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 8400

Article: 8400
Subject: Xilinx Configuration Problem
From: "Acromag Web Surfer" <webuser@acromag.com>
Date: 12 Dec 97 21:30:55 GMT
Links: << >>  << T >>  << A >>
We have used the Xilinx XC5200 FPGA family and XC17XXXD serial PROM parts
in a number of designs.  We configure via master serial mode with the INIT~
line tied to the PROM OE~ line with an external 4.7K pullup resistor. 
After a number of customer complaints reporting product malfunctions, we
investigated and learned that the Xilinx FPGA will not reliable configure
on power up if the power supply has a slower then 4ms (typically) rise time
from 3 volts to 4.75 volts.  Note that 4ms is a short time since we have a
number of supplies that require greater then 50ms to rise from 3 volts to
4.75 volts.  This is a serious problem!

Has any one else experienced this problem?  Do you have any recommended
solutions?

One solution is to have all our customers use power supplies that power up
quickly.  This is not a realistic request.  Another solution is to hold off
configuration until a safe configuration voltage is reached (i.e 4.75
volts).  This requires a power supply monitor circuit that would hold the
INIT~ line low until 4.75 volts is detected.  This circuit can easily be
included on the configuration PROM dice. Since the PROM's OE~ signal is
often tied to the INIT~ signal, the PROM can then hold off configuration
(with its built in power supply monitor circuit) until a safe programming
voltage is detected.   Does anyone know of such a PROM?


Article: 8401
Subject: ATMEL 17C256 problems solved ...
From: Tobias Hilpert <thilpert@osc.de>
Date: Fri, 12 Dec 1997 23:36:10 +0100
Links: << >>  << T >>  << A >>
Hello to everyone,

About two weeks ago, I posted an article concerning configuration
troubles of a XC4013E-2 using an ATMEL 17C256 EEPROM.

First - many thanks for the great number of helpful replies and Emails I
received.

An email message from ATMEL finally gave the "wall-breaking" hint:
Some early delivered devices do not sense the RESET polarity on power-up
correctly! They wrote, that devices with the following codes (number
under the device designation 17C256) seem to have this problem:
        6D0852
        6D0853
        6D0854
        6D0855

I looked at my components and had one one of them ... :(

Now, I managed to get newer ones (7Axxx code) and they work without any
problems :) .

Best Regards,

  Tobias Hilpert
Article: 8402
Subject: bus design in Altera 10K, how to increase speed
From: muzok@pacbell.net (muzo)
Date: Fri, 12 Dec 1997 22:43:10 GMT
Links: << >>  << T >>  << A >>
hi,
I have a design which uses %55 of a 10K100 and within this design I have 12 8 bit
registers which need to be selectively loaded into another register, IOW, I have
an 8 bit bus from which 12 8 bit registers hang. Because 10K doesn't have
internal tristates, I have a big mux which selects the register. Currently this
mux is the bottleneck in my design which limits my speed to  11 MHz in a -3 part.
Any ideas how to increase the speed in this design? The design is in Verilog.

Are there any ways to get more performance by using the floor plan tool ? I am
already using FAST synthesis and maximum speed in the compiler options.

thanks

muzo

WDM & NT Kernel Driver Development Consulting <muzok@pacbell.net>
Article: 8403
Subject: Re: what is metastability time of a flip_flop
From: Chuck Parsons <chuck@CatenaryScientific.com>
Date: Fri, 12 Dec 1997 17:44:03 -0500
Links: << >>  << T >>  << A >>


John Woodgate wrote:

  [ Poorly worded statement that expected magnitude of integrals of
    random variables with symmetric distributions  about zero is usually
    proportional to the square root of the integration length deleted. ]

> Yes, of course. But you have assumed, AFAICS, in there somewhere a
> non-linearity, either that the pdf of the noise is asymmetrical and the right
> way round to produce a net movement away from metastability (which I
> think is petitio principii) or that the integration somehow involves a
> rectification. I would regard your case as still unproven by argument,
> although I can accept that the effect is real.

  Well that leaves math or actual circuits. But the whole issue is becoming
large and therefore somewhat confused. This is my fault, but probability
in of itself is quite a large subject, and the integral referred to above is not
IMHO trivial by any means even if it should turn out to be mathematically trivial
depending on the particular PDF (Probability Density Function).

   As I read your paragraph, two things I've written are intermeshed. One is
the idea that in general integrating random displacements commonly has
a non-zero expected result. An example is  take a resistor to ground
and place a integrator on the other end which integrates the noise current
onto a capacitor. Now clearly noise currents average to a DC zero.
However the expected magnitude of the output voltage of the integrator
grows proportionally to sqrt(time). This is a classic random walk problem.
The most likely place to be at any later time T is still 0 and the probability
of having a negative or positive result is equal, but the average value
of V(t)^2 will be found to be proportional to t and hence the average
magnitude of V(t) is proportional to sqrt(t).  I brought up this thought
in response to your statement that the high frequency noise seems to
do nothing. I intended to point out that very likely it will cause the
initial voltage to wander in time irrespective of the nonlinear behavior
in the system or for that matter "diode action".  This is by no means certain
with out nailing down the specifics of the problem, but it is the "usual" result.

  So far in this discussion neither the circuit or the noise has been fully
specified
so its not possible to complete the math. For example in the above resistor noise
integrator problem if we add a large capacitor in series with the integrator
then the sqrt(t) behavior will only be valid for small times compared to RC.

   Now the nonlinear response function in this problem works in a manner opposite
to the capacitor added above it causes a small divergence to blow up, instead of
pushing them back to zero. This is a separate point from the one above which
considers
the simplest case with no push one way or the other.

   I think it is time to resort to hard measurements or the literature. I am
seriously
considering building a test circuit.

Chuck


Article: 8404
Subject: Re: what is metastability time of a flip_flop
From: Chuck Parsons <chuck@CatenaryScientific.com>
Date: Fri, 12 Dec 1997 17:45:32 -0500
Links: << >>  << T >>  << A >>


John Woodgate wrote:

  [ Poorly worded statement that expected magnitude of integrals of
    random variables with symmetric distributions  about zero is usually
    proportional to the square root of the integration length deleted. ]

> Yes, of course. But you have assumed, AFAICS, in there somewhere a
> non-linearity, either that the pdf of the noise is asymmetrical and the right
> way round to produce a net movement away from metastability (which I
> think is petitio principii) or that the integration somehow involves a
> rectification. I would regard your case as still unproven by argument,
> although I can accept that the effect is real.

  Well that leaves math or actual circuits. But the whole issue is becoming
large and therefore somewhat confused. This is my fault, but probability
in of itself is quite a large subject, and the integral referred to above is not
IMHO trivial by any means even if it should turn out to be mathematically trivial
depending on the particular PDF (Probability Density Function).

