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 25375

Article: 25375
Subject: Re: 3.3/2.5 voltage regulators
From: Austin Lesea <austin.lesea@xilinx.com>
Date: Fri, 08 Sep 2000 13:52:00 -0700
Links: << >>  << T >>  << A >>
Rick,

Floored?  To start up a chip with 210 million transistors?

Spartan2 transistors are smaller than Virtex transistors, and there are
fewer of them, and the engineers (just went a'visiting) improved the design.

The increase in current at -40 C is gone in Spartan2 (different design).

The characterization is proceeding as we speak, so I would need which parts
you want to use (size, package) and I could get you this information ahead
of the regular release.  Please email me at austin@xilinx.com.

For now, the commercial specification for the Virtex parts, would apply to
Spartan2 (< 500 mA), even for Spartan2 I grade parts.

As for rise time, you can't turn anything on any faster than about 1 ms (it
would require many amps to charge the power supply capacitors that fast --
to charge 1000uF to 2.5 V in 1 ms takes 2.5 amps, or I=Cdv/dt), and Spartan2
has an objective of being perfectly OK with 100 ms ramp up time.  Again,
this is something we have to verify (at the same time as the current).

Austin



rickman wrote:

> I must say that I am floored by this information. All the time I have
> been using Xilinx parts, I have never been aware of the high current
> required for startup. I guess I was lucky that the systems I worked on
> had plenty of power for startup.
>
> But now I am designing boards that will have very little extra power
> supply capacity since the low voltages for powering the FPGAs are
> generated on board. I was planning on using 4 of the smaller Spartan II
> FPGAs on my board with a 1 Amp 2.5 Volt power converter. Now I realize
> that I may need as much as 8 Amps to bring these parts up if I can't
> control the output rampup of the power converter.
>
> I am pretty sure that this will be impossible to acheive on the board I
> am designing. If there is no better way to do this, I will be forced to
> stick with the Lucent Orca parts I am currently using.
>
> Any idea of how the Spartan II parts will work? I am assuming that they
> are the same as the current Virtex parts and that the industrial parts
> will need 2 Amps each worse case. Is this correct?
>
> Is there any way to prevent this current surge to the part? Will holding
> the PRGM- pin low prevent this high power draw? Can the parts be
> sequenced to limit the total power draw? Or will I have to add switches
> to the Vdd feeds to turn them on one at a time?
>
> Austin Lesea wrote:
> >
> > Alain,
> >
> > We recently (> 1 year ago) implemented a Power On Ramp Up current
> > specification for all parts.  We have not gone all the way back to the
> > original 4K family, but the data sheet now specifies the current
> > capacity of the power supply required for clean out and startup prior
> > to configuration.
> >
> > I would recommend that 1 amp be allowed for each older part (4K, 4KE).
> >
> > I know the 4KXL, and all subsequent parts are characterized AND
> > TESTED.
> >
> > There are also things you can do which make this start up current
> > worse.  A generally rising voltage, that rises no faster than 2
> > milliseconds, and no slower than 50 milliseconds, and starts from near
> > 0 Vdc (< 300 mV) is always the best way to go.  Starting from a
> > voltage around 450 mV to 700 mV from a previously configured part, or
> > holding INIT to prevent configuration, or passing through the POR trip
> > point and then going below the POR trip point, are common causes of
> > higher currents.
> >
> > In all cases, check the latest website data sheet.  An example is
> > here:
> >
> > http://www.support.xilinx.com/partinfo/ds005.pdf
> >
> > page 2 of 16
> >
> > For parts that have virtually no current requirement at starup, use
> > the 4KXLA, 4KXV, SpartanXL, or contact your sales office and FAE for
> > assitance in selection.
> >
> > Austin Lesea
> > IC Design, Xilinx
> >
> > Alain Cloet wrote:
> >
> > > "Andy Peters n o a o [.] e d u>" <"apeters <"@> wrote in message
> > > news:8p6288$5k7$2@noao.edu...
> > > > Alexandr V Shuvalov wrote:
> > > >
> > > > > Which voltage regulators ICs are commonly used to power supply
> > > Xilinx XL
> > > > > and other low voltage (3.3-2.5v) devices?
> > > >
> > > > I used a National LM3940IS-3.3 to drop a VME 5V supply down to
> > > 3.3V @ 1
> > > > A.
> > > >
> > > Can you power-up several Xilinx-FPGA's with this element ; or is
> > > there a
> > > work-around ?
> > >
> > > We had this problem recently (not completelly solved), and the most
> > > recent
> > > idea is to supply 3 Xilinx FPGA's with one National LM3940 (other
> > > version :
> > > WG ?).
> > >
> > > The element we used before had a fast power-up time which caused the
> > > FPGA
> > > (4013) to take up to 1 A, and the regulator couldn't give 3A (it
> > > went ok a
> > > normal temperature, but in a cold phase at -40C it didn't go).
> > >
> > > Can the LM3940 do the job ?
> > >
> > > TIA,
> > > Alain
>
> --
>
> Rick Collins
>
> rick.collins@XYarius.com
>
> Ignore the reply address. To email me use the above address with the XY
> removed.
>
> Arius - A Signal Processing Solutions Company
> Specializing in DSP and FPGA design
>
> Arius
> 4 King Ave
> Frederick, MD 21701-3110
> 301-682-7772 Voice
> 301-682-7666 FAX
>
> Internet URL http://www.arius.com

