Site Home   Archive Home   FAQ Home   How to search the Archive   How to Navigate the Archive   
Compare FPGA features and resources   

Threads starting:
1994JulAugSepOctNovDec1994
1995JanFebMarAprMayJunJulAugSepOctNovDec1995
1996JanFebMarAprMayJunJulAugSepOctNovDec1996
1997JanFebMarAprMayJunJulAugSepOctNovDec1997
1998JanFebMarAprMayJunJulAugSepOctNovDec1998
1999JanFebMarAprMayJunJulAugSepOctNovDec1999
2000JanFebMarAprMayJunJulAugSepOctNovDec2000
2001JanFebMarAprMayJunJulAugSepOctNovDec2001
2002JanFebMarAprMayJunJulAugSepOctNovDec2002
2003JanFebMarAprMayJunJulAugSepOctNovDec2003
2004JanFebMarAprMayJunJulAugSepOctNovDec2004
2005JanFebMarAprMayJunJulAugSepOctNovDec2005
2006JanFebMarAprMayJunJulAugSepOctNovDec2006
2007JanFebMarAprMayJunJulAugSepOctNovDec2007
2008JanFebMarAprMayJunJulAugSepOctNovDec2008
2009JanFebMarAprMayJunJulAugSepOctNovDec2009
2010JanFebMarAprMayJunJulAugSepOctNovDec2010
2011JanFebMarAprMayJunJulAugSepOctNovDec2011
2012JanFebMarAprMayJunJulAugSepOctNovDec2012
2013JanFebMarAprMayJunJulAugSepOctNovDec2013
2014JanFebMarAprMayJunJulAugSepOctNovDec2014
2015JanFebMarAprMayJunJulAugSepOctNovDec2015
2016JanFebMarAprMayJunJulAugSepOctNovDec2016
2017JanFebMarAprMayJunJulAugSepOctNovDec2017
2018JanFebMarAprMayJunJulAugSepOctNovDec2018
2019JanFebMarAprMayJunJulAugSepOctNovDec2019
2020JanFebMarAprMay2020

Authors:A B C D E F G H I J K L M N O P Q R S T U V W X Y Z

Custom Search

Messages from 27450

Article: 27450
Subject: Re: Clock Skew : Does Xilinx know what they're doing?
From: "Walter Haas" <walter_haas@pmc-sierra.com>
Date: Wed, 22 Nov 2000 07:35:35 -0800
Links: << >>  << T >>  << A >>
Hi Magnus,

Yes, skew is the difference between the arrival time of two signals, but again, it's the same clock net with the same driver that's have the skew problem. How can one part of one net have a maximum Tpd and another part of that same net have a minimum? Enough so that it makes a big difference, like in the ns range?

And yes, on chip routing is slow. It's all capacitive loading of the entire net plus the gate of the receiver(s). That is not in dispute at all.

What I'm saying is that skew on a signal net that is driven by a single driver can only come from the differences in propagation delay, or buffering elements in between. If all elements get their clock from a single clock net, how can there be skew in the range they are reporting?

I hope this makes it a little clearer...

Cheers,

Wally

Article: 27451
Subject: Re: Using FPGA as PCI target
From: Eckhard Hammer <Eckhard.Hammer@rsd.rsd.de>
Date: Wed, 22 Nov 2000 16:45:07 +0100
Links: << >>  << T >>  << A >>


eml@riverside-machines.com.NOSPAM schrieb:
> 
> On Fri, 17 Nov 2000 14:46:43 +0100, Wolfgang Kufer
> <wolfgang.kufer@rsd.rsd.de> wrote:
> 
> >Hi All,
> >
> >I want to use an FPGA as PCI target.
> >
> >What are the possibilities of configuring this device?
> >Must I use an onboard flash/eeprom device as configuration memory or is
> >there any way to download the configuration data via the pci bus?
> 
> This is a nasty problem. In principle, it should be possible to do it
> by adding some extra circuitry which latches the address of
> configuration writes, and resistor-isolating it so as not to violate
> the PCI loading specs (hey, Mr. Xilinx, have you thought about
> hard-coding this inside the FPGA?). If it works, it should be cheaper
> than storing the configuration on your card.
> 
> This was discussed a couple of years ago here and in
> comp.os.ms-windows.programmer.vxd, but deja doesn't seem to be fixed
> yet so you probably wont be able to find the threads. If you're
> desperate, try emailing Joseph Allen (jhallen@world.std.com) to see if
> he got it to work.
> 
> Evan
But with this method you can't configure your PCI-Device bevor the host
didn't boot the fpga. Is it possible to reconfigure one PCI-Device? Do I
have problems with Win NT? Which drivers do I need??