   As I read your paragraph, two things I've written are intermeshed. One is
the idea that in general integrating random displacements commonly has
a non-zero expected result. An example is  take a resistor to ground
and place a integrator on the other end which integrates the noise current
onto a capacitor. Now clearly noise currents average to a DC zero.
However the expected magnitude of the output voltage of the integrator
grows proportionally to sqrt(time). This is a classic random walk problem.
The most likely place to be at any later time T is still 0 and the probability
of having a negative or positive result is equal, but the average value
of V(t)^2 will be found to be proportional to t and hence the average
magnitude of V(t) is proportional to sqrt(t).  I brought up this thought
in response to your statement that the high frequency noise seems to
do nothing. I intended to point out that very likely it will cause the
initial voltage to wander in time irrespective of the nonlinear behavior
in the system or for that matter "diode action".  This is by no means certain
with out nailing down the specifics of the problem, but it is the "usual" result.

  So far in this discussion neither the circuit or the noise has been fully
specified
so its not possible to complete the math. For example in the above resistor noise
integrator problem if we add a large capacitor in series with the integrator
then the sqrt(t) behavior will only be valid for small times compared to RC.

   Now the nonlinear response function in this problem works in a manner opposite
to the capacitor added above it causes a small divergence to blow up, instead of
pushing them back to zero. This is a separate point from the one above which
considers
the simplest case with no push one way or the other.

   I think it is time to resort to hard measurements or the literature. I am
seriously
considering building a test circuit.

Chuck


Article: 8405
Subject: Re: what is metastability time of a flip_flop
From: Chuck Parsons <chuck@CatenaryScientific.com>
Date: Fri, 12 Dec 1997 17:46:33 -0500
Links: << >>  << T >>  << A >>


John Woodgate wrote:

  [ Poorly worded statement that expected magnitude of integrals of
    random variables with symmetric distributions  about zero is usually
    proportional to the square root of the integration length deleted. ]

> Yes, of course. But you have assumed, AFAICS, in there somewhere a
> non-linearity, either that the pdf of the noise is asymmetrical and the right
> way round to produce a net movement away from metastability (which I
> think is petitio principii) or that the integration somehow involves a
> rectification. I would regard your case as still unproven by argument,
> although I can accept that the effect is real.

  Well that leaves math or actual circuits. But the whole issue is becoming
large and therefore somewhat confused. This is my fault, but probability
in of itself is quite a large subject, and the integral referred to above is not
IMHO trivial by any means even if it should turn out to be mathematically trivial
depending on the particular PDF (Probability Density Function).

   As I read your paragraph, two things I've written are intermeshed. One is
the idea that in general integrating random displacements commonly has
a non-zero expected result. An example is  take a resistor to ground
and place a integrator on the other end which integrates the noise current
onto a capacitor. Now clearly noise currents average to a DC zero.
However the expected magnitude of the output voltage of the integrator
grows proportionally to sqrt(time). This is a classic random walk problem.
The most likely place to be at any later time T is still 0 and the probability
of having a negative or positive result is equal, but the average value
of V(t)^2 will be found to be proportional to t and hence the average
magnitude of V(t) is proportional to sqrt(t).  I brought up this thought
in response to your statement that the high frequency noise seems to
do nothing. I intended to point out that very likely it will cause the
initial voltage to wander in time irrespective of the nonlinear behavior
in the system or for that matter "diode action".  This is by no means certain
with out nailing down the specifics of the problem, but it is the "usual" result.

  So far in this discussion neither the circuit or the noise has been fully
specified
so its not possible to complete the math. For example in the above resistor noise
integrator problem if we add a large capacitor in series with the integrator
then the sqrt(t) behavior will only be valid for small times compared to RC.

   Now the nonlinear response function in this problem works in a manner opposite
to the capacitor added above it causes a small divergence to blow up, instead of
pushing them back to zero. This is a separate point from the one above which
considers
the simplest case with no push one way or the other.

   I think it is time to resort to hard measurements or the literature. I am
seriously
considering building a test circuit.

Chuck


Article: 8406
Subject: Re: what is metastability time of a flip_flop
From: Peter Alfke <peter@xilinx.com>
Date: Fri, 12 Dec 1997 15:00:30 -0800
Links: << >>  << T >>  << A >>
rk wrote:

> here are some calcuations i did using chip express cx2001 technology,
> based
> on their flip-flop parameters and example in the CX Technology Design
> Manual, so i could see how this technology performs.  The CX2001
> series
> uses a channelled module architecture (gate array) with each module
> consisting of three 2:1 muxes and an AND gate (a bit differently set
> up
> than Act 1 but not all that dissimilar); there are no hardwired
> flip-flops,
> as there is in the xilinx devices which peter a. previously talked
> about.
> I assumed a 50 MHz clock, a 10 MHz average incoming data rate, and
> made the
> extra settling time a paramter.  here's what i got:
>
>         extra delay               mtbf               mtbf
>         (nsec)           (years)                 (years)
>
>         1.0000e+0               448.25e-6               14.214e-12
>         2.0000e+0               180.84e-3               5.7344e-9
>         3.0000e+0               72.956e+0               2.3134e-6
>         4.0000e+0               29.432e+3               933.29e-6
>         5.0000e+0               11.874e+6               376.52e-3
>         6.0000e+0               4.7903e+9               151.90e+0
>         7.0000e+0               1.9325e+12              61.280e+3
>  

Correction: The center column must be mtbf in seconds, not years.

My opinion:
Interesting results, dramatically showing the superiority of dedicated,
tightly-designed and laid out flip-flops.
Under the above conditions with the product of the two frequencies 500
times higher than in the Xilinx data book ( page 13-43  ) , just 2 ns of
extra allowable resolution time give an MTBF of about 20,000 years in
the 2-year old XC4005E-3. The gate array with its, -excuse the harsh
word-, "spread-out" flip-flops, takes almost 7 ns to achieve the same
MTBF.

There is no substitute for high gain-bandwidth in the feedback path of
the master latch, if you want to resolve metastability reasonably fast.

Peter Alfke, Xilinx Applications
 

Article: 8407
Subject: Re: what is metastability time of a flip_flop
From: John Woodgate <jmw@jmwa.demon.co.uk>
Date: Fri, 12 Dec 1997 23:05:09 +0000
Links: << >>  << T >>  << A >>
In article <3491BE33.36C5A13B@CatenaryScientific.com>, Chuck Parsons
<chuck@CatenaryScientific.com> writes
>
>
>John Woodgate wrote:
>
>  [ Poorly worded statement that expected magnitude of integrals of
>    random variables with symmetric distributions  about zero is usually
>    proportional to the square root of the integration length deleted. ]
>
>> Yes, of course. But you have assumed, AFAICS, in there somewhere a
>> non-linearity, either that the pdf of the noise is asymmetrical and the right
>> way round to produce a net movement away from metastability (which I
>> think is petitio principii) or that the integration somehow involves a
>> rectification. I would regard your case as still unproven by argument,
>> although I can accept that the effect is real.
>
>  Well that leaves math or actual circuits. But the whole issue is becoming
>large and therefore somewhat confused. This is my fault, but probability
>in of itself is quite a large subject, and the integral referred to above is not
>IMHO trivial by any means even if it should turn out to be mathematically 
>trivial
>depending on the particular PDF (Probability Density Function).
>
>   As I read your paragraph, two things I've written are intermeshed. One is
>the idea that in general integrating random displacements commonly has
>a non-zero expected result. An example is  take a resistor to ground
>and place a integrator on the other end which integrates the noise current
>onto a capacitor. Now clearly noise currents average to a DC zero.
>However the expected magnitude of the output voltage of the integrator
>grows proportionally to sqrt(time). This is a classic random walk problem.
>The most likely place to be at any later time T is still 0 and the probability
>of having a negative or positive result is equal, but the average value
>of V(t)^2 will be found to be proportional to t and hence the average
>magnitude of V(t) is proportional to sqrt(t). 