Article: 25376
Subject: Re: Numerically-Controlled Crystal Oscillator (NCXO) or
From: Theron Hicks <hicksthe@egr.msu.edu>
Date: Fri, 08 Sep 2000 16:52:23 -0400
Links: << >>  << T >>  << A >>
Look at

http://www.micronetworks.com/

for programmable master clocks (M115 series).  I heard that that part was
being dropped so it may not be available but it sounds like just what you
wanted.  Note that the output levels were PECLl (Positive ECL) but there
are level translators available quite cheaply.  The M115 was a full blown
oscillator and the last time I used one it went for about $175.  Frequency
range was 100MHz to 1000MHz depending on model number.

Nestor wrote:

> Hi.
>
> Does anyone know any manufacturer who fabricates
> numerically-controlled crystal oscillators (NCXO), also known as
> digitally-controlled crystal oscillators (DCXO) which are suitable for
> digital phase-locked loop designs in VHDL and FPGAs?
>
> Although these blocks resemble a numerically-controlled oscillator
> (NCO), they differ in that the NCXO is not oversampled to generate the
> required output signal (an NCO needs to be oversampled by at least
> 8-times in order to have an acceptable low jitter output).  Rather, a
> digital input word is fed to the NCXO and it synthesizes the required
> output frequency using a standard, low-cost crystal oscillator.  The
> output is also a square wave, just like the standard crystal.  In
> general, the NCXO has a narrow tuning range similar to a
> voltage-controlled crystal oscillator (VCXO), e.g. +/-150ppm relative
> to a frequency in the MHz range.
>
> The NCXO technology is fairly recent from what I understand, but
> allows one to replace a circuit composed of a digital-to-analog
> converter (DAC) and a VCXO by one chip that performs the exact same
> task will less design hassles.  The DCXO is ideal for custom-made
> phase-locked loop (PLL) circuits using digital sections that can be
> implemented in VHDL and FPGAs.
>
> Since I haven't been able to find any NCXO manufacturers over the web,
> I am now looking to the knowledgeable engineers, designers and friends
> that frequent these newsgroups for some potential referrals and/or
> links.
>
> Thanks in advance for your help.
>
> Nestor

