I can design an I/O to give you exactly the edge rate you want on any given
pin of a custom IC. Sometimes that's exactly what a customer pays for. Most
of the time, though, they want "How fast can I make my critical path" -- and
they're not necessarily talking about the hardware.
I/O library development is a large part of bringing out a new process technology.
Thanks to core-vs-I/O voltage mismatches there's a lot of redesign every time
around. Since the effort to redesign, requalify, and recharacterize a library
increases with the size of the library we have a STRONG reason to keep the
libraries simple.
There's only one edge rate that we KNOW will be required in any library: full
speed, minimum intrinsic, and let the customer deal with the SI. Aside from
that, every customer who wants controlled edge rates wants a *different* edge
rate. One wants 1000ps, because he's running only 50 MHz synchronous across
ten cm of point-to-pont trace. Another wants 5000ps to use at 16 MHz in a
random-routed bus. And so forth.
Keep in mind that I have it relatively easy. *MY* customers know what the PWB
environment will be in advance. Now what is my opposite number over in Philips
logic products going to do with his new 74xxx244 part (Or the one at Altera
doing a new line of PLDs)? Make it fast enough to suit the DIMM buffer
customers, or slow enough to use with IEEE 1284? Maybe somewhere in the middle?
Should we have several edge-rate variants within each family of standard logic
(multiply the number of catalog items in each generation by the number of
desired edge rates.)
While we're cranking away at the 10ns library our competitors will be shipping
product for the next generation and designing cells for the one after that.
Then when we finally release the 0.5 micron CMOS library for customer design,
both of the ones who waited that long will use the fast cells. One because
she needs the speed and the other because it was specified in the RFP as being
available from multiple vendors and the PHB who picked us doesn't even know
what edge-rate control IS.
Now none of these are insurmountable obstacles. Just don't expect the Si
manufacturers to do it all in hopes of being noticed (been there, done that,
can show you the scars.) We're in the business of providing customers with
things that THEY think are important, and our test of their priorities is
straightforward: are they (that means YOU) willing to pay for it? Will it
disqualify us from a (large) design win? If customers aren't hard-currency
serious we'd be fools to waste resources on an empty gesture to the Right
Way to Do Things.
So, people, the ball is in your court. If you really need controlled edge
rates, make them part of your purchase qualification process. Pass that
bit of news along to your suppliers, and I can promise you that if enough
of you vote with your purchase orders you WILL get those controlled edge
rates. Even if you're the only one who wants them -- because that's what
I do for a living.
> Robert Tsai <[email protected]> on 07/12/99 04:24:02 PM
>
> Please respond to [email protected]
>
> Sent by: Robert Tsai <[email protected]>
>
> To: [email protected]
> cc: (Roy Leventhal/MW/US/3Com)
> Subject: Re: [SI-LIST] : Some Semiconductors are Unnecessarily Fast
>
> Roy Leventhal wrote:
>
> > Several recent threads have commented that few nets on a board used to be a
> > signal integrity challenge in the past and now almost all are.
> >
> > What we are seeing are a number of instances where the semiconductor companies
> > are producing parts with edge rates way faster than the clock and application
> > calls for. Parts with 500ps rise times or less and clock periods of 100ns or
> > more. SI engineers are seeing more demand for their skills what with shrinking
> > geometries and lack of (care? concern?) edge rate control on drivers.
> >
> > But, this is "make work" and is not the road to world class competitiveness
> for
> > our companies. There are enough real world signal integrity problems for us to
> > tackle without any "help" of this sort from the semiconductor manufactures.
> >
> > Roy Leventhal
> >
> > **** To unsubscribe from si-list: send e-mail to [email protected].
> 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 ****
>
> Hi Roy,
>
> As many of you already know that the edge rate is directly related to the
> channel length of the drivers. When the semiconductor technology shrinks its
> feature sizes from 0.5 to 0.35 to 0.18 um the edge rate goes up accordingly.
> Unless special designs are implemented to intentionly slow down the edge rate,
> the edge rate will keep going up. We are not making everybody's life tough
> purposely.
>
> Think positively, this is job security for all of you.
>
> Robert Tsai
>
> **** To unsubscribe from si-list: send e-mail to [email protected]. 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 protected]. 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 ****
-- D. C. Sessions [email protected]**** To unsubscribe from si-list: send e-mail to [email protected]. 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 ****