Thanks to all!
Eckhard

Article: 27452
Subject: Virtex-PCI-Boards
From: Christian Werner <cwerner@informatik.hu-berlin.de>
Date: Wed, 22 Nov 2000 17:07:12 +0100
Links: << >>  << T >>  << A >>
Hi!

I'm looking for Virtex prototyping boards with a PCI-interface.

The PCI-interface shold be controlled by a seperate chip, so that the FPGA
does not need to provide the functionality for the PCI-Bus.

Is there anybody who has some (good or bad) experiences with such a board
and can give me any hints which board I should buy?

As I don't know much about PCI-programming I would appreciate a board
that is shipped with ready-to-use PCI-drivers for Linux and/or Windows and 
some examples how to use them.

Thanks in advance,

Christian Werner


Article: 27453
Subject: Re: Virtex-PCI-Boards
From: Steven Derrien <sderrien@irisa.fr>
Date: Wed, 22 Nov 2000 17:52:48 +0100
Links: << >>  << T >>  << A >>


Christian Werner wrote:
> 
> Hi!
> 
> I'm looking for Virtex prototyping boards with a PCI-interface.
> 
> The PCI-interface shold be controlled by a seperate chip, so that the FPGA
> does not need to provide the functionality for the PCI-Bus.
> 
> Is there anybody who has some (good or bad) experiences with such a board
> and can give me any hints which board I should buy?
> 
> As I don't know much about PCI-programming I would appreciate a board
> that is shipped with ready-to-use PCI-drivers for Linux and/or Windows and
> some examples how to use them.

There are 4 suppliers (as far as I can tell) for this type of board,
here are their website:

http://www.alphadata.co.uk/dsheet/adc-rc1000.html
http://www.annapmicro.com/
http://www.nallatech.com/products/index.htm

http://www.x2e.de/ (Spyder X2)

That's the one I have, It seems that these type of board are more
targeted at fast prototyping rather than for doing a reconfigurable
accelerator,
the design is fine, but it reaaly lack a on board programable clock
generator.

PCI performance are not great (at most 40Mb/sec on slave DMA) but as far
as i heard it is not much worse than for wildforce and co.

BTW, If someone have experience with one of the listed board above, I'd
be happy to hear from him (especially in regards of the software driver
flexibilty and PCI performance)

Steven





> 
> Thanks in advance,
> 
> Christian Werner

Article: 27454
Subject: Re: Clock Skew : Does Xilinx know what they're doing?
From: Muzaffer Kal <muzaffer@dspia.com>
Date: Wed, 22 Nov 2000 16:56:47 GMT
Links: << >>  << T >>  << A >>
On Wed, 22 Nov 2000 07:35:35 -0800, "Walter Haas"
<walter_haas@pmc-sierra.com> wrote:

>Hi Magnus,
>
>Yes, skew is the difference between the arrival time of two signals, but again, it's the same clock net with the same driver that's have the skew problem. How can one part of one net have a maximum Tpd and another part of that same net have a minimum? Enough so that it makes a big difference, like in the ns range?
>
>And yes, on chip routing is slow. It's all capacitive loading of the entire net plus the gate of the receiver(s). That is not in dispute at all.
>
>What I'm saying is that skew on a signal net that is driven by a single driver can only come from the differences in propagation delay, or buffering elements in between. If all elements get their clock from a single clock net, how can there be skew in the range they are reporting?
>
>I hope this makes it a little clearer...
>
>Cheers,
>
>Wally

I think you are missing the fact that you need to look at this as a
transmission line. The clock signal doesn't arrive at all the inputs
at the same time. If you are closer to the source you see the clock
earlier.