Article: 25377
Subject: Re: VirtexE availability?
From: "S. Ramirez" <sramirez@deleet.cfl.rr.com>
Date: Fri, 08 Sep 2000 22:11:53 GMT
Links: << >>  << T >>  << A >>
Andy,
     Earlier, I posted a message to the newsgroup that explained how the
disties work and what their motivations are.  A large amount of their time
is spent "registering" parts.
     You are correct in saying that they will jump all over you asking those
silly questions, because they are trying to build a case for registering the
design socket with, of all people, the Xilinx manufacturer's rep (Xilinx
rep)!
     What you should do is get in touch with the Xilinx rep yourself.  This
is the person that the disty will go to to get availability and price
information, so why not go there yourself?  With Virtex-E parts,
availability and other information are scarce, and the disties are forced to
go to the Xilinx rep for the answers.  Tell the rep that you do not want any
disties in on your design.  Explain to him/her that you are not interested
in a bunch of people bothering you "on this one."  They will keep the
information to themselves and not involve the disties.
     The exception to the above is if you need parts that are well
established.  In this case, the disties may have parts on their shelves.  If
they do, great, but if they don't, they will have to go to the rep again.
In either case, you will have the salesperson or the FAE knocking on your
door or calling you up for the same reason -- to gather information in order
to register the part.
    So my advice is, if you don't want the disties to come give you a 2 hour
PowerPoint presentation on why they are the chosen ones, to call the Xilinx
rep yourself regardless of the part.  Let the rep find the parts for you at
the disty warehouse or factory.  That way you will minimize disty PowerPoint
presentations and obnoxious questions.
     By the way, this is true of most semiconductor companies, not just
Xilinx.
-Simon Ramirez, Consultant
 Synchronous Design, Inc.



"Andy Peters n o a o [.] e d u>" <"apeters <"@> wrote in message
news:8pbfj5$16qm$1@noao.edu...
> I really hate to ask silly questions like this, but I hate calling the
> distros because once you tell them you're thinking about considering
> thinking about using a part in a design, they jump all over you because
> they think you're gonna be writing a req for 10,000 parts tomorrow ...
>
> I'm looking at doing a VirtexE design.  XCV50E should be big enough.
> Mainly, I want the LVDS I/O.  Are they available now, or will they be
> available in the next couple of months?
>
> Of course, I could probably do the design with external LVDS parts, but
> this seems like a "neat" solution.
>
> -- andy
> ----------------------------
> 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
>


Article: 25378
Subject: Re: Numerically-Controlled Crystal Oscillator (NCXO) or Digitally-Controlled Crystal Oscillator (DCXO) Designs
From: berd_kalamunda@techemail.com (Rolie Baldock)
Date: Fri, 08 Sep 2000 22:55:59 GMT
Links: << >>  << T >>  << A >>
What frequency range are you interested in and what resolution do you
require?

On Fri, 08 Sep 2000 20:27:24 GMT, nestor@stansync.com (Nestor) wrote:

>Hi.
>
>Does anyone know any manufacturer who fabricates
>numerically-controlled crystal oscillators (NCXO), also known as
>digitally-controlled crystal oscillators (DCXO) which are suitable for
>digital phase-locked loop designs in VHDL and FPGAs?
>
>Although these blocks resemble a numerically-controlled oscillator
>(NCO), they differ in that the NCXO is not oversampled to generate the
>required output signal (an NCO needs to be oversampled by at least
>8-times in order to have an acceptable low jitter output).  Rather, a
>digital input word is fed to the NCXO and it synthesizes the required
>output frequency using a standard, low-cost crystal oscillator.  The
>output is also a square wave, just like the standard crystal.  In
>general, the NCXO has a narrow tuning range similar to a
>voltage-controlled crystal oscillator (VCXO), e.g. +/-150ppm relative
>to a frequency in the MHz range.
>
>The NCXO technology is fairly recent from what I understand, but
>allows one to replace a circuit composed of a digital-to-analog
>converter (DAC) and a VCXO by one chip that performs the exact same
>task will less design hassles.  The DCXO is ideal for custom-made
>phase-locked loop (PLL) circuits using digital sections that can be
>implemented in VHDL and FPGAs.
>
>Since I haven't been able to find any NCXO manufacturers over the web,
>I am now looking to the knowledgeable engineers, designers and friends
>that frequent these newsgroups for some potential referrals and/or
>links.
>
>Thanks in advance for your help.
>
>Nestor
>
>
>

--Rolie Baldock.  email:  <berd_kalamunda@'nospam'techemail.com>
Article: 25379
Subject: How many 4005s (4010s) does it take to make a general purpose CPU?
From: "Tom Kerrigan" <Thomas.Kerrigan@Colorado.EDU>
Date: Fri, 8 Sep 2000 21:27:13 -0600
Links: << >>  << T >>  << A >>
I'm in a senior project class at the University of Colorado. My group has
opted to design a general purpose CPU, and we have easy access to 4005s (and
some 4010s) and the associated Xilinx tools.

