Thanks for your reply. I agree with you that measuring (and simulating) the
power ground impedances on a complex board is not an easy task. In a
real-life situation it is complicated by the fact that boards are not
designed with probe points for this purpose, therefore attaching the probes
may be extremely difficult. I also agree with your derivation which
illustrates how much noise you would pick up with adjacent loops in a
typical situation. This clearly shows that we need to go an extra distance
to achieve our goal. Your derivation also suggests the direction where the
solution can be found: the only help is to reduce the mutual inductance,
since reducing the source voltage would reduce dI/dt but would also reduce
the voltage to be measured by the same ratio. By using smaller loops
(closer connection points between the shield and signal wire, and very short
exposed signal and shield connections), the 'noise floor' of the measurement
can be reduced to about 25 milliohms up to about one GHz, which may be
enough today for some of the boards. This noise floor can be achieved even
if the connection points are just 0.1 inches away, which brings up another
interesting point: on a power-ground distribution system we want to know not
only how much noise will propagate from one device to the other, but also it
is important to figure out how much noise a particular device is generating
at its own location. Having the full impedance matrix (or any other
electrical matrix for that matter) gives the possibility to sum up the noise
contributions from several devices at the desired locations. To have the
self-impedance measured (Z11), we clearly need to connect the source and the
probe so close that at our measurement frequencies the electrical separation
is negligible. By using semirigid for both the source cable and probe cable
and using very short interconnections, we were able to measure self and
transfer impedances and to correlate to simulated values within less than
one dB error up to 100MHz and within less than 3dB error up to one GHz. To
achieve the good correlation it is also necessary to use the proper complex
impedances in the equivalent schematic of voltage divider.
But finally let me repeat again that I agree with you, it is not an easy
task.
Best regards
Istvan Novak, SI Engineer
SUN Microsystems
-----Original Message-----
From: Howard Johnson <howiej@sigcon.com>
To: Istvan NOVAK <istvan.novak@worldnet.att.net>
Cc: si-list@silab.Eng.Sun.COM <si-list@silab.Eng.Sun.COM>
Date: Tuesday, June 09, 1998 6:22 PM
Subject: Re: [SI-LIST] : Re: MEASURING POWER GROUND IMPEDANCE
>NOTE: Since this question was copied to the S.I. list reflector,
>I will copy my answer there as well. I of course always welcome
>any other comments on the subject, pro or con. Measuring the
>power and ground impedance on a complex board is never easy.
>
>
>Hello Istvan,
>
>Regarding the comment below about "polluting your measurement",
>the issue is that we are trying to measure a very small signal (the
>Vcc noise voltage), and any direct crosstalk between the signal
>source and the probe may overwhelm the signal you are trying to
>measure.
>
>Example: suppose we have a 50-ohm sinewave source, and a 50-ohm probe.
>Adjust the sinewave source so we see 1V p-p when the source is plugged
>straight into the probe (no power system connected).
>The total driving point impedance at the probe point is 25 ohms (fifty
>from the source in parallel with fifty from the probe).
>Now connect both probe and scope to the power system on your bare
>board (no active parts installed - just the bypass caps).
>If the Vcc-ground impedance is 0.1 ohms (a reasonable value for a good
>power system), then the Vcc-ground impedance will reduce the voltage
>received at the scope by a ratio of (0.1)/(25) -- that's the
>resistor divider theorem. (In the analysis below I just converted this
>number to a dB figure.)
>The voltage at the scope will now be about 4 mV. This is a small,
>but clearly visible and easily measured amount of noise.
>
>Now let's look at the direct coupling between the source and the probe.
>Couple both source and probe to the board using coax (I
>like to use RG-174, because it's thin, flexible, and easy to solder
>to the board). Let's say that at the source attachment point, the
>distance between the coax shield attach point (to board ground) and
>the coax signal attach point (to board Vcc) is about 1/2 inch,
>and that the coax signal conductor stands up about
>1/8 inch above the surface of the board. The total exposed area of this
>loop between the signal conductor and the ground conductor
>is (1/2)*(1/8) = 0.0625 square inches.
>Assume the same for the probe connection.
>
>If the source and probe coax cables are connected to points
>separated by about 1 inch, the mutual inductive coupling between
>these two loops will be on the order of 0.02 nH. The total
>di/dt flowing through the source loop, times the mutual inductance
>between loops, will coupled crosstalk noise voltages directly
>into the probe loop.
>
>At a frequency of, say, 500 MHz, the total noise voltage
>will be: Lmutual*(di/dt = 0.02nH*2*(pi)*500MHz*(1V/25ohms) = 2.5 mV
>
>The noise voltage is almost as large as the signal you are trying to
>measure. Keep the source and probe cables searated by at least
>an inches, and keep the length of the exposed coax signal conductor
>as short as practical.
>
>Also, at frequencies in the multiple-megahertz range and beyond,
>you will notice that the impedance is a function of the
>position of the source and probe cables, due to various resonance
>effects between the power and ground planes. If you're interested
>in that topic, check out the cool simulations from SIGRITY (and probably
>many others) on the subject.
>
>
>Best regards,
>Dr. Howard Johnson
>
>
>
>
>At 08:08 AM 6/1/98 -0400, you wrote:
>>Dr. Johnson,
>>
>>I am following with great interest the On-line Newsletters of High-Speed
>>Digital Design. Inj one of your recent responses regarding the
measurement
>>of power/ground planes, you are suggesting not to connect the probes to
>>nearby points.
>>Can you elaborate somewhat further your reasoning for that ('local
crosstalk
>>will pollute your measurement')?
