Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

RE: 10 Gigabit EMI

Dan (and others),

Warning: Jonathan removes his chair hat.

I fully understand the heresy I spoke on my append on EMI. I fully
understand the difference it makes in having high speed lines that have no
sharp turns (if any turns at all), avoiding the use of passive components
due to the radiation, keeping the line lengths exactly matched, not having
screws protrude from the back of a system, not allowing breaks in the ground
loop on either the board OR THE CHIP, putting decoupling capacitors right on
top of the power and ground pins for a high speed output, putting special
coatings on LCD displays, etc ad nausium.

I've been in far too many customer's offices trying to help them resolve EMI
issues to not take this seriously. Those who have worked with me are
unlikely to ever forget that I told them early in their gigabit design
efforts that EMI would be the thing that would cause them to miss their
dates. Most learned the hard way.

I understand!

And for Jamie, I even brought Dr. Johnson in to teach a class to my
customers so that they would get it.

I understand!!

During my initial presentation a year ago during the kick-off meeting I
raised a number of these same issues, including 8B/10B coding.

I understand!!!

Now, is anyone going to give serious attention to the holes that the optics
are going to stick through (yes, fan holes too)? Anyone that thinks that the
correct application of EMI hygiene will make that issue go away is smoking
something. Perhaps, just perhaps, this is an "implementation" issue and not
worth discussion in this forum. Perhaps, just perhaps, this is an issue that
will delay the rollout of 10 gig products (at least the serial ones) by more
than a year. 

Yes, I am absolutely paranoid! 

Now (deep, cleansing breath) that I have vented....

Jonathan puts his chair hat back on.

It is entirely appropriate for use to look at methods to improve EMI
radiation using coding modifications. This is something that the world will
have to live with for a long time that is not in the control of the

It is also appropriate for us to encourage the industry to fix other
problems that are mandatory for the success of the market development of our
future standard which are outside our scope.  These are not things we should
dwell on. Even so, it would be irresponsible for us to remain silent.

So, rather than calling the writer of this very note out of order (not
within the scope of our objectives), I will offer instead to compile a list
(starting with this) of similar issues that people wish to forward to me.
Simply put on the "Subject:" line "Out of scope issue -- ". Send it either
to me or to the reflector. 

At very least, it will be an interesting exercise.

> -----Original Message-----
> From: DOVE,DANIEL J (HP-Roseville,ex1) [mailto:dan_dove@xxxxxx]
> Sent: Wednesday, March 15, 2000 4:12 PM
> To: 'HSSG_reflector'
> Subject: RE: 10 Gigabit EMI
> >  In the world of EMI, my personal experience is that the 
> >  problems are a combination of religion, black magic, and 
> >  some practical experience (with a little knowledge of 
> >  Maxwell's equations thrown in). Historically, it has been 
> >  very difficult to get "experts" to step up to the plate 
> >  because invariably there are problems in the actual 
> >  implementations that the "experts" simply can't explain. 
> >  This is not a dig on the experts, just a point on the 
> >  extreme difficulty of the problem.
> Actually, it is a matter of understanding that not all of the
> components in your design, are on the schematic. Once you realize
> that, the problems get easier to understand.
> >  It is my opinion that the coding will be an important, but 
> >  second order effect. My opinion is that the rise times are a 
> >  second order effect. My opinion is that the actual bit rate 
> >  is a second order effect. I know, I know. You can bring me 
> >  Maxwell's equations and show me the math that demonstrate 
> >  the direct relationship between these parameters and the emissions.
> Whether a problem is first-order or second-order depends on the 
> magnitude of your first-order problem. If your box has a TON of 
> high frequency attenuation, these problems may appear as second-order.
> On the other hand, if your box is constrained by manufacturing 
> technology, time-to-market, etc.. they may become first order.
> The basic equation is EMI  =  Source Magnitude * Channel Attenuation
> For any given frequency, the Source Magnitude is determined by the
> baud rate, rise time, repetitiveness of edges, amplitude of 
> the signal,
> and imbalances in the electrical channel.
> Channel Attenuation is harder to grasp because there are many paths
> for EMI to get out of a box. Ground/Power planes, air, traces, slots,
> etc.
> One thing for sure... If you have a line-code that is inherently 
> going to be an order of magnitude lower than its alternative in EMI, 
> it is much more likely to fall into a second-order bucket than its
> alternative.
> >  But, in the end, it is the holes is the system that 
> >  matter!!! It is the little antennas built into the boxes. It 
> >  is the phase array of optical transceivers.
> It might be the fan holes, or the power-cord entry for that matter. 
> It depends on the knowledge of your system designer. But whatever the
> "channel" for propagation, it always comes back to the source. What 
> is the source and how big is it?
> >  Most transceiver companies will claim that the majority of 
> >  the noise that radiates from the nose of the optical 
> >  transceiver is generated inside the system, not inside the 
> >  transceiver. The new "small form factor" transceivers, with 
> >  ~1/2 the aperture size of the SC, are not inherently going 
> >  to solve the EMI problem at 10 Gig.
> But there are shielding techniques to reduce the impact of these
> aperatures substantially. Whatever the shielding requirement is, it
> will be based on the spectral content inside the box.
> >  In short, from an implementation perspective, we have a very 
> >  fundamental problem: the way we design systems today as 
> >  racks of modules with arrays of optical transceivers poking 
> >  out through the EMI enclosure may simply not hack it.
> Maybe.
> >  Up to Gig E, we have had the luxury (admittedly, with a fair 
> >  amount of pain) of having a "convenience based, user 
> >  friendly" box design. With 10 Gig, have we finally hit the wall?
> This might have been said about 10 or 100 or 1000. We will see. Rather
> than choose a "spikey" code and roll the dice, I would recommend
> using careful consideration for the code.. then roll the dice. :)
> I liken EMI design to firefighting.. It is better to prevent fires, 
> than try to to contain them. 
> 8B10B might be too hot to handle without some improvements.
> Dan