My professor told us that our CPU should be 8-bit, with ~20 instructions,
and that such a design would take 3 or 4 FPGAs.

With that in mind, I just visited this web page, and was shocked:
http://www.io.com/~guccione/HW_list.html

One guy says he fit a 32-bit RISC design (among other things) on a single
4010. Another guy says he made a 16-bit RISC design that only takes up 75%
of a 4005.

I'm worried about doing a 16-bit design because my professor thinks the
project would be doomed to fail. But now I'm not so sure anymore. Does
anyone have any comments or suggestions about this stuff?

Please let me know (via e-mail) if you post a reply. Sometimes my news
server is flakey.

Thanks,
Tom Kerrigan


Article: 25380
Subject: Re: 3.3/2.5 voltage regulators
From: rickman <spamgoeshere4@yahoo.com>
Date: Sat, 09 Sep 2000 01:10:50 -0400
Links: << >>  << T >>  << A >>
Now I am really floored to think that it takes 210 million transistors
to make a chip with only 57,906 system gates! The startup current of the
XCV50 industrial grade is rated at 2 Amps just like the XCV1000. And
yes, I am amazed that the startup current of any of these devices is so
large. When the quiesent current is between 50 and 100 mA, I find it
amazing that the device is not designed to come up in a well defined
state that is not creating a near short between power and ground. 

I have never seen this type of warning on any of the other dozens and
hundreds of other chips I have designed with, so I can only assume that
this is unique to the Xilinx configuration process. 


Austin Lesea wrote:
> 
> Rick,
> 
> Floored?  To start up a chip with 210 million transistors?
> 
> Spartan2 transistors are smaller than Virtex transistors, and there are
> fewer of them, and the engineers (just went a'visiting) improved the design.
> 
> The increase in current at -40 C is gone in Spartan2 (different design).
> 
> The characterization is proceeding as we speak, so I would need which parts
> you want to use (size, package) and I could get you this information ahead
> of the regular release.  Please email me at austin@xilinx.com.
> 
> For now, the commercial specification for the Virtex parts, would apply to
> Spartan2 (< 500 mA), even for Spartan2 I grade parts.
> 
> As for rise time, you can't turn anything on any faster than about 1 ms (it
> would require many amps to charge the power supply capacitors that fast --
> to charge 1000uF to 2.5 V in 1 ms takes 2.5 amps, or I=Cdv/dt), and Spartan2
> has an objective of being perfectly OK with 100 ms ramp up time.  Again,
> this is something we have to verify (at the same time as the current).

I don't know where you got your information on my power supply, I don't
think I gave any details. We have a 100 uF capacitor (low ESR) on the
output of the 1 Amp switcher we are using in our current design. But
even so, I don't think it ramps up in 1 mS. I will check my notes on it
when I have a chance. Since I am now aware that this is a critical spec
for our boards, I will pay more attention to it. The FPGAs we are
currently using do require a stated amount of current on powerup (10
mA), but this is only during the time the voltage is crossing the 1 to
1.5 volt region. I assume that this is the time during which the
circuitry is first coming alive as the Vdd exceeds the gate thresholds.
Otherwise I have seen no spec on high startup currents on these chips,
all four of them. 

I am looking at using the XC2S15, 30 in the TQ100 package and the 50 and
100 in the 256 or 456 pin FPBGA. If this information is still
preliminary, can you tell me when it is expected to be released as
standard data? I am not looking to rush into production with these
parts. I will be waiting for them to be fully characterized. 

BTW, you have mentioned the term "clean out" several times in this
thread. Is this the same as clearing the configuration memory? Is this a
reset that is done all at once or are the configuration bits cleared
sequentially? 