No, I can't go along with that. The circuit you describe is
(theoretically, anyway) linear, and cannot produce what is effectively a
growing d.c. voltage from the noise. Your math reasoning does not take
into account that the sqrt function is two-valued. What grows
proprtionally to sqrt(t) is not sqrt(v^2), which can be either positive
or negative, but |sqrt(V^2)|, which is positive.

> I brought up this thought
>in response to your statement that the high frequency noise seems to
>do nothing. I intended to point out that very likely it will cause the
>initial voltage to wander in time irrespective of the nonlinear behavior
>in the system or for that matter "diode action".  This is by no means certain
>with out nailing down the specifics of the problem, but it is the "usual" 
>result.

I suppose because there is always some non-linearity, and since the
metastable condition is  (theoretically) a single point in the circuit's
state-space, whichever way the bias is, it moves the conditions away
from the metastable point.
>
>  So far in this discussion neither the circuit or the noise has been fully
>specified
>so its not possible to complete the math. For example in the above resistor 
>noise
>integrator problem if we add a large capacitor in series with the integrator
>then the sqrt(t) behavior will only be valid for small times compared to RC.
>
>   Now the nonlinear response function in this problem works in a manner 
>opposite
>to the capacitor added above it causes a small divergence to blow up, instead of
>pushing them back to zero. This is a separate point from the one above which
>considers
>the simplest case with no push one way or the other.
>
>   I think it is time to resort to hard measurements or the literature. I am
>seriously
>considering building a test circuit.
>
>Chuck
>
>
Good idea!
-- 
Regards, John Woodgate, Phone +44 (0)1268 747839 Fax +44 (0)1268 777124. 
OOO - Own Opinions Only. It is useless to threaten a strong man - he will
ignore you. It is dangerous to threaten a weak man - he will kill you if he can.
Article: 8408
Subject: dynamic power in Xilinx designs
From: gah@u.washington.edu (G. Herrmannsfeldt)
Date: 12 Dec 1997 23:33:46 GMT
Links: << >>  << T >>  << A >>
I am working on a design for an XC4013 with 30 16bit adders, plus
lots of latches, in an XC4013.  It should be able to go up to about
30MHz with the 4013-2, but I am wondering about the dynamic power.
If on the average a signal changes every other clock cycle, 
how much power should I expect out of this?

I don't want a big heatsink and fan for each chip!

I tried to find this in "the programmable logic data book," but
I didn't find it.  I only need rough numbers right now, but I don't
even have that,

thanks,

-- glen
Article: 8409
Subject: Re: what is metastability time of a flip_flop
From: "rk" <stellare@erols.com>
Date: 13 Dec 1997 00:59:47 GMT
Links: << >>  << T >>  << A >>
Peter Alfke <peter@xilinx.com> wrote in article
<3491C135.D65118EF@xilinx.com>...
> rk wrote:
> 
> > here are some calcuations i did using chip express cx2001 technology,
> > based

<snip>

> >
> >         extra delay               mtbf               mtbf
> >         (nsec)           (years)                 (years)
> >
> >         1.0000e+0               448.25e-6               14.214e-12
> >         2.0000e+0               180.84e-3               5.7344e-9
> >  

<snip>

> 
> Correction: The center column must be mtbf in seconds, not years.

**** yup, good eye-balls pete.  i cut and pasted the numbers but the titles
**** didn't come over (sigma splot) so i did those by hand.  i was sort of
**** paranoid and checked my numbers vs. chipx' for the case they did,
**** checked each row for about the right divider from sec -> years, but
**** missed the obvious.

> My opinion:
> Interesting results, dramatically showing the superiority of dedicated,
> tightly-designed and laid out flip-flops.

**** i thought it would make a nice comparison and hope from the
description
**** it was plain for all that it was a 'routed' flip-flop design, not a
**** hardwired one.

> Under the above conditions with the product of the two frequencies 500
> times higher than in the Xilinx data book ( page 13-43  ) , just 2 ns of
> extra allowable resolution time give an MTBF of about 20,000 years in
> the 2-year old XC4005E-3. The gate array with its, -excuse the harsh
> word-, "spread-out" flip-flops, takes almost 7 ns to achieve the same
> MTBF.

**** no problem, don't consider it harsh, that's just the way it is.  i
**** call it a 'routed flip-flop' vs. a 'hard-wired flip-flop.'  now, i'm
**** not sure if there are official terms for these by those in the chip
**** business.  if so, please let us now so we all use consistent and 
**** correct language.

> 
> There is no substitute for high gain-bandwidth in the feedback path of
> the master latch, if you want to resolve metastability reasonably fast.

**** yup, going through the routing network slows things down and it shows
**** in this parameter.  
****
**** an interesting aside, however, is some simulations on the performance
**** of routed vs. hard-wired flip-flops in the Actel RH1280 and the Act 3
**** devices (either A1460A or A14100A), where you can select either type
**** of flip-flop.  The results showed that for this technology the choice
**** of implementation had a significant effect on tsu, not counting the
**** ability to 'combine' in their s-module, which gives even a better tsu.
**** however, clk -> q performance for the global routed clocks showed no
**** significant difference.  by careful logic design, the slower c-module
**** flip-flops can be used with no loss in system performance. 
additionally, 
**** the slower routed flip-flops generally offer better radiation
performance,
**** so in the aerospace community they still have some advantages.

**** lastly, i think it would be interesting to compare a lot of the
different
**** devices and technologies out there.  so, if anyone wants to email me
the
**** parameters for the device that they are familiar with, i'll compute
the
**** results for some sample cases and post the results.

> Peter Alfke, Xilinx Applications

--------------------------------------------------------------
rk

"there's nothing like real data to screw up a great theory" 
- me (modified from original, slightly more colorful version)
--------------------------------------------------------------

Article: 8410
Subject: JTAG configuration of Xilinx XC4000E FPGAs?
From: "Peter Fenn" <PeteFenn@iafrica.com>
Date: Sat, 13 Dec 1997 04:38:11 +0200
Links: << >>  << T >>  << A >>
JTAG configuration of Xilinx XC4000E FPGAs?