Muzaffer

http://www.dspia.com

Article: 27455
Subject: 10 Pcs. Of Paper Money From Around the World 5551
From: fhipvl@hotmail.com
Date: Wed, 22 Nov 2000 17:15:20 GMT
Links: << >>  << T >>  << A >>
Two days ago I ordered 10 pieces of paper money from 10 different countries from a company called Perth Numismatics. Lo and behold they arrived today and they are very nice and colourful. They even have a bill from Antarctica. I didn't even know they existed. These are perfect for stocking stuffers or people who are hard to buy for. The website address is www.perthmoney.com
Good luck and Merry Christmas

Cynthia Reeves

clvpyoqypqrdpjkswtuvwyzd



Article: 27456
Subject: Re: Clock Skew : Does Xilinx know what they're doing?
From: Andy Peters <"apeters <"@> n o a o [.] e d u>
Date: Wed, 22 Nov 2000 10:51:54 -0700
Links: << >>  << T >>  << A >>
walter haas wrote:

/* They are both clocked by the exact same clock net, but a clock net
that is on local routing.  */

The words "local routing" are the key.  Look at the design in the FPGA
Editor, and see exactly how that routing is achieved.  Your clock might
be going through all sorts of stuff (LUTs as pass-throughs, etc) that
contribute to delay on the clock net and hence skew.

-- a
----------------------------
Andy Peters
Sr. Electrical Engineer
National Optical Astronomy Observatory
950 N Cherry Ave
Tucson, AZ 85719
apeters (at) n o a o [dot] e d u

"It is better to be silent and thought a fool, 
 than to send an e-mail to the entire company
 and remove all doubt."

Article: 27457
Subject: Re: Clock Skew : Does Xilinx know what they're doing?
From: Greg Neff <gregneff@my-deja.com>
Date: Wed, 22 Nov 2000 17:55:26 GMT
Links: << >>  << T >>  << A >>
In article <ee6ec96.-1@WebX.sUN8CHnE>,
  "walter haas" <walter_haas@pmc-sierra.com> wrote:
> Hi Everybody,
>
(snip)
>
> I have a design where I have two synchronous elements, a register
feeding a RAM. They are both clocked by the exact same clock net, but a
clock net that is on local routing. The Xilinx tools report that I have
skew between the data signal  and the clock signal between those two
elements.
>
(snip)

It sounds to me like you are not using global clock resources for this
clock.  If this is true, then you will have clock skew problems.
Routing of non-global signals is via horizontal and vertical metal
lines, which are connected via buffers and switches.  If you are not
using a global clock, then two adjacent logic blocks can have the
clocks routed to them in very different ways, depending on how the
local routing resources are allocated.

You *must* use global clock resources for synchronous logic in FPGAs.
If you don't then you are asking for all kinds of trouble.

--
Greg Neff
VP Engineering
*Microsym* Computers Inc.
greg@guesswhichwordgoeshere.com


Sent via Deja.com http://www.deja.com/
Before you buy.

Article: 27458
Subject: Re: Clock Skew : Does Xilinx know what they're doing?
From: "Walter Haas" <walter_haas@pmc-sierra.com>
Date: Wed, 22 Nov 2000 11:32:15 -0800
Links: << >>  << T >>  << A >>
Hi Muzaffer,

Yes, if Xilinx did model these lines as transmission lines, then the closer element would see the clock faster.

However, a point I neglected to mention which I should have, is that Xilinx also told me they do not differentiate between capacitive delay and propagation delay, which would imply not using transmission lines.

Quoting from Norman Johnson, you can lump the parameters of a circuit (as in, ignore transmission line effects) if the length of the net divided by 6 is larger than the rise time (unloaded) of the buffer divided by the speed of electrical material. The speed for FR4 is 180ps/inch.

So, with the rise time being ~1ns (all I could find in the databook), using double the electrical speed of FR4 to 360ps/inch, gives me a lumped parameter cut-off of 1/2 an inch. With no trace being 1/2 inch, there is no effective transmission line. All the capacitance in lumped together. So, I'm back to my original problem. 

And Andy, in regards to the editor, looking at the FPGA editor, the net does not go through any buffering elements of any kind, just through the routing matrix.

But, I am getting good discussion on this. Cheers.