>>Also, can you give some more reasoning behind your suggestion of using the
>>resistance divider theorem in this case? Do you mean here the magnitudes
of
>>impedances?
>>
>>Thank you
>>
>>Istvan Novak, SI Engineer, SUN Microsystems
>>
>>-----Original Message-----
>>From: High-Speed Digital Design Mailing List <hsdd@accessone.com>
>>To: High-Speed Digital Design Newsletter Subscriber(4) : ; <High-Speed
>>Digital Design Newsletter Subscriber(4) : ;>
>>Date: Tuesday, May 26, 1998 4:28 PM
>>Subject: MEASURING POWER GROUND IMPEDANCE
>>
>>
>>>*---------------------------------------------------------------*
>>> H i g h - S p e e d D i g i t a l D e s i g n
>>>
>>> *On-Line Newsletter*
>>>
>>> Dr. Howard Johnson, Vol. 2 Issue 14
>>>*---------------------------------------------------------------*
>>>
>>>*------------------------(ANNOUNCEMENTS)------------------------*
>>>
>>>Next Public High-Speed Digital Design Seminars:
>>>
>>> U. of Oxford, UK June 22-23, 1998
>>> San Jose, CA September 21-22, 1998
>>>
>>>Registration & information at:
>>>http://signalintegrity.com/seminar.htm
>>>
>>>Tell Your Co-Workers!
>>>
>>>*--------------------------(QUESTION)---------------------------*
>>>
>>>MEASURING POWER GROUND IMPEDANCE
>>>Larry Smith of Sun Microsystems writes:
>>>
>>> Dr. Johnson - thanks for your newsletter, I have just
>>> subscribed. I completely agree with the comments pertaining
>>> to power plane impedance and the 'single node' assumption
>>> below the frequency where the board resonates (Volume 2 Issue
>>> 14). We have checked the impedance of power planes with a
>>> network analyzer. With no capacitors present, you can see
>>> interesting resonances that depend on the 1/4 wavelength from
>>> the probes to the card edge (valleys) and half wavelengths
>>> that fit into the card dimensions. But these measurements
>>> always come in dB and my spice simulations come out in ohms
>>> (after I force 1 Amp). Do you have any ideas on how to
>>> correlate between them? I would like to be able to measure
>>> the plane impedance in Ohms!
>>>
>>> A minor point... We are looking at using HIGHer dielectric
>>> constants for the material between power planes in order to
>>> gain more decoupling capacitance. That is going to lower the
>>> resonant frequencies of the power planes, possibly into
>>> frequencies of interest (near the clock).
>>>
>>> Best Regards,
>>>
>>>*-------------------(REPLY FROM DR. JOHNSON)--------------------*
>>>
>>> Thanks for your interest in High-Speed Digital Design.
>>>
>>> When using a network analyzer to measure power and ground
>>> impedance, we drive the power and ground planes at one point
>>> on the board with a sine wave, and the measure how much
>>> voltage appears at a different point.
>>>
>>> Don't set the IN and OUT cable attachment points on the board
>>> too close together or else their local crosstalk will pollute
>>> your measurement.
>>>
>>> In this setup, we can relate ohms to dB if (1) we know the
>>> driving point impedance of the network analyzer test setup
>>> and (2) the network analyzer is calibrated in terms of dB
>>> gain, where 0 dB means there is no device under test
>>> connected (that is, the IN cable is directly connected to the
>>> OUT cable).
>>>
>>> To establish this relation, we just use the resistance
>>> divider theorem:
>>>
>>> dB gain = 20*log(Vmeasured / Vreference)
>>>
>>> dB gain = 20*log( Rpwr-gnd / (Rpwr-gnd + Rsource) )
>>>
>>> where Rsource is the driving point impedance of the test
>>> setup.
>>>
>>> If your network analyzer has a 50-ohm output, and a 50-ohm
>>> input, then the driving point impedance at the device-under-
>>> test point is 25 ohms. Rsource = 25 ohms.
>>>
>>> Now we can simplify things a little if we use the fact that,
>>> even when resonating, the power and ground planes have an
>>> impedance much less than 25 ohms (if they didn't the whole
>>> system wouldn't even come close to working). Therefore we can
>>> ignore the term Rpwr-gnd in the denominator, and just use the
>>> approximation:
>>>
>>> dB gain = 20*log( Rpwr-gnd / Rsource )
>>>
>>> Converting this formula to express Rpwr-gnd in terms of dB,
>>> we get:
>>>
>>> Rpwr-gnd = Rsource *{10 <raised to the power of> [(dB
>>> gain)/20]}
>>>
>>>
>>>Best regards,
>>>Dr. Howard Johnson
>>>
>>>*---------------------------------------------------------------*
>>>
>>> Comments welcome! hsdd@signalintegrity.com
>>>
>>> Newsletter Archives:
>>> http://signalintegrity.com/newsletter.htm
>>>
>>
>_________________________________________________
>Dr. Howard Johnson, Signal Consulting, Inc.
>tel 425.556.0800 fax 425.881.6149 email howiej@sigcon.com
>
>High-Speed Digital Design seminars:
>June 22-23 at Oxford U., in the U.K.,
>September 21-22 in San Jose, CA
>S E E - - - >>> WWW.sigcon.com
>**** To unsubscribe from si-list: send e-mail to
majordomo@silab.eng.sun.com. In the BODY of message put: UNSUBSCRIBE
si-list, for more 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 majordomo@silab.eng.sun.com. In the BODY of message put: UNSUBSCRIBE si-list, for more more help, put HELP. si-list archives are accessible at http://www.qsl.net/wb6tpu/si-list ****