Urgent! I need to do JTAG configuration of a single FPGA (Xilinx XC4013E).
Can anyone advise/supply 3rd-party software to do this?

Xilinx documentation details that it's indeed possible to configure a XC4000
FPGA via its JTAG interface, but Xilinx does not extend their Xchecker
software to support of this. Any suggestions ?

Thanks-

Peter Fenn
--------------------------------------
            "CodeLogic"
Digital & Software Design Services
--------------------------------------
   PeteFenn@iafrica.com
   TEL: (+27 21) 855-1354
   FAX: (+27 21) 855-2807
--------------------------------------
   P.O.Box 5098
   Helderberg, 7135.
   South Africa
--------------------------------------




Article: 8411
Subject: JTAG configuration of Xilinx XC4000E FPGAs?
From: "Peter Fenn" <PeteFenn@iafrica.com>
Date: Sat, 13 Dec 1997 04:38:11 +0200
Links: << >>  << T >>  << A >>
JTAG configuration of Xilinx XC4000E FPGAs?

Urgent! I need to do JTAG configuration of a single FPGA (Xilinx XC4013E).
Can anyone advise/supply 3rd-party software to do this?

Xilinx documentation details that it's indeed possible to configure a XC4000
FPGA via its JTAG interface, but Xilinx does not extend their Xchecker
software to support of this. Any suggestions ?

Thanks-

Peter Fenn
--------------------------------------
            "CodeLogic"
Digital & Software Design Services
--------------------------------------
   PeteFenn@iafrica.com
   TEL: (+27 21) 855-1354
   FAX: (+27 21) 855-2807
--------------------------------------
   P.O.Box 5098
   Helderberg, 7135.
   South Africa
--------------------------------------




Article: 8412
Subject: Re: combinational multipliers
From: Ray Andraka <no_spam_randraka@ids.net>
Date: Fri, 12 Dec 1997 23:12:07 -0500
Links: << >>  << T >>  << A >>
Carmen Baena Oliva wrote:
> 
> I'm trying to obtain a combinational multiplier using a Xilinx FPGA.
> Can anybody give me some references about good structures for the
> multiplier?
> 
> Thanks in advance.
> Carmen

For the 4-LUT based Xilinx FPGA, your best bet is probably to produce a
1x, 2x (1x shifted) and 3x copy of one input.  Then for every 2 bits in
the other input, select one of these or zero to get a 2xN partial
product.  Then add all the partial products together with the
appropriate  alignment of the radix point to obtain the result.  The
adding is done in a tree structure to minimize the prop delays.  I think
the Xilinx DSP LogiCore tools use this approach for their multipliers. 
You can get the tools from Xilinx.

-Ray Andraka, P.E.
President, the Andraka Consulting Group, Inc.
401/884-7930     Fax 401/884-7950
email randraka@ids.net
http://users.ids.net/~randraka
Article: 8413
Subject: Re: bus design in Altera 10K, how to increase speed
From: dbl@hydra1.tyrvos.caltech.edu (Daniel Lang)
Date: 12 Dec 1997 20:19 PST
Links: << >>  << T >>  << A >>
In article <3491bc91.80619134@news.walltech.com>, muzok@pacbell.net writes...
>hi,
>I have a design which uses %55 of a 10K100 and within this design I have 12 8 bit
>registers which need to be selectively loaded into another register, IOW, I have
>an 8 bit bus from which 12 8 bit registers hang. Because 10K doesn't have
>internal tristates, I have a big mux which selects the register. Currently this
>mux is the bottleneck in my design which limits my speed to  11 MHz in a -3 part.
>Any ideas how to increase the speed in this design? The design is in Verilog.
> 
>Are there any ways to get more performance by using the floor plan tool ? I am
>already using FAST synthesis and maximum speed in the compiler options.

I have done my programming in AHDL, not Verilog, but I assume you have a way
to access the LCELL and CASCADE primitives through Verilog.  What I would do
is write out the multiplexor levels explicitly.  Use the LSB select bit
to multiplex from 12 sets of lines to 6 sets with LCELLs on the output.
Repeat this with the 2nd LSB to get down to 3 sets of 8 lines, again with
LCELLs on the output.  Finally, take the 2 MSB select bis along with 2
of the 3 remaining sets of 8 lines and cascade the 8 outputs into the last
8 lcells with the remaining set of 8 lines and select bits.  This assures that
any signal only has to propagate through 3 levels of LCELLs plus a small
extra delay for the cascade.  (I don't recommend trying to get down to 1 level
of LCELLs using the cascade primitive as you may have fitting problems.)

If this does not buy you enough speed, you might be able to convert the middle
level of LCELLs to DFFs (an extra pipeline stage).

Daniel Lang  dbl@hydra0.caltech.edu

Article: 8414
Subject: Re: bus design in Altera 10K, how to increase speed
From: s_clubb@die.spammer.netcomuk.co.uk (Stuart Clubb)
Date: Sat, 13 Dec 1997 14:03:25 GMT
Links: << >>  << T >>  << A >>
On Fri, 12 Dec 1997 22:43:10 GMT, muzok@pacbell.net (muzo) wrote:

>I have a design which uses %55 of a 10K100 and within this design I have 12 8 bit
>registers which need to be selectively loaded into another register, IOW, I have
>an 8 bit bus from which 12 8 bit registers hang. Because 10K doesn't have
>internal tristates, I have a big mux which selects the register. Currently this
>mux is the bottleneck in my design which limits my speed to  11 MHz in a -3 part.

What speed do you need?

>Any ideas how to increase the speed in this design? The design is in Verilog.

So am I right in saying that a 12 to 1 multiplexer, register to
register (hence, including routing) takes 90 ns? As Altera claims a 16
to 1 mux takes 9.4 ns in a -3 speed grade there are three
possibilities:
1
You need a new synthesis tool (which are you using?)
2
There's more than the mux in your critical path. (how many layers of
logic and routing in the path?)
3
Altera isn't up to it, has a "printing error" in it's data sheets, and
you need a different FPGA.

1 & 2 seem more probable.

Stuart
--
For Email remove "die.spammer." from the address
Article: 8415
Subject: Re: what is metastability time of a flip_flop
From: Chuck Parsons <chuck@CatenaryScientific.com>
Date: Sat, 13 Dec 1997 12:03:41 -0500
Links: << >>  << T >>  << A >>
Several times by E-mail I have been asked the following, which I quote from the last
person who asked, without attribution since he E-mailed instead of posting


>>I've been following your arguements and I'm
>>mostly convinced, but I've also seen the arguement
>>that if all it took was noise, then surely the
>>manufacturere would be adding noise.  If you've
>>already addressed this point, sorry, I may
>>have missed it.  There have been a lot of
>>articles about this subject, most, like yours
>>have been interesting.

  This is a good question, and of course the answer is they are!

   Someone in an earlier post stated that things had improved by thirteen orders of
magnitude relative to some old logic. (Sorry wish I could remember who.) Many have
said that increase in gain are responsible for this. Sure increases in gain
definitely help, and decreasing channel length leads to increased gain in a FET but
decreased channel width leads to decreased gain. How
much has gain improved? Someone out there should have hard numbers I don't.

