose language to discuss separately two different SSN effects, only one of which is truly ground bounce. In a system using CMOS gates you can look at the rising edge of a signal separately from the falling
edge. When the gates switch from LO to HI, you have a current path (as those gates charge their loads) that is different from the current path when those same gates switch from HI to LO and those gates drain their loads. I
refer to the LO to HI swich case (which draws current from the supply) as Supply Sag. It is the other case - HI to LO switching - which causes ground bounce. The reason to evaluate these separately is that you would usually
look at different physical locations to fix each.
While both can be called SSN, you measure Supply Sag differently from Ground Bounce. Supply sag can be measured between the power and ground pins of whichever IC is concerning you most. Ground bounce is much more difficult and
can only be obtained indirectly. One technique would be drive one gate to LO and then measure the output voltage of that gate while it is trying to hold a LO as the rest of the bus/gates switch from HI to LO. This is why we
care about ground bounce as that output voltage is (or can cause) false switching. False switching due to Supply sag can also be obtained indirectly, but the gate should be trying to hold a HI. And Supply Sag can also cause
other ill effects that ground bounce can not. <begin tongue in cheek mode> Say, could we use Supply Sag to slow down the gates and fix Roy's tooo fast gates problem? <end tongue in cheek>
But getting back to the original thread of Noise voltages VS EMI, I will use this comparison:
1) While a Square is always a Rectangle, a Rectangle is not always a Square.
2) While your Ground bounce and Supply sag always cause a least some EMI, your EMI is not always caused by those two SSN effects.
In other words, there are other sources of Noise Voltages which can cause EMI as well as current loops which radiate magnetic fields and cause EMI (without much source voltage). So I would not try to correlate Noise Voltages to
EMI levels. I would miss the EMI due to current loop sources.
BTW, I would like to add to Larry's comments about the "loop area between the hot side of the probe and the ground connection" with the comment that the measurement bandwidth of that measurement is inversely proportional to the
size of the loop area. Keep it small!
Hope this helps,
Larry Smith wrote:
> Ron - I think you are saying that you can make a measurement by
> touching the hot side of the probe to a node and just not worry about
> the ground side of the probe. You end up making a measurement of that
> node with respect to some "electrical mass".
> I would agree with you to some extent and have used the same technique
> myself. It is good enough to find out if a signal is alive and
> approximately what it is doing. Depending on where the "electrical
> mass" is, it might even be an indication of what that node is doing WRT
> the center of the earth or spice node 0. But the measurement is not
> very well controlled. The FET in the high impedance probe will measure
> a voltage WRT it's local circuit ground. The local circuit ground is
> not connected to the circuit that you are trying to measure but is
> heaviliy influenced by capacitance to your hand, surrounding metal and
> a highly inductive path back down the cable that attaches the probe to
> the measurement instrument.
> >From an SI perspective, I do not believe the 'groundless' measurement
> is a good one. There is simply too many things that you miss. Local
> receivers measure an incoming signal WRT local Vdd and local Gnd. A
> single ended high bandwidth probe does a very good job of measureing
> exaclty what the receiver sees WRT local ground, if the connections are
> short. It is a little more difficult to obtain the incoming signal WRT
> to Vdd, but it can be done. With the probe ground present, you can see
> detail in the signal that you will not see with the ground absent. On
> a bidirectional net, it is possible to tell when the circuit is driving
> and when it is receiving by the precise signature of the waveform.
> Let me give another pertinant example. One time I wanted to know the
> voltage on the Vdd plane at the edge of the board and at the center of
> the board. We were using a spectrum analyzer and were interested in
> noise at a frequency that was near the half wave resonance of the
> board, 700 MHz. It was an EMI problem frequency. At first, the
> measurements were taken with the probe not grounded. All points on the
> board were about the same. Then we went back and made the same
> measurements with the same single ended probe, but this time connected
> the ground side of the probe to the ground plane and the hot side of
> the probe to the Vdd plane at unpopulated decoupling capacitor sites.
> The difference was astounding. The ends of the board had high voltage
> at that frequency but the center of the board was quiet. This is
> exactly what you would expect from a half wave resonator with open
> ends. The ends have maximum voltage and the center has maximum current.
> It confirmed our thinking on decoupling capacitor placement. We
> would not have gotten that information if we had not used a tight
> connection to circuit ground.
> In the SI world, I think it is important to specifically measure every
> signal WRT something else. That is the signal that circuits respond
> to. When I speak of ground bounce in an SSN problem, it is always
> ground bounce WRT another local node.
> The 'groundless' probe measurement may have meaning in the EMI world
> but my feeling is that the measurement is a little out of control. It
> may not be repeatable or consistent with measurements in other
> Larry Smith
> Sun Microsystems
> > Ron Miller wrote:
> > Hi Larry
> > I like your discussion on ground bounce, and I used to think the same
> > way you do. However, over the many years I have modified my origional
> > thoughts.
> > TIME REFERENCE
> > Ground bounce may also be considered with your reference not being
> > another physical point but with itself versus time. At first this seems
> > frivilous because it seems impossible to measure.