> rickman wrote:
> 
> > I must say that I am floored by this information. All the time I have
> > been using Xilinx parts, I have never been aware of the high current
> > required for startup. I guess I was lucky that the systems I worked on
> > had plenty of power for startup.
> >
> > But now I am designing boards that will have very little extra power
> > supply capacity since the low voltages for powering the FPGAs are
> > generated on board. I was planning on using 4 of the smaller Spartan II
> > FPGAs on my board with a 1 Amp 2.5 Volt power converter. Now I realize
> > that I may need as much as 8 Amps to bring these parts up if I can't
> > control the output rampup of the power converter.
> >
> > I am pretty sure that this will be impossible to acheive on the board I
> > am designing. If there is no better way to do this, I will be forced to
> > stick with the Lucent Orca parts I am currently using.
> >
> > Any idea of how the Spartan II parts will work? I am assuming that they
> > are the same as the current Virtex parts and that the industrial parts
> > will need 2 Amps each worse case. Is this correct?
> >
> > Is there any way to prevent this current surge to the part? Will holding
> > the PRGM- pin low prevent this high power draw? Can the parts be
> > sequenced to limit the total power draw? Or will I have to add switches
> > to the Vdd feeds to turn them on one at a time?
> >
> > Austin Lesea wrote:
> > >
> > > Alain,
> > >
> > > We recently (> 1 year ago) implemented a Power On Ramp Up current
> > > specification for all parts.  We have not gone all the way back to the
> > > original 4K family, but the data sheet now specifies the current
> > > capacity of the power supply required for clean out and startup prior
> > > to configuration.
> > >
> > > I would recommend that 1 amp be allowed for each older part (4K, 4KE).
> > >
> > > I know the 4KXL, and all subsequent parts are characterized AND
> > > TESTED.
> > >
> > > There are also things you can do which make this start up current
> > > worse.  A generally rising voltage, that rises no faster than 2
> > > milliseconds, and no slower than 50 milliseconds, and starts from near
> > > 0 Vdc (< 300 mV) is always the best way to go.  Starting from a
> > > voltage around 450 mV to 700 mV from a previously configured part, or
> > > holding INIT to prevent configuration, or passing through the POR trip
> > > point and then going below the POR trip point, are common causes of
> > > higher currents.
> > >
> > > In all cases, check the latest website data sheet.  An example is
> > > here:
> > >
> > > http://www.support.xilinx.com/partinfo/ds005.pdf
> > >
> > > page 2 of 16
> > >
> > > For parts that have virtually no current requirement at starup, use
> > > the 4KXLA, 4KXV, SpartanXL, or contact your sales office and FAE for
> > > assitance in selection.
> > >
> > > Austin Lesea
> > > IC Design, Xilinx
> > >
> > > Alain Cloet wrote:
> > >
> > > > "Andy Peters n o a o [.] e d u>" <"apeters <"@> wrote in message
> > > > news:8p6288$5k7$2@noao.edu...
> > > > > Alexandr V Shuvalov wrote:
> > > > >
> > > > > > Which voltage regulators ICs are commonly used to power supply
> > > > Xilinx XL
> > > > > > and other low voltage (3.3-2.5v) devices?
> > > > >
> > > > > I used a National LM3940IS-3.3 to drop a VME 5V supply down to
> > > > 3.3V @ 1
> > > > > A.
> > > > >
> > > > Can you power-up several Xilinx-FPGA's with this element ; or is
> > > > there a
> > > > work-around ?
> > > >
> > > > We had this problem recently (not completelly solved), and the most
> > > > recent
> > > > idea is to supply 3 Xilinx FPGA's with one National LM3940 (other
> > > > version :
> > > > WG ?).
> > > >
> > > > The element we used before had a fast power-up time which caused the
> > > > FPGA
> > > > (4013) to take up to 1 A, and the regulator couldn't give 3A (it
> > > > went ok a
> > > > normal temperature, but in a cold phase at -40C it didn't go).
> > > >
> > > > Can the LM3940 do the job ?
> > > >
> > > > TIA,
> > > > Alain
> >
> > --
> >
> > Rick Collins
> >
> > rick.collins@XYarius.com
> >
> > Ignore the reply address. To email me use the above address with the XY
> > removed.
> >
> > Arius - A Signal Processing Solutions Company
> > Specializing in DSP and FPGA design
> >
> > Arius
> > 4 King Ave
> > Frederick, MD 21701-3110
> > 301-682-7772 Voice
> > 301-682-7666 FAX
> >
> > Internet URL http://www.arius.com

-- 

Rick Collins

rick.collins@XYarius.com

Ignore the reply address. To email me use the above address with the XY
removed.