   I don't dispute that that makes a big difference, but I would stipulate that a far
bigger difference is made by the reduced capacitance of the modern circuits. This
increases the slew rate of the circuit (without necessarily improving gain but only
GBW) allowing it to move out of the metastable
region faster for smaller perturbations. In addition, it also has to following
interesting effect.

   The expected noise voltage energy any capacitor will be at least kT/2 where k is
the Boltzmann constant. This leads to the equation 1/2CV^2=kT/2.  We can solve this
for V finding |V|=sqrt(kT/C) Boltzmann's constant has a value of 1.4E-23J/K.  At a
temperature of 300K this is 4.2E-21J. Or
4.2E-21 Farad Volt^2. This leads to:

   |Vnoise| = 65nV/sqrt(C) = 2mV/sqrt( C in femto farads)


  The noise will randomly flucuate positive and negative. The time that his takes to
happen will be determined by the impedance of the network it is connected to and the
effective RC time constant. For a 10 femtofarad capacitor, sitting unconnected we
only have the leakage resistance which might be 10^15 ohms leading to flucuation
times of about a 10 seconds. For a femtofarad capacitor hooked up in a 100 ohm
circuit the fluctuation times would be about 100fS. This is quite a range, and I
bring it up to point out that calculating the noise voltage on a capacitor without
considering bandwidth, is usually not useful.

   Now perhaps someone could post what the typical  gate+interconnect capacitance in
a modern
process are as well as the drive+interconnect resistance.

    For the sake of example I will assume they are something like 100fF and 100 ohm.
This leads to a noise voltage of 200uV and a characteristic fluctuation time of 10ps.
If a older process had 1000fF and 1000ohm ( How high was poly silicon on a longish
run?) this would lead to 63uV of noise and a fluctuation time of 1ns.

    A final important fact is that as the input+stray capacitance goes down leading
to faster fluctuation noise the speed of the circuit goes up as well assuming this is
the dominate load on the output or
that the other loads on the output reduce capacitance at the same rate. So the
circuit is able to
track these faster fluctuations.

   To sum up lower capacitance --> higher noise and faster noise and a circuit that
can track the
faster noise. --> Quicker expulsion from the metastable region.

    Secondly, of course if you just inject noise without increasing slew rate, a
number of bad things
happen, particularly a increase in jitter. But basically, except for the metastable
problem everything I can think of gets worse with increased noise, so one would want
to study the tradeoffs of doing it carefully.

 As a postscript it is enlightening to consider the following post:

>rk wrote:
>>
>>here are some calcuations i did using chip express cx2001 technology,
>> based
>> on their flip-flop parameters and example in the CX Technology Design
>> Manual, so i could see how this technology performs.  The CX2001
>> series
>> uses a channelled module architecture (gate array) with each module
>> consisting of three 2:1 muxes and an AND gate (a bit differently set
>> up
>> than Act 1 but not all that dissimilar); there are no hardwired
>> flip-flops,
>> as there is in the xilinx devices which peter a. previously talked
>> about.
>> I assumed a 50 MHz clock, a 10 MHz average incoming data rate, and
>> made the
>> extra settling time a paramter.  here's what i got:
>>
>>         extra delay               mtbf               mtbf
>>         (nsec)           (years)                 (years)
>>
>>         1.0000e+0               448.25e-6               14.214e-12
>>         2.0000e+0               180.84e-3               5.7344e-9
>>         3.0000e+0               72.956e+0               2.3134e-6
>>         4.0000e+0               29.432e+3               933.29e-6
>>         5.0000e+0               11.874e+6               376.52e-3
>>         6.0000e+0               4.7903e+9               151.90e+0
>>         7.0000e+0               1.9325e+12              61.280e+3
>>

>Correction: The center column must be mtbf in seconds, not years.
>
>My opinion:
>Interesting results, dramatically showing the superiority of dedicated,
>tightly-designed and laid out flip-flops.
>Under the above conditions with the product of the two frequencies 500
>times higher than in the Xilinx data book ( page 13-43  ) , just 2 ns of
>extra allowable resolution time give an MTBF of about 20,000 years in
>the 2-year old XC4005E-3. The gate array with its, -excuse the harsh
>word-, "spread-out" flip-flops, takes almost 7 ns to achieve the same
>MTBF.
>
>There is no substitute for high gain-bandwidth in the feedback path of
>the master latch, if you want to resolve metastability reasonably fast.
>
>Peter Alfke, Xilinx Applications

  I think the increased noise of a low capacitance layout helps too.

Chuck

Article: 8416
Subject: Re: what is metastability time of a flip_flop
From: Chuck Parsons <chuck@CatenaryScientific.com>
Date: Sat, 13 Dec 1997 12:21:35 -0500
Links: << >>  << T >>  << A >>


John Woodgate wrote:

> In article <3491BE33.36C5A13B@CatenaryScientific.com>, Chuck Parsons
> <chuck@CatenaryScientific.com> writes
>



> >   As I read your paragraph, two things I've written are intermeshed. One is
> >the idea that in general integrating random displacements commonly has
> >a non-zero expected result. An example is  take a resistor to ground
> >and place a integrator on the other end which integrates the noise current
> >onto a capacitor. Now clearly noise currents average to a DC zero.
> >However the expected magnitude of the output voltage of the integrator
> >grows proportionally to sqrt(time). This is a classic random walk problem.
> >The most likely place to be at any later time T is still 0 and the probability
> >of having a negative or positive result is equal, but the average value
> >of V(t)^2 will be found to be proportional to t and hence the average
> >magnitude of V(t) is proportional to sqrt(t).
>
> No, I can't go along with that. The circuit you describe is
> (theoretically, anyway) linear, and cannot produce what is effectively a
> growing d.c. voltage from the noise. Your math reasoning does not take
> into account that the sqrt function is two-valued. What grows
> proprtionally to sqrt(t) is not sqrt(v^2), which can be either positive
> or negative, but |sqrt(V^2)|, which is positive.
>

   Yes it does, that is why I said "magnitude of V(t)". In fact, you don't know
whether the output will be positive or negative, but you can calculate that  on
average the _magnitude_  of the difference from the initial voltage grows  with time
as I described. One can for instance calculate the statistically average  time it
takes the integrator to peg against it output range limits. If you use
higher supply limits and double the output range of the the integrator. It will take
on average 4 times as long before the integrator pegs and stops working. If we start
in the middle of the range, and any input bias current is perfectly nulled , 50% of
the time the integrator will peg negative and 50% will peg positive, but  it in both
cases. But if you reset the circuits 1000 times and calculate the average
time to pegging the later circuit will take four times as long on average to peg.