Wally

Article: 27459
Subject: Re: Using FPGA as PCI target
From: eml@riverside-machines.com.NOSPAM
Date: Wed, 22 Nov 2000 20:00:40 GMT
Links: << >>  << T >>  << A >>
On Wed, 22 Nov 2000 16:45:07 +0100, Eckhard Hammer
<Eckhard.Hammer@rsd.rsd.de> wrote:

>But with this method you can't configure your PCI-Device bevor the host
>didn't boot the fpga. Is it possible to reconfigure one PCI-Device? Do I
>have problems with Win NT? Which drivers do I need??
>
>Thanks to all!
>Eckhard

Don't know about NT - try asking on
comp.os.ms-windows.programmer.nt.kernel-mode

Evan

Article: 27460
Subject: Re: Clock Skew : Does Xilinx know what they're doing?
From: eml@riverside-machines.com.NOSPAM
Date: Wed, 22 Nov 2000 20:01:22 GMT
Links: << >>  << T >>  << A >>
On Wed, 22 Nov 2000 07:35:35 -0800, "Walter Haas"
<walter_haas@pmc-sierra.com> wrote:

>Hi Magnus,
>
>Yes, skew is the difference between the arrival time of two signals, but again, it's the same clock net with the same driver that's have the skew problem. How can one part of one net have a maximum Tpd and another part of that same net have a minimum? Enough so that it makes a big difference, like in the ns range?
>
>And yes, on chip routing is slow. It's all capacitive loading of the entire net plus the gate of the receiver(s). That is not in dispute at all.
>
>What I'm saying is that skew on a signal net that is driven by a single driver can only come from the differences in propagation delay, or buffering elements in between. If all elements get their clock from a single clock net, how can there be skew in the range they are reporting?
>
>I hope this makes it a little clearer...
>
>Cheers,
>
>Wally

I'm not sure if I follow your argument, but you seem to be saying that
that if points A and B are on the same line, and are driven from point
C, then C sees a single lumped capacitance and this is therefore not
relevant to the skew calculation. The skew is therefore just a
function of the distance between A and B and the speed of the
wavefront. If so, this is very simplistic, and it doesn't work. I
believe the original Neocad software did this, but certainly Lucent,
and almost certainly Xilinx, have much more sophisticated distributed
element models now. You could easily get real skews of several
nanoseconds across a device, even though the distance between the
points is minimal.

BTW, of all the 'experts' I've come across, 'signal integrity experts'
are easily the least expert...

Evan

Article: 27461
Subject: Re: Virtex-PCI-Boards
From: "Zhen Luo" <zhenluo@ee.princeton.edu>
Date: Wed, 22 Nov 2000 15:47:12 -0500
Links: << >>  << T >>  << A >>
Go to http://www.optimagic.com/ and click on boards. You will find plenty of
choices.

-- Zhen

"Christian Werner" <cwerner@informatik.hu-berlin.de> wrote in message
news:Pine.GSO.4.02.10011221458100.4926-100000@rock...
> Hi!
>
> I'm looking for Virtex prototyping boards with a PCI-interface.
>
> The PCI-interface shold be controlled by a seperate chip, so that the FPGA
> does not need to provide the functionality for the PCI-Bus.
>
> Is there anybody who has some (good or bad) experiences with such a board
> and can give me any hints which board I should buy?
>
> As I don't know much about PCI-programming I would appreciate a board
> that is shipped with ready-to-use PCI-drivers for Linux and/or Windows and
> some examples how to use them.
>
> Thanks in advance,
>
> Christian Werner
>



Article: 27462
Subject: Re: Power consumption FPGA...
From: Theron Hicks <hicksthe@egr.msu.edu>
Date: Wed, 22 Nov 2000 16:16:15 -0500
Links: << >>  << T >>  << A >>


Jurjen Boss wrote:

> Hi,
>
> We want to implement a Xilinx Spartan-II FPGA, type XC2S50, in our
> (excisting) application. I'm trying to estimate the current supply of this
> device by using the datasheets ("Spartan-II 2.5V FPGA Family: DC and
> Switching Charateristics"). From this datasheet I get the following
> information:
> -    Iccintq = 50mA;    (supply current internal part)
> -    Iccoq = 2mA;       (supply current I/O-part)
> -    Iccpo=500mA;      (ramp rate 2ms)
>
> Does this FPGA needs so much current at power-up? How can this ramp rate be
> adjusted? I just can't imagine that this FPGA needs 500mA at power-up. I
> thought these devices where low-power!?! Is there some information available
> on this subject?
>
> Regards,
>
> Jurjen

If you consider that the device is being programmed very rapidly the 500mA is
quite reasonable.  There are millions of CMOS gates being turned on and off at
once so 500mA is quite reasonable.  More to the point... They do indeed require
a high current a programming.  That is why Xilinx says so.  You had better
design for it or you will be sorry.  Note that this is a transient state and
thus the total energy (V*I* time product) required is quite small.  After
programming they are much like any other CMOS device.  namely ... Fast clock
speed means high power.  Zero clock speed means very low power.  Obviously all
sections of the device do not clock at the same rate so you take a summation
over the whole device and integrate over time.