Arius - A Signal Processing Solutions Company
Specializing in DSP and FPGA design

Arius
4 King Ave
Frederick, MD 21701-3110
301-682-7772 Voice
301-682-7666 FAX

Internet URL http://www.arius.com
Article: 25381
Subject: Re: 3.3/2.5 voltage regulators
From: rickman <spamgoeshere4@yahoo.com>
Date: Sat, 09 Sep 2000 01:13:08 -0400
Links: << >>  << T >>  << A >>
Austin Lesea wrote:
> 
> Henryk,
> 
> The INIT holdoff warning applies to 4K only.  It does not apply to
> Virtex, and Virtex architecture derivatives.
> 
> I am sorry for the confusion.
> 
> In 4K, holding INIT and preventing clean out does not make the device
> HOT -- it may be that the 4K device is in contention from the Vcc not
> going down below a few hundred millivolts, and then the Vcc returns,
> and the 4K device is in a partially configured state, and drawing
> current.  So the device is already HOT and getting hotter, and INIT
> prevents the clean out.
> 
> Again, Virtex, Virtex E, Spartan2 do not have this behavior.  The
> design is such that the means of contention that were caused by memory
> contents which occurred in 4K do not exist in Virtex.

Can you be a little more specific as to what behaviour you are talking
about? The Virtex data sheet claims it needs up to 2 Amps of current for
startup. So I assume that you are saying that the Virtex does not
continue to draw heavy current when INIT is held low? Is that right? 


> Henryk Cieslak wrote:
> 
> > So, what is the smart method of delaying Virtex configuration? I
> > need to
> > keep Virtex non-configured for an arbitrary time. The config mode
> > can vary -
> > master or slave.
> > Previous data sheets have specified that PROGRAM can be held low to
> > make a
> > delay, but the latest data sheet (2.2) does not mention it - I think
> > it
> > means this method is not recommended.
> > Now you say that keeping INIT low can make the chip consume plenty
> > of
> > current = make hot.
> > What to do? Let it start configuration and wait infinitely for CCLK
> > (slave)
> > or configure with random data until it stops due to checksum error?
> >
> > Henryk Cieslak
> > Becker Elektronic Polska
> >
> > Austin Lesea wrote in message <39B7BF1E.B24BE76F@xilinx.com>...
> > >Alain,
> > >
> > >We recently (> 1 year ago) implemented a Power On Ramp Up current
> > specification
> > >for all parts.  We have not gone all the way back to the original
> > 4K
> > family, but
> > >the data sheet now specifies the current capacity of the power
> > supply
> > required
> > >for clean out and startup prior to configuration.
> > >
> > >I would recommend that 1 amp be allowed for each older part (4K,
> > 4KE).
> > >
> > >I know the 4KXL, and all subsequent parts are characterized AND
> > TESTED.
> > >
> > >There are also things you can do which make this start up current
> > worse.  A
> > >generally rising voltage, that rises no faster than 2 milliseconds,
> > and no
> > >slower than 50 milliseconds, and starts from near 0 Vdc (< 300 mV)
> > is
> > always the
> > >best way to go.  Starting from a voltage around 450 mV to 700 mV
> > from a
> > >previously configured part, or holding INIT to prevent
> > configuration, or
> > passing
> > >through the POR trip point and then going below the POR trip point,
> > are
> > common
> > >causes of higher currents.
> > >
> > >In all cases, check the latest website data sheet.  An example is
> > here:
> > >
> > >http://www.support.xilinx.com/partinfo/ds005.pdf
> > >
> > >page 2 of 16
> > >
> > >For parts that have virtually no current requirement at starup, use
> > the
> > 4KXLA,
> > >4KXV, SpartanXL, or contact your sales office and FAE for assitance
> > in
> > >selection.
> > >
> > >Austin Lesea
> > >IC Design, Xilinx
> > >
> > >Alain Cloet wrote:
> > >
> > >> "Andy Peters n o a o [.] e d u>" <"apeters <"@> wrote in message
> > >> news:8p6288$5k7$2@noao.edu...
> > >> > Alexandr V Shuvalov wrote:
> > >> >
> > >> > > Which voltage regulators ICs are commonly used to power
> > supply Xilinx
> > XL
> > >> > > and other low voltage (3.3-2.5v) devices?
> > >> >
> > >> > I used a National LM3940IS-3.3 to drop a VME 5V supply down to
> > 3.3V @ 1
> > >> > A.
> > >> >
> > >> Can you power-up several Xilinx-FPGA's with this element ; or is
> > there a
> > >> work-around ?
> > >>
> > >> We had this problem recently (not completelly solved), and the
> > most
> > recent
> > >> idea is to supply 3 Xilinx FPGA's with one National LM3940 (other
> > version
> > :
> > >> WG ?).
> > >>
> > >> The element we used before had a fast power-up time which caused
> > the FPGA
> > >> (4013) to take up to 1 A, and the regulator couldn't give 3A (it
> > went ok
> > a
> > >> normal temperature, but in a cold phase at -40C it didn't go).
> > >>
> > >> Can the LM3940 do the job ?
> > >>
> > >> TIA,
> > >> Alain
> > >

