Re: [SI-LIST] : Preferred PWB impedances

D. C. Sessions (dc.sessions@tempe.vlsi.com)
Wed, 19 Nov 1997 09:31:49 -0700

Fred Rosenberger wrote:

> Sorry for being so ignorant but I thought that a "9ma driver" was designed
> to give min of 9ma at 0.4V out for low state (and -9ma at Vdd -0.4V for high
> state if symmetrical). That would be max impedance of about 40 ohms, and
> probably typical impedance of 25 or so ohms. I realize its non-linear
> and the impedance increases some with increasing voltage but not
> a whole lot over the operating range.

On the contrary, the output current is close to saturation near
midrange.
The values that I posted were from simulations of our correlated device
models, so they're pretty close to reality. A CMOS output at half of
supply is nearly a constant-current device.

> I believe some people here have found (via HSPICE simulation) that
> 4ma drivers are best choice for series terminated lines.

Perhaps the objective was to avoid overshoot under best-case conditions.

> Of course we
> should probably have a new spec that gives the max and min voltage into
> a 60 ohm resistance to better indicate the t-line driving characteristics

We (the signal-integrity group) are in something of a dilemma here
because the technology marketing group insists on keeping the old
TTL characterizations. Their reasons are pretty good if you figure
that the majority of customers are clueless, and that seems to be
sadly true.

OTOH, we want to make sure that the diligent and clueful customers have
the tools that they need to do the job right. The compromise that we
hammered out last week is that we'll design the I/Os to let the 'good'
people do a good job, but keep the databook characterizations in the
old format. Customers who are clueful will use the IBIS models and
more complete characterizations anyway.

We WILL publish an ap note describing these as 50-ohm (or whatever,
etc.) drivers.

> > > If you could have only four 3.3v CMOS driver types, what
> > > would you have.
> >
> > DC,
> > Very good question!
> >
> > I don't know if it's possible to compare 3.3V I/O's from
> > different sources, but we recently selected a 9mA buffer as a
> > "standard" buffer.
> >
> > The 9mA buffer corresponds to a 60 ohm reflected wave driver.
> > (Are there standard definitions of voltage levels which define
> > a "reflected-wave driver"?)

Properly, a reflected-wave CMOS driver is one that drives an incident
wave of half of supply. Thus a 60-ohm CMOS reflected-wave driver would
have a worst-case output current of (1500mV/60ohms=25mA) at 1.5v.

> > 2 cases where the 9mA buffer is not used:
> >
> > a) Really short connections between two ASICs.
> > Even at 100Mbit/s, it can be OK to have multiple reflections
> > (you need to use your imagination to see them!). So, in order
> > to reduce switching noise, a weaker driver is used.

I'm assuming that you mean the drivers are weaker than 9mA and
that it takes several reflections to stairstep up to the rails.
That's a good way to avoid overshoot under best-case conditions
but there's no real virtue to making the outputs weaker than
what would give you reflected-wave at best-case.

> > b) PCI
> > If you stick to the spec, you have to use a 30-ohm
> > reflected wave driver, and lots of power/ground pairs.

Intel calls PCI a reflected-wave switching environment, but
it's really just the same old TTL-type signalling with some
limits to keep the worst violations from getting out of hand.
It's a cut-to-fit job instead of a designed-from-first-principles
high-speed bus. That's OK, but we shouldn't kid ourselves.

> > So, that makes only 3 buffer types: 30-ohm, 60-ohm and something weaker.
> >
> > Looking forward to hearing other opinions.
> > Will you post a summary of responses?

> > D. C. Sessions wrote:
> > >
> > > I had an interesting meeting late last week. VLSI is
> > > drafting the specs for our newest I/O library and rather
> > > than wet-finger it, we're asking for a little help in
> > > making them meet real-world needs.
> > >
> > > Here's the question for the floor:
> > >
> > > If you could have only four 3.3v CMOS driver types, what
> > > would you have. (Aside from HSTL, SSTL, etc.) One obvious
> > > candidate is a reflected-wave driver for 50-ohm nominal
> > > lines; that comes to about 10mA at 350mV from the rails
> > > under worst case conditions. A 65-ohm reflected-wave
> > > driver would run about 25% lighter, or about 8mA.
> > > Incident-wave is another matter entirely.
> > >
> > > SO! the floor is open for nominations. Keep in mind that
> > > excessively strong drivers are both inherently slower and
> > > more vulnerable to SSO degradation.

-- 
D. C. Sessions
dc.sessions@tempe.vlsi.com