Article: 27463
Subject: Re: Clock Skew : Does Xilinx know what they're doing?
From: Greg Neff <gregneff@my-deja.com>
Date: Wed, 22 Nov 2000 21:53:47 GMT
Links: << >>  << T >>  << A >>
In article <ee6ec96.5@WebX.sUN8CHnE>,
  "Walter Haas" <walter_haas@pmc-sierra.com> wrote:
> Hi Muzaffer,
>
(snip)
>
> And Andy, in regards to the editor, looking at the FPGA editor, the
net does not go through any buffering elements of any kind, just
through the routing matrix.
>
(snip)

The point Andy and I are trying to make is that local routing resources
are not designed for delivering low-skew clocks.  The routing matrix
includes active elements (switches and buffers) that appear only as
route matrix intersections in FPGA editor.  Only the global clock
routing resources provide the balanced buffer delays and net
propagation delays needed for low skew clock routing.

I have had to fix designs for companies where engineers used local
routing resources for synchronous clocks.  These designs had symptoms
like temperature and lot dependent failures.  Please trust me when I
tell you that forcing synchronous clocks through local routing
resources is a *bad* idea.

--
Greg Neff
VP Engineering
*Microsym* Computers Inc.
greg@guesswhichwordgoeshere.com


Sent via Deja.com http://www.deja.com/
Before you buy.

Article: 27464
Subject: Re: Clock Skew : Does Xilinx know what they're doing?
From: Magnus Homann <d0asta@mis.dtek.chalmers.se>
Date: 22 Nov 2000 23:28:23 +0100
Links: << >>  << T >>  << A >>
"Walter Haas" <walter_haas@pmc-sierra.com> writes:

> However, a point I neglected to mention which I should have, is that
> Xilinx also told me they do not differentiate between capacitive
> delay and propagation delay, which would imply not using
> transmission lines.

What do you mean? I don't think they model the capicatance and the
length separately, but instead the measure the actual time througha
conductor, and from that you either calculate an "effective Er" or an
"effective capacitance" or an "effective lngth", if you want. But what
is interesting is the Tpd, not why it is there.

Surely, they have more intricate models than this, but that's just
what they are, models.

> Quoting from Norman Johnson, you can lump the parameters of a
> circuit (as in, ignore transmission line effects) if the length of
> the net divided by 6 is larger than the rise time (unloaded) of the
> buffer divided by the speed of electrical material. The speed for
> FR4 is 180ps/inch.

Would that be Howard Johnson? If you have the chance to attend one of
his classes, try to! Your quote doesn't sound right: The length of the
net multiplied by speed gives you the delay. That delay should be less
than 1/6 of the rise time. Have a look at

http://www.sigcon.com/news/3_18.htm

I do not understand why you have the speed for FR4. Surely, the
substrate is of silicon, and not FR4. There might be FR4 in the BGA,
but that's outside the die. The speed of the electrcial material is,
in this case the Tpd of the net/divided by the length of it.

If you have the Tpd of the net in the FPGA at 2ns (either due to
distributed cap, length or Er), it is not likely that you can use any lumped
model. Say the driver has a (linear) Tr of 1ns. Also say that the
receiver have a threshold of 50 %. If you place one receiver halfway
(no stub) and the other in the end, there is no way you get the skew
less than 1 ns. In fact, add to that 20% variation of speed for the
line between receivers, and then add 50% of Tr for different
thresholds at the receivers... It all adds up. Now, if you route from
driver to he recievers completely different way, you could end up with
much more than that.

I have even heard people mumble about transmission line effects on
chip (over/undershoot/reflection etc).

> And Andy, in regards to the editor, looking at the FPGA editor, the
> net does not go through any buffering elements of any kind, just
> through the routing matrix.

Just out of curiosity, are the recievers close to eachother? Are they
routed different ways?