-- 

Rick Collins

rick.collins@XYarius.com

Ignore the reply address. To email me use the above address with the XY
removed.



Arius - A Signal Processing Solutions Company
Specializing in DSP and FPGA design

Arius
4 King Ave
Frederick, MD 21701-3110
301-682-7772 Voice
301-682-7666 FAX

Internet URL http://www.arius.com
Article: 25382
Subject: Re: How many 4005s (4010s) does it take to make a general purpose CPU?
From: Ray Andraka <ray@andraka.com>
Date: Sat, 09 Sep 2000 05:24:05 GMT
Links: << >>  << T >>  << A >>
Jan Gray did a series of articles in circuit cellar magazine this past spring on
a risc machine in a small FPGA.  He was specifically targeting the XESS board. 
ANyway I know it can be done.  I did a machine a while back that was kind of a
cross between a PIC and an RCA1802 that occupied 8x10 XC4000 CLBs.  Find Jan's
article series, it'll give you a good head start.

Tom Kerrigan wrote:
> 
> I'm in a senior project class at the University of Colorado. My group has
> opted to design a general purpose CPU, and we have easy access to 4005s (and
> some 4010s) and the associated Xilinx tools.
> 
> My professor told us that our CPU should be 8-bit, with ~20 instructions,
> and that such a design would take 3 or 4 FPGAs.
> 
> With that in mind, I just visited this web page, and was shocked:
> http://www.io.com/~guccione/HW_list.html
> 
> One guy says he fit a 32-bit RISC design (among other things) on a single
> 4010. Another guy says he made a 16-bit RISC design that only takes up 75%
> of a 4005.
> 
> I'm worried about doing a 16-bit design because my professor thinks the
> project would be doomed to fail. But now I'm not so sure anymore. Does
> anyone have any comments or suggestions about this stuff?
> 
> Please let me know (via e-mail) if you post a reply. Sometimes my news
> server is flakey.
> 
> Thanks,
> Tom Kerrigan

-- 
-Ray Andraka, P.E.
President, the Andraka Consulting Group, Inc.
401/884-7930     Fax 401/884-7950
email ray@andraka.com  
http://www.andraka.com  or http://www.fpga-guru.com
Article: 25383
Subject: Re: Numerically-Controlled Crystal Oscillator (NCXO) or
From: rickman <spamgoeshere4@yahoo.com>
Date: Sat, 09 Sep 2000 01:27:32 -0400
Links: << >>  << T >>  << A >>
I believe the VCXO circuit you are describing is a standard crystal
oscillator with a variactor diode to control the frequency. All crystal
circuits can be tuned slightly by varying the capacitance in the
circuit. The variactor diode allows you to use a DC voltage to adjust
this capacitance. 

I am not familiar with the NCXO or DCXO, but I would be willing to bet
that this is just a DAC combined with a VCXO. No magic here, just a
matter of combining mulitple devices in one package. 