     The situation is quite similar to two school children matching pennies. If they
each start out with 100 pennies and don't cheat. Eventually one or the  other will
run out of pennies. Obviously it is impossible to happen in less than 100 matches.
But in fact it is very unlikely to happen in 400 matches. Mathematically one can
calculate how long on average this takes. I'm sorry, but I don't remember the precise
answer but it is considerably less than 10,000 matches. The probabilistic question
here is: "What is the average number of matches until the difference in wins equals
100". A closely related problem I do remember the answer to is "What is the
probability distribution of winnings and losses after N matches?" The answer is it is
gaussian distributed with a sigma equal to the square root of N. In fact the exact
answer is binomial not gaussian, but for large N the two functions converge very
quickly.  After 10,000 matches the distribution  of expected winnings (losses) will
be almost  perfectly gaussian with a sigma of 100.  After 40,000 matches the
distribution  will have a sigma of 200. If each school child has 200 pennies it takes
4 times as long  on average though. We find that the expected magnitude of the  grows
as the square root of the number of matches. Just as for the integrator I described
were the expected deviation of the output from the starting point (breaking even)
grows with the square root of time.

Article: 8417
Subject: Xilinx Copy Protection
From: Rick Collins <rickman@erols.com>
Date: Sat, 13 Dec 1997 12:42:01 -0500
Links: << >>  << T >>  << A >>
We just bought the Xilinx development software. However we found that
the copy protection works differently from the past. We were expecting
to be able to install the software on several workstations and share a
dongle. But now they protect the licence by keying the software to the
serial number on your diskdrive.

I have been told that this serial number can be changed so that we can
once again let each engineer work at his own desk. Does anyone know how
to do this? Can you point me to a source of this information?


Rick Collins

rickman@erols.com



Article: 8418
Subject: Re: what is metastability time of a flip_flop
From: John Woodgate <jmw@jmwa.demon.co.uk>
Date: Sat, 13 Dec 1997 18:03:48 +0000
Links: << >>  << T >>  << A >>
In article <3492C41F.67A722C5@CatenaryScientific.com>, Chuck Parsons
<chuck@CatenaryScientific.com> writes
>
>   Yes it does, that is why I said "magnitude of V(t)". In fact, you don't know
>whether the output will be positive or negative, but you can calculate that  on
>average the _magnitude_  of the difference from the initial voltage grows  with 
>time
>as I described. One can for instance calculate the statistically average  time 
>it
>takes the integrator to peg against it output range limits. If you use
>higher supply limits and double the output range of the the integrator. It will 
>take
>on average 4 times as long before the integrator pegs and stops working. If we 
>start
>in the middle of the range, and any input bias current is perfectly nulled , 50% 
>of
>the time the integrator will peg negative and 50% will peg positive, but  it in 
>both
>cases. But if you reset the circuits 1000 times and calculate the average
>time to pegging the later circuit will take four times as long on average to 
>peg.

I don't think I've quite grasped why you have brought in this doubling
of the number of pennies.
>
>     The situation is quite similar to two school children matching pennies. If 
>they
>each start out with 100 pennies and don't cheat. Eventually one or the  other 
>will
>run out of pennies. 

Well, that is clearly not inevitable, AFAICS. An infinite exchange can
be envisaged.

>Obviously it is impossible to happen in less than 100 
>matches.
>But in fact it is very unlikely to happen in 400 matches. Mathematically one can
>calculate how long on average this takes. I'm sorry, but I don't remember the 
>precise
>answer but it is considerably less than 10,000 matches. The probabilistic 
>question
>here is: "What is the average number of matches until the difference in wins 
>equals
>100". A closely related problem I do remember the answer to is "What is the
>probability distribution of winnings and losses after N matches?" The answer is 
>it is
>gaussian distributed with a sigma equal to the square root of N. In fact the 
>exact
>answer is binomial not gaussian, but for large N the two functions converge very
>quickly.  After 10,000 matches the distribution  of expected winnings (losses) 
>will
>be almost  perfectly gaussian with a sigma of 100.  After 40,000 matches the
>distribution  will have a sigma of 200. If each school child has 200 pennies it 
>takes
>4 times as long  on average though. We find that the expected magnitude of the  
>grows
>as the square root of the number of matches. Just as for the integrator I 
>described
>were the expected deviation of the output from the starting point (breaking 
>even)
>grows with the square root of time.
>
I won't argue with your reasoning: questions of probability often have
counter-intuitive answers. However, looking at the subject another way,
the resistor/integrator device seems to be able to extract noise energy,
and that means thermal energy, from the resistor and store a net amount
of it in the capacitor. What happens then to the temperature of the
resistor? What is wrong with this crazy result?
-- 
Regards, John Woodgate, Phone +44 (0)1268 747839 Fax +44 (0)1268 777124. 
OOO - Own Opinions Only. It is useless to threaten a strong man - he will
ignore you. It is dangerous to threaten a weak man - he will kill you if he can.
Article: 8419
Subject: Re: JTAG configuration of Xilinx XC4000E FPGAs?
From: "Steven K. Knapp" <sknapp@optimagic.com>
Date: Sat, 13 Dec 1997 12:36:14 -0800
Links: << >>  << T >>  << A >>
See the info on http://www.xilinx.com/techdocs/940.htm.  This, plus the
Xilinx documentation, should get you up to speed.

-----------------------------------------------------------
Steven K. Knapp
OptiMagic, Inc. -- "Great Designs Happen 'OptiMagic'-ally"
E-mail:  sknapp@optimagic.com
   Web:  http://www.optimagic.com
-----------------------------------------------------------

Peter Fenn wrote in message <66ssic$5pm$1@news01.iafrica.com>...
>JTAG configuration of Xilinx XC4000E FPGAs?
>
>Urgent! I need to do JTAG configuration of a single FPGA (Xilinx XC4013E).
>Can anyone advise/supply 3rd-party software to do this?
>
>Xilinx documentation details that it's indeed possible to configure a
XC4000
>FPGA via its JTAG interface, but Xilinx does not extend their Xchecker
>software to support of this. Any suggestions ?
>
>Thanks-
>
>Peter Fenn
>--------------------------------------
>            "CodeLogic"
>Digital & Software Design Services
>--------------------------------------
>   PeteFenn@iafrica.com
>   TEL: (+27 21) 855-1354
>   FAX: (+27 21) 855-2807
>--------------------------------------
>   P.O.Box 5098
>   Helderberg, 7135.
>   South Africa
>--------------------------------------
>
>
>
>


Article: 8420
Subject: Re: JTAG configuration of Xilinx XC4000E FPGAs?
From: "Steven K. Knapp" <sknapp@optimagic.com>
Date: Sat, 13 Dec 1997 12:47:37 -0800
Links: << >>  << T >>  << A >>
The following is from Xilinx' Answer Database at
http://www.xilinx.com/techdocs/940.htm