Homann
-- 
Magnus Homann, M.Sc. CS & E
d0asta@dtek.chalmers.se

Article: 27465
Subject: Re: Clock Skew : Does Xilinx know what they're doing?
From: Rick Filipkiewicz <rick@algor.co.uk>
Date: Wed, 22 Nov 2000 22:33:44 +0000
Links: << >>  << T >>  << A >>


Magnus Homann wrote:

>
>
> I would guess that they take the max Tpd to the receiver as skew. Not
> many FPGA (any?) manufacturer have implemented min. Tpd calculation. A
> min Tpd of 25% of Max, would still get you 1.50 ns skew. Skew is diff
> between max delay to one element and min delay to the other. If you
> have a min of 0, skew equals max delay.
>

This analysis is way too pessimistic. The difference between Tpd(max) & Tpd(min) is over all variations of temp, voltage, silicon. However clock skew
analysis applies to a specific device where

o the voltage is constant

o the temp may vary and there may be localised hot spots but not not over the range  of a few CLBs - assuming in the situation described here the reg &
RAM are placed reasonably close together.

o The silicon doping levels will be constant.

Therefore all the internal devices will be at or near the same point in their min-max range. [Walter you might try to see if this is allowed for in the
Xilinx skew calculation by setting the Temp & Voltage derate parameters in the UCF/PCF].> "The way in which we implement and characterize our internal
routing is proprietary.  What I can tell you is this- believe what the tool tells you:

> > there is skew between those synchronous elements.  You may infer what you'd
> > like about the meaning of that (existance of buffers and other such elements
> > that would cause skew / type of material we use to implement routing),
> > but the fact remains- you have a race condition."
> >
> > Umm...OK. Anybody have any ideas on what this proprietary way is? Or, is it just a way to tell me to be quiet and just use the tools as is?
>

That's just a warning that Xilinx are about to take out a patent on Physics.

>
> It's a way telling you to stick to the specified interface, so that
> Xilinx can redesign internally as needed. Could be said in a nicer
> way, but i basically agree with him/her.
>

Taken to it conclusion this argument means we should all go back to pushing the big friendly green GUI button and forget about trying to make the tools
do what we want.

Don't forget its not Xilinx that has to get products out the door with these *&~$!!## devices, its us.


Article: 27466
Subject: Re: Clock Skew : Does Xilinx know what they're doing?
From: gah@ugcs.caltech.edu (glen herrmannsfeldt)
Date: 23 Nov 2000 00:14:42 GMT
Links: << >>  << T >>  << A >>
"Walter Haas" <walter_haas@pmc-sierra.com> writes:

>Yes, if Xilinx did model these lines as transmission lines, 
>then the closer element would see the clock faster.

>However, a point I neglected to mention which I should have, 
>is that Xilinx also told me they do not differentiate between 
>capacitive delay and propagation delay, which would imply not 
>using transmission lines.

>Quoting from Norman Johnson, you can lump the parameters of a 
>circuit (as in, ignore transmission line effects) if the length 
>of the net divided by 6 is larger than the rise time (unloaded) 
>of the buffer divided by the speed of electrical material. 
>The speed for FR4 is 180ps/inch.

>So, with the rise time being ~1ns (all I could find in the databook), 
>using double the electrical speed of FR4 to 360ps/inch, gives me a 
>lumped parameter cut-off of 1/2 an inch. With no trace being 1/2 inch, 
>there is no effective transmission line. All the capacitance in lumped 
>together. So, I'm back to my original problem. 

Someone else probably knows the details better, but as I understand
it the lumped capacitance model didn't work as of about 0.3 micron
technology.  The problem is that the resistance of the lines is
much more significant than the source impedance of the driver.
(The metal gets thinner and narowwer, so the resistance increases
with the square of the scale factor.  Capacitance decreases linearly.)

Simple transmission line theory assumes no loss, so that doesn't
help much here.  For PC boards, the wire resistance is small enough
that simple transmission line theory works.  That is, use only
the capacitance and inductance terms.  In this case, you need the
resistive and inductive terms, and it is more like a diffusion 
equation than a wave equation.  Note also that the dielectric
constant of Si is around 10.

-- glen

Article: 27467
Subject: Survey on design methodologies
From: "Zhen Luo" <zhenluo@ee.princeton.edu>
Date: Wed, 22 Nov 2000 21:46:55 -0500
Links: << >>  << T >>  << A >>
Thanks for the replies for my previous posting. They are really helpful.
Here are more questions about general aspects of configurable computing.

