Re: (Fwd) Buried Capacitance

John R Benham ([email protected])
Fri, 7 Jun 1996 02:15:50 -0400 (EDT)

>>From: "John D. Allen" <[email protected]>
>>To: [email protected]
>>Date: Tue, 4 Jun 1996 07:16:28 -0500
>>Subject: (Fwd) Buried Capacitance
>>Priority: normal
>>X-mailer: Pegasus Mail for Windows (v2.22)

>>John - would you care to comment on this - there is a lot of interest
>>in this technology and how well it pays off.

>>Regards,
>>John Allen

[email protected] writes:

>>>About four or five years ago Zycon Corp began licensing it's buried
>>>capacitance technology to other PCB manufacturers. The technology was
>>>simple, a very thin (~2mil) layer of FR4 sandwiched by 1oz copper
>>>layers. The technology received a somewhat lukewarm reception in the
>>>industry as the max capacitance per square inch was way lower than
>>>what people needed for the mean operating frequency and device
>>>technologies then in vogue. It required about 3 BC stacks to achieve a
>>>usable capacitance (usable means that all the .001uf and .01uf caps
>>>went away). I think Zycon showed a technology roadmap where the dielectric
>>>>constant of the laminate material continued to increase over time.

Well, as you know I spent some time thinking about and modeling decoupling
processes in logic and power systems in a previous reincarnation, and the
conclusion I came to at that time was that Zycon had a good idea, but
unfortunately seemed to be trying to sell it on the basis of the wrong
technical arguments.

If you look at the spatial current demand distribution of logic boards
in the frequency domain you find that the local current sinks have a wide
range of characteristics , varying from large ( 10s of amps) demands in
the low MegaHertz region due to processes like memory refresh, to lower
current demands (0.5-1 amps) in the high MegaHertz/ low GigaHertz region
due to processes like the assertion of data/address buses etc. In order
to supply the large low-frequency bulk currents needed to decouple memory
refresh cycles etc , large tantalum caps (100uF) and 0.01uF, 0.001uF
capacitors are adequate and represent the lowest cost/least risk solution.

However, if you take into account the significant inductance of the
capacitor packages (roughly 0.25nH-> ~0.6nH for SMT ceramics and
~1.6-2ish nH for tantalums) together with the phase delay due to the
typical physical separations between the current sink location and the
location of the decoupling capacitor(s) you find that no matter how many
decoupling capacitors (of any type) are piled on the design, it is still
impossible to really adequately decouple devices like fast bus driver,
cpus etc.

However if you analyse the behavior of closely spaced power-ground plane
systems you find that the variations in the local sink currents propogate
quasi-cylindrical wavefronts into the power/ground plane transmission line
system, and that (after some math) the equivalent inductance of the
power/ground plane system as seen by the current sink is proportional to
the inverse of the laminate material's dielectric constant, and proportional
to the power-ground plane separation [raised to some power which I can't now
recall without digging up my notes :-) ]. The upshot is that for power
-ground plane separations in the range 0.5 - 2 mil, and dielectric
constants around 4 -5, the predicted effective series inductance of
BC systems are roughly 2-3 orders of magnitude lower than can ever be
achieved by discrete decoupling capacitors alone - and it therefore provides
excellent high frequency decoupling.

I won't go into details of the computer model(s) I developed for this, by
the time you've superposed the behavior of the board decoupling, power
distribution system and power supply feedback response and output
filtering it got, well, complex.

The results can be summarized as: you still need decoupling capacitors to
handle the low frequency/ large current demands, but BC, or something very
like it is the only way the industry will reliably meet the decoupling
demands of faster logic and bus systems. BC also has distinct advantages
in the area of EMI and FCC compliance, but again the arguments are lengthy
and I've droned on quite enough here already.....................

Until a couple of weeks ago I thought this analysis was unique. However I
recently came across some discussion on the net which indicated that a
group at the University of Missouri Loyola have recently looked at this
problem and published some papers which apparently seem to arrive at the
same general conclusions. If you're interested I'll hunt around and see if I
can dig up some names and references.

Thanks for the question John, its been a really pleasant change have
a legitimate excuse to think about something other than antennas and high
temperature superconducting RF devices for a while.................

Best regards

John

-----------------------------------------------------------------------
MegaWave Corporation * P.O. Box 614 * 200 Shrewsbury Street
* Boylston, MA 01505 * (508)-869-2700 tel
* (508)-869-2741 fax * [email protected]