--- BEGIN ---
General Description:
With the exception of the XC3000 family, it is possible to configure XC4000
fami
ly
devices and XC5200 devices via the boundary scan pins TMS,TCK,TDI, & TDO.
This
solution applies the XC4000(including XC4000EX) family and the XC5200 family
of devices.NOTE: This solution record is the document on configuring an
XC4000
or XC5200 FPGA via the JTAG TAP.  Information in this solution
record superceeds all other documents on FPGA JTAG configuration.Solution 1:
*NOTE*To understand this solution, you must have an understanding of
JTAG/Boundary Scan.  This solution applies to the XC4000 family
and XC5200 family of devices.CONFIGURE - Steps to Follow to configure a
Xilinx
XC4000, or XC5200 via JTAG:
The bitstream format is identical for all configuration
modes. A user can use a .bit file or a .rbt file, depending on
whether the user wants to read a binary file(.bit) or an ascii
file(.rbt). Express mode bitstreams cannot be used in configuring
via boundary scan.
Xilinx also recommends that the mode pins of the device be tied low
before starting the configuration.1. Turn `on' the boundary scan circuitry.
This can be done one of three ways, either via powerup
or via a configured device with boundary scan enabled,
or by pulling the /PROGRAM pin low.
If you want to do this via powerup, then just hold the
INIT pin low when power is turned on. When VCC has
reached VCC(min), then the TAP can be toggled to enter
JTAG instructions. The INIT pin can be held low one of
two ways, either manually or with a pulldown. If you
choose to manually hold the INIT low, then the INIT pin
must be held low until the CONFIGURE instruction is
the current instruction. If you choose a pulldown, use a
pulldown which pulls the INIT pin down to approxi-
mately 0.5V. The pulldown has the merit of holding INIT
low whenever the FPGA is powered-up, and letting the
user `see' an attenuated INIT pin during configuration.
After the FPGA has been configured, if you want to
reconfigure a configured device that has boundary scan
enabled after configuration, then just start toggling theboundary scan TAP
pins.
2. Load the Xilinx Configure instruction into the IR.
The Xilinx Configure instruction is 101(I2 I1 I0). I0 is the
bit shifted in first into the IR.
3. After shifting in the Xilinx CONFIGURE instruction,
make the CONFIGURE instruction the current JTAG
instruction by going to the update-IR state. When TCK
goes low in the update-IR state, the FPGA is now in the
JTAG configuration mode and will start clearing the con-
figuration memory; The CONFIGURE instruction is now
the current instruction, which must be followed by a ris-
ing edge on TCK. If you chose to manually hold the INIT
pin low, then the INIT pin must be held low until the
CONFIGURE instruction is the current instruction.
4. Once the Xilinx CONFIGURE instruction has been
made the current instruction, the user must go to the
run-test/idle state, and remain in the run-test/idle state
until the FPGA has finished clearing it's configurationmemory.
The approximate time it takes to clear an FPGA's con-
figuration memory is: 2 * 1 us * (# of frames per devicebitstream).
When the FPGA has finished clearing it's configuration
memory, the open-collector INIT has gone high impedance.At this point, the
user
should advance to the shift-dr state. Once the TAP is in
the shift-dr state and the INIT pin has been released, clocks
on the TCK pin will be considered configuration clocksfor data and length
count.
5. In the shift-DR state, start shifting in the bitstream. Con-
tinue shifting in the bitstream until DONE has gone high
and the startup sequence has finished.
During the time you are shifting in the bitstream via the
TAP, the configuration pins LDC, HDC, INIT, PRO-
GRAM, DOUT, and DONE all function as they normally
do during non-JTAG configuration. These pins can be
probed by the user, or after completion of configuration, or if
configuration failed, the SAMPLE/PRELOAD instruction can be
used to view these IOB's(except PROGRAM or DONE).
LDC is low during configuration. HDC is high during
configuration. INIT will be high impednace during configuration, but
if a CRC error or frame error is detected, INIT will go
low; If a pulldown is present on INIT then the user must
probe INIT with a meter or scope;With a pulldown(as in step 1)
attached to the INIT pin, the user will see a drop from
approximately 0.5V to OV if INIT drops low to indicate a
data error. PROGRAM can still be used to abort the
configuration process. DOUT and TDO will echo TDI
until the preamble and length count are shifted into TDI;
When the preamble and length count have been shifted
into the FPGA, DOUT will remain high. DONE will go
high when configuration is finished. Until configuration
is finished, DONE will remain low.Some Additional Notes:
(a) It is possible to configure several 4K, and/or 5K
devices in a JTAG chain. But unlike non-JTAG daisy-
chain configuration, this doesn't necessarily mean merging all the
bitstreams into one bitstream. In the case of JTAG con-
figuration of Xilinx devices in a JTAG chain, all devices,
except the one being configured, will be placed in
BYPASS mode. The one device in CONFIGURE will
have its bitstream downloaded to it. After configuring
this device it will be placed in BYPASS, and another
device will be taken out of BYPASS into CONFIGURE.
(b)  If you are configuring a 'long' daisy-chain of
JTAG devices(TDI connected to TDO of the previous device),
the bitstream for the device with the CONFIGURE instruction
may need to have its bitstream modified.
For example, lets say that the a user has the followingdaisy-chain of
devices:
device1---device2-----device3
Device1's TDO pin is connected to device2's TDI pin;
Device2's TDO pin is connected to device3's TDO pin.
The way to configure this chain is to place one
device in CONFIGURE, and the other two in BYPASS. Let's
further assume that device1 and device2 configure in this
way, but device3 never configures.  Specifically, device3's
DONE pin never goes high; The problem is the bitstream
length count.  A possible cause, aside from bitstream
corruption, is that the final value of the length count
computed by the user/software was reached beforethe loading was complete.
There are two solutions. One solution involves just continually
clocking TCK(for about 15 seconds) until DONE goes high.  The other solution
is
to modify the bitstream;Increase the length count by the number
of devices ahead of the device under configuration.
For example, in the testcase above, the user increase
the length count value by 2.  (In a daisy-chain of
devices configuring via boundary scan, devices inBYPASS will supply
the extra 1's needed at the head of the bitstream)
(c) In general for the XC4000 and XC5200, if you are
configuring these devices via JTAG, finish configuring
the device first before executing any other JTAG instruc-
tions; Once configuration through boundary scan is
started, the configuration operation must be finished.
(d) If boundary scan is not included in the design being
configured, then make sure that the release of I/Os is
the last event in the startup sequence.
If boundary scan is not available, the FPGA is config-
ured, and the I/Os are released before the startup
sequence is finished, the FPGA will not respond to
input signals and outputs won't respond at all.
(e) Re-issuing a boundary scan CONFIGURE instruction
after the the clearing of configuration memory will cancel
the configure instruction.
The proper method of re-issuing a CONFIGURE instruction
after the configuration memory is cleared is to issuue
another boundary scan instruction, and the follow it by the
CONFIGURE instruction.
(f) If configuration through boundary scan fails, there are
only two boundary scan instructions available: sample/pre-
load and bypass. If a another reconfiguration wants to be
attempted, then the PROGRAM pin must be pulled low, or
the FPGA must be repowered.
(g) When the CONFIGURE instruction is the current instruction,
clocks on the TCK pin are not considered configuration clocks
until the /INIT pin has gone high impedance, and the TAP is in the
shift-dr state.(h) If the user is attempting to configure a chain of
devices, it is recommended that the user only configure the
chain in all boundary scan mode, or use the non-boundary scan
configuration modes.  It is possible to configure a daisy-chain
of devices, some in boundary scan and some in non-boundary scan
configuration. Configuring in a mixed mode will not necessarily
give the user a continuous boundary scan chain, which may or
may not be a problem for a particular user's applications.
(j) Currently, there is no software to configure a Xilinx FPGA
via the boundary scan pins.  The user must provide this.NOTE: The intention
of
configuration for a daisy-chain was to use either all the
devices in boundary scan, or all the devices in non-boundary
scanconfiguration.
(k) When configuring a chain of Xilinx FPGA's via boundary scan,
this doesn't require merging all the bitstreams into one bitstream, as in
non-boundary scan configuration daisy-chains.  When the FPGA is in
boudary scan configuration, the same configuration circuitry used
for non-boundary scan configuration is used.  So, if a user would like,
it is possible for a user to merge all bitstreams into one bitstream, using
the Prom Formater or makeprom/promgen. In a case where the user wants to
merge the bitstreams into one bitstream, the user should configure
as in note (a) above.  Additionally, the user will have to tie all /INIT
pins together. All DONE pins will also have to be tied together.
NOTE: The intention of configuration
for a daisy-chain was to use either all the devices in boundary scan, or all
the devices in non-boundary scan configuration.