I think most FPGA designers today go for the HDL synthesis and automatic
placement and routing approach. And in this process, designers would try to
include the hard macros like Xilinx cores in their design to improve the
overall performance. However, one of my reviewers believed most designers
for commercial products would probably still go for manual placement and
routing.

Secondly, I state that configurable computing shortens the development cycle
by comparison to ASIC design. However, one of my reviewers argue that ASIC
design can also be done in a comparable time frame through design
automation. I wonder if that is really the case.

Please tell me your opinions on these two issues and I would be very
grateful for you answers.

Thanks!

-- Zhen Luo





Article: 27468
Subject: Re: Clock Skew : Does Xilinx know what they're doing?
From: Greg Neff <gregneff@my-deja.com>
Date: Thu, 23 Nov 2000 04:33:46 GMT
Links: << >>  << T >>  << A >>
In article <3A1C49C8.DBCFFD61@algor.co.uk>,
  Rick Filipkiewicz <rick@algor.co.uk> wrote:
(snip)
>
> This analysis is way too pessimistic. The difference between Tpd(max)
& Tpd(min) is over all variations of temp, voltage, silicon. However
clock skew
> analysis applies to a specific device where
>
> o the voltage is constant
>
> o the temp may vary and there may be localised hot spots but not not
over the range  of a few CLBs - assuming in the situation described
here the reg &
> RAM are placed reasonably close together.
>
> o The silicon doping levels will be constant.
>
> Therefore all the internal devices will be at or near the same point
in their min-max range.
(snip)

This is only partially true.  I have seen repeatable FPGA FF hold-time
failures at ambient temperature extremes, and in different FPGA lots,
in designs where clocks were routed using local resources.  Things like
counters getting stuck, and shift registers scrambling bits.  I'm not
talking theoretical analysis here, I'm talking about actual observed
failure modes and effects.

I'll say this one more time: routing clocks using local routing
resources is a *bad* idea.  If you do this, it will come back to haunt
you.  Unfortunately, many engineers that do this don't understand the
problem they are introducing, and are then at a loss to explain why the
design is flaky in the field.  For example, a system that failed to
respond to a fire sensor input, but only when the system got hot...

--
Greg Neff
VP Engineering
*Microsym* Computers Inc.
greg@guesswhichwordgoeshere.com


Sent via Deja.com http://www.deja.com/
Before you buy.

Article: 27469
Subject: work from home
From: rcat1204@aol.com (Rcat1204)
Date: 23 Nov 2000 05:11:32 GMT
Links: << >>  << T >>  << A >>
Earn $2000+/month working from home. 

simply by reading e-mails and surfing the internet.

-No, cost to you! Everything is free!

 To find out details visit :

 http://rcat1204.tripod.com/pages

Article: 27470
Subject: Xilinx XC4000** Speed Grades
From: "Akito" <..@no.com>
Date: Thu, 23 Nov 2000 06:25:44 GMT
Links: << >>  << T >>  << A >>
Greetings, I've been wondering for some time now what the different
"speed grade" versions really mean.

I'll compile a design for a XC4005XL-3C and it may run as fast as 80MHz.
From some semi-ambiguous graphs on Xilinx's site, 3C is only supposed to do
about 40MHz...

Pardon for the lame question, I just cant find any clear definition.



Article: 27471
Subject: How to reduce the Tco
From: "Frank Z.F Xie" <frank.xie@latticesemi.com>
Date: Thu, 23 Nov 2000 15:54:36 +0800
Links: << >>  << T >>  << A >>
Hi, there

Currently I'm doing a design with VirtexE, but have problem to achieve Tco
performance:

AD is a bidi bus with oe control
AD <= D_OUT when OE = '1' else "ZZZZ";

here D_OUT is a registered internal signal, and also used somewhere else in
the design.

When I run it in Foundation ISE 3.1i + SP5, the SW keeps implement the
tristate with internal tristate bus(BUFT), which will add more delay, and
the Tco is as long as 10ns. Is there anyway to disable the internal tristate
in the SW? How to?