> > However, If you ever used an RF high impedance probe you may have
> > noticed that the ground wire can often be removed and the signal is
> > still there. I have seen this many times with spectrum analyzers when
> > looking into the gigahertz range.
> > PRACTICAL GROUND RETURN
> > In this setup the ground return need only be an electrical mass with
> > storage capacity and lossiness, as in your hand and body, or it could be
> > a chassis or cover above the board being measured. The ground path
> > through the chassis is unimportant, but the stored charge in the cover is
> > relatively stable compared to the trace on the board. The charge is
> > averaged by the capacitance over that area and as long as the probe
> > does not draw appreciable current, it does not disturb the chassis or body
> > ground reference.
> > So, a differential fet probe might seem to be best with one input connected
> > to the local chassis. However, the ground on a single ended fet probe does not
> > load the reference because the wavelength gets so short that and the fet
> > ground path has no effect. So, a differential probe is unnecessary, and since
> > single ended probes have a higher cutoff frequency they are preferrable.
> > PRACTICAL MEASUREMENTS:
> > So, when you float the probe with your hand on the ground wire, the bounce you
> > measure in either the ground or the power is real. Because you are also picking
> > up lots of other interference from florescent lights and such you need to vary
> > the sweep rate and triggering of your scope to minimize interference and maximize
> > the ground bounce. An external trigger and/or a sampling scope may be needed.
> > Ron Miller
> > *****************************************************************
> > Larry Smith wrote:
> > > This is a very important and timely discussion. I think it would help
> > > if we carefully defined our terms. We are certainly familiar the the
> > > term "ground bounce" but those words mean a lot of different things to
> > > a lot of different people.
> > >
> > > Voltages are always with respect to something else. To me, ground
> > > bounce means the voltage on a power (Vdd) node with respect to a nearby
> > > ground node. In modern systems, it can be measured using a single
> > > ended probe connected vertically between the Vdd plane and Gnd plane
> > > using _very_ short leads. The loop area between the hot side of the
> > > probe and the ground connection should be less than 1/16 of a square
> > > inch, otherwise you will be measuring the magnetic flux in your probe
> > > loop instead of the noise between Vdd and Gnd. Ground bounce and Vdd
> > > bounce cannot be sepparated because they are referenced to each other.
> > >
> > > I do not know of any way to directly measure the voltage across a
> > > ground plane. It is a valid thoretical concept because the common mode
> > > voltage across the ground plane can radiate, creating EMI noise. But
> > > there is no way to measure it using a scope or spectrum analyzer
> > > probe. There is a time delay across any significant distance (inches
> > > or cm). What does it mean to measure a voltage across time? The noise
> > > you measure will actually be from magnetic flux penetrating the
> > > loop area of the probe. The only valid measuremnt technique that I
> > > know of involves some kind of antenna. An antenna can sense the
> > > magnetic or electric field that propagates from the plane and turn it
> > > into a measureable voltage. This is exactly what is done in an EMI lab.
> > >
> > > From an SI perspective, the only thing that is important is Vdd-Gnd.
> > > That is the voltage that all circuits respond to. There probably is a
> > > noise voltage between the Gnd plane and the center of the earth, or
> > > from the Vdd plane to the center of the earth, but I know of no way to
> > > measure it, without an antenna. In spice, we report the Vdd voltage
> > > and Gnd voltage with respect to spice node 0, but these voltages are
> > > not important to circuits, only local Vdd-Gnd is important. Spice node
> > > 0 is a lot like the center of the earth. It is very difficult to model
> > > the impedances from our circuit to the center of the earth, and it is
> > > very rarely done. In SI analysis, we commonly connect spice node 0 to
> > > some "Gnd" node, but this is only for convenience. You can dump a lot
> > > of current into spice node 0 or into the center of the earth and see
> > > no change in voltage. There is no node in our system where we
> > > can do this.
> > >
> > > When an EMI engineer talks about ground bounce, many times he is
> > > discussing the voltage on a ground node with respect to the center of
> > > the earth. Measurements and models for this voltage are important for
> > > EMI purposes, but are beyond the scope of what an SI engineer usually
> > > does. When we discuss EMI and SI topics, it is very important to
> > > define our terms.
> > >
> > > I hope this helps the discussion. This is a very important topic for
> > > those of us who are trying to make circuits that both work and comply
> > > to EMI standards at the same time.
> > >
> > > regards,
> > > Larry Smith
> > > Sun Microsystems
> **** To unsubscribe from si-list: send e-mail to firstname.lastname@example.org. In the BODY of message put: UNSUBSCRIBE si-list, for more help, put HELP. si-list archives are accessible at http://www.qsl.net/wb6tpu/si-list ****
**** To unsubscribe from si-list: send e-mail to email@example.com. In the BODY of message put: UNSUBSCRIBE si-list, for more help, put HELP. si-list archives are accessible at http://www.qsl.net/wb6tpu/si-list ****