-- END --

-----------------------------------------------------------
Steven K. Knapp
OptiMagic, Inc. -- "Great Designs Happen 'OptiMagic'-ally"
E-mail:  sknapp@optimagic.com
   Web:  http://www.optimagic.com
-----------------------------------------------------------



Article: 8421
Subject: Re: Z80 in FPGA: clockspeed?
From: z80@ds.com (Peter)
Date: Sat, 13 Dec 1997 23:05:46 GMT
Links: << >>  << T >>  << A >>

>I have heard that the code in the article was deliberately incomplete,
>per an agreement with Zilog. They (Zilog) didn't want people to be
>able to reproduce the Z80 core. Can't say that I'd blame them.

Correct. EDN said as much at the time. This was done in VHDL.

This was a pity, since IMHO anyone can just sit down and design a
Z80-opcode-compatible processor and publish it, perfectly legally. I
was told by Zilog that this was what NEC did with the UPD780 - they
got no help whatever from Zilog. And now Toshiba to some
Z80-compatible micros too.

And such a design, published for all, would have been truly useful, to
anyone doing an ASIC and looking for a decent CPU (and I don't mean a
6805 etc) to go inside it.

Regarding the subject header: doing a Z80 in an FPGA would be
unbelievably inefficient and expensive. FPGAs are not efficient for
decoder-rich logic like micros tend to be. And by the time one
implemented all the opcode decoders using cascaded CLBs etc, the speed
would not be that good, either.

Just for a laugh, I recall someone in late 1970s doing a 8080 emulator
using ECL (as a PhD post-grad project), and this ran at 80MHz.


Peter.

Return address is invalid to help stop junk mail.
E-mail replies to z80@digiXYZserve.com but
remove the XYZ.
Article: 8422
Subject: Re: dynamic power in Xilinx designs
From: z80@ds.com (Peter)
Date: Sat, 13 Dec 1997 23:05:47 GMT
Links: << >>  << T >>  << A >>
Whatever anwers you get, you will be able to improve on it by an order
of magnitude if you can gate clocks, or avoid clocking altogether. But
gate clocking is dangerous, unless the design is carefully done to
tolerate skews.

Peter.

Return address is invalid to help stop junk mail.
E-mail replies to z80@digiXYZserve.com but
remove the XYZ.
Article: 8423
Subject: Re: what is metastability time of a flip_flop
From: Chuck Parsons <chuck@CatenaryScientific.com>
Date: Sat, 13 Dec 1997 18:28:32 -0500
Links: << >>  << T >>  << A >>


John Woodgate wrote:

> I won't argue with your reasoning: questions of probability often have
> counter-intuitive answers. However, looking at the subject another way,
> the resistor/integrator device seems to be able to extract noise energy,
> and that means thermal energy, from the resistor and store a net amount
> of it in the capacitor. What happens then to the temperature of the
> resistor? What is wrong with this crazy result?

  Well lets not get sidetracked on that. John, I think you may be trying to exhaust
me ;-) The input node of an amp (as discussed in the thread on Johnson noise) can
have a lower than thermal temperature, than a resistor and yes heat flows from the
resistor to the amp. It is a very small amount similar in magnitude to the thermal
conductivity of a single atom in a tube, because there is only one degree of freedom.
Anyone, wanting to know more about this please ask Kendall ;-)

   But John, you know this energy is not what is being stored on the capacitor. The
energy to charge the capacitor comes from the amplifier power supplies, at least the
vast bulk of it. Just like in an amplifier the energy from the microphone doesn't
wind up in the speakers but drains down the base of a transistor to the supply rail.
For instance as the voltage on the capacitor grows the power being stored on it goes
as V*I which is very different for large and small V. The thermal heat extraction
from the resistor is basically constant. Well it fluctuates based on chance, but it
doesn't care what the voltage on the output of the integrator is.

Article: 8424
Subject: Re: what is metastability time of a flip_flop
From: Chuck Parsons <chuck@CatenaryScientific.com>
Date: Sat, 13 Dec 1997 18:33:17 -0500
Links: << >>  << T >>  << A >>
Left out a important word.

Chuck Parsons wrote:

> John Woodgate wrote:
>
> > I won't argue with your reasoning: questions of probability often have
> > counter-intuitive answers. However, looking at the subject another way,
> > the resistor/integrator device seems to be able to extract noise energy,
> > and that means thermal energy, from the resistor and store a net amount
> > of it in the capacitor. What happens then to the temperature of the
> > resistor? What is wrong with this crazy result?
>
>   Well lets not get sidetracked on that. John, I think you may be trying to exhaust
> me ;-) The input node of an amp (as discussed in the thread on Johnson noise) can
>



> have a lower than thermal **NOISE** temperature, than a resistor and yes heat flows
> from the
>



> resistor to the amp. It is a very small amount similar in magnitude to the thermal
> conductivity of a single atom in a tube, because there is only one degree of freedom.
> Anyone, wanting to know more about this please ask Kendall ;-)
>
>    But John, you know this energy is not what is being stored on the capacitor. The
> energy to charge the capacitor comes from the amplifier power supplies, at least the
> vast bulk of it. Just like in an amplifier the energy from the microphone doesn't
> wind up in the speakers but drains down the base of a transistor to the supply rail.
> For instance as the voltage on the capacitor grows the power being stored on it goes
> as V*I which is very different for large and small V. The thermal heat extraction
> from the resistor is basically constant. Well it fluctuates based on chance, but it
> doesn't care what the voltage on the output of the integrator is.





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