Thanks

Frank




Article: 27472
Subject: Re: Clock Skew : Does Xilinx know what they're doing?
From: Magnus Homann <d0asta@licia.dtek.chalmers.se>
Date: 23 Nov 2000 09:18:37 +0100
Links: << >>  << T >>  << A >>
Rick Filipkiewicz <rick@algor.co.uk> writes:

> This analysis is way too pessimistic. The difference between
> Tpd(max) & Tpd(min) is over all variations of temp, voltage,
> silicon. However clock skew analysis applies to a specific device
> where
> 
> o the voltage is constant
> 
> o the temp may vary and there may be localised hot spots but not not over the range  of a few CLBs - assuming in the situation described here the reg &
> RAM are placed reasonably close together.
> 
> o The silicon doping levels will be constant.

You are of course right. In another post I noted that all FPGA
manufacturers considers max Tpd only. I also said that I would be
happy if they even managed to characterize the ouput hold of an I/O.

After they have have done that, I would also be interested in skew
on-chip.

> > It's a way telling you to stick to the specified interface, so that
> > Xilinx can redesign internally as needed. Could be said in a nicer
> > way, but i basically agree with him/her.

> Taken to it conclusion this argument means we should all go back to
> pushing the big friendly green GUI button and forget about trying to
> make the tools do what we want.

No. If you want to do a worst case design, I recommend always stick to
the public spec. That goes for all digital design.

Noit many digital devices have publicied there skew, I've had this
problem with all sorts of devices.

"Skew? Zero, of course!"

Homann
-- 
Magnus Homann, M.Sc. CS & E
d0asta@dtek.chalmers.se

Article: 27473
Subject: Re: Spartan and XC4000 configuration
From: mohankumar <m_nagalingam@hotmail.com>
Date: Thu, 23 Nov 2000 01:44:49 -0800
Links: << >>  << T >>  << A >>
hello everybody,
           i am working with XILINX foundation series v1.5(students editionn). in my design work, when i synthesize one of my macro (schematic type) it is giving right pins for Virtex family but does not work for spartan family. can any one help to me.

Article: 27474
Subject: survey of fpga application
From: "Jianyong Niu" <cop00jn@shef.ac.uk>
Date: Thu, 23 Nov 2000 10:01:20 -0000
Links: << >>  << T >>  << A >>
hi, all. I am plowing a lieteratural survey on fpga applications. I found
that there are six major potential fields: communication &networks,
consumer, industry, automotive, data processing and military& aerospace. but
I can not find the details of them and some examples. anybody can give me
some valuable reference?

cheers

Jianyong.Niu





Site Home   Archive Home   FAQ Home   How to search the Archive   How to Navigate the Archive   
Compare FPGA features and resources   

Threads starting:
1994JulAugSepOctNovDec1994
1995JanFebMarAprMayJunJulAugSepOctNovDec1995
1996JanFebMarAprMayJunJulAugSepOctNovDec1996
1997JanFebMarAprMayJunJulAugSepOctNovDec1997
1998JanFebMarAprMayJunJulAugSepOctNovDec1998
1999JanFebMarAprMayJunJulAugSepOctNovDec1999
2000JanFebMarAprMayJunJulAugSepOctNovDec2000
2001JanFebMarAprMayJunJulAugSepOctNovDec2001
2002JanFebMarAprMayJunJulAugSepOctNovDec2002
2003JanFebMarAprMayJunJulAugSepOctNovDec2003
2004JanFebMarAprMayJunJulAugSepOctNovDec2004
2005JanFebMarAprMayJunJulAugSepOctNovDec2005
2006JanFebMarAprMayJunJulAugSepOctNovDec2006
2007JanFebMarAprMayJunJulAugSepOctNovDec2007
2008JanFebMarAprMayJunJulAugSepOctNovDec2008
2009JanFebMarAprMayJunJulAugSepOctNovDec2009
2010JanFebMarAprMayJunJulAugSepOctNovDec2010
2011JanFebMarAprMayJunJulAugSepOctNovDec2011
2012JanFebMarAprMayJunJulAugSepOctNovDec2012
2013JanFebMarAprMayJunJulAugSepOctNovDec2013
2014JanFebMarAprMayJunJulAugSepOctNovDec2014
2015JanFebMarAprMayJunJulAugSepOctNovDec2015
2016JanFebMarAprMayJunJulAugSepOctNovDec2016
2017JanFebMarAprMayJunJulAugSepOctNovDec2017
2018JanFebMarAprMayJunJulAugSepOctNovDec2018
2019JanFebMarAprMayJunJulAugSepOctNovDec2019
2020JanFebMarAprMay2020

Authors:A B C D E F G H I J K L M N O P Q R S T U V W X Y Z

Custom Search