Nestor wrote:
> 
> Hi.
> 
> Does anyone know any manufacturer who fabricates
> numerically-controlled crystal oscillators (NCXO), also known as
> digitally-controlled crystal oscillators (DCXO) which are suitable for
> digital phase-locked loop designs in VHDL and FPGAs?
> 
> Although these blocks resemble a numerically-controlled oscillator
> (NCO), they differ in that the NCXO is not oversampled to generate the
> required output signal (an NCO needs to be oversampled by at least
> 8-times in order to have an acceptable low jitter output).  Rather, a
> digital input word is fed to the NCXO and it synthesizes the required
> output frequency using a standard, low-cost crystal oscillator.  The
> output is also a square wave, just like the standard crystal.  In
> general, the NCXO has a narrow tuning range similar to a
> voltage-controlled crystal oscillator (VCXO), e.g. +/-150ppm relative
> to a frequency in the MHz range.
> 
> The NCXO technology is fairly recent from what I understand, but
> allows one to replace a circuit composed of a digital-to-analog
> converter (DAC) and a VCXO by one chip that performs the exact same
> task will less design hassles.  The DCXO is ideal for custom-made
> phase-locked loop (PLL) circuits using digital sections that can be
> implemented in VHDL and FPGAs.
> 
> Since I haven't been able to find any NCXO manufacturers over the web,
> I am now looking to the knowledgeable engineers, designers and friends
> that frequent these newsgroups for some potential referrals and/or
> links.
> 
> Thanks in advance for your help.
> 
> Nestor

-- 

Rick Collins

rick.collins@XYarius.com

Ignore the reply address. To email me use the above address with the XY
removed.



Arius - A Signal Processing Solutions Company
Specializing in DSP and FPGA design

Arius
4 King Ave
Frederick, MD 21701-3110
301-682-7772 Voice
301-682-7666 FAX

Internet URL http://www.arius.com
Article: 25384
Subject: Re: How many 4005s (4010s) does it take to make a general purpose CPU?
From: rickman <spamgoeshere4@yahoo.com>
Date: Sat, 09 Sep 2000 01:32:25 -0400
Links: << >>  << T >>  << A >>
Aside from the larger number of gates required and the slower paths for
carry propagation, and signal routing, you should not have any more
difficulty in designing a 16 bit CPU compared to an 8 bit one. The logic
is identical and the circuitry is the same except that the paths have to
be 16 bits wide. 

In what ways do you expect a 16 bit CPU to be harder to build than an 8
bit one? Of course if you make your instructions 16 bits wide and make
them more complex, the 16 bit machine will be more difficult to design.
But if you still use 8 bit instructions and the same level of complexity
in the instruction encoding you will see little difference. 


Tom Kerrigan wrote:
> 
> I'm in a senior project class at the University of Colorado. My group has
> opted to design a general purpose CPU, and we have easy access to 4005s (and
> some 4010s) and the associated Xilinx tools.
> 
> My professor told us that our CPU should be 8-bit, with ~20 instructions,
> and that such a design would take 3 or 4 FPGAs.
> 
> With that in mind, I just visited this web page, and was shocked:
> http://www.io.com/~guccione/HW_list.html
> 
> One guy says he fit a 32-bit RISC design (among other things) on a single
> 4010. Another guy says he made a 16-bit RISC design that only takes up 75%
> of a 4005.
> 
> I'm worried about doing a 16-bit design because my professor thinks the
> project would be doomed to fail. But now I'm not so sure anymore. Does
> anyone have any comments or suggestions about this stuff?
> 
> Please let me know (via e-mail) if you post a reply. Sometimes my news
> server is flakey.
> 
> Thanks,
> Tom Kerrigan

-- 

Rick Collins

rick.collins@XYarius.com

Ignore the reply address. To email me use the above address with the XY
removed.



Arius - A Signal Processing Solutions Company
Specializing in DSP and FPGA design

Arius
4 King Ave
Frederick, MD 21701-3110
301-682-7772 Voice
301-682-7666 FAX

Internet URL http://www.arius.com
Article: 25385
Subject: Re: How many 4005s (4010s) does it take to make a general purpose CPU?
From: "Tom Kerrigan" <Thomas.Kerrigan@Colorado.EDU>
Date: Fri, 8 Sep 2000 23:45:18 -0600
Links: << >>  << T >>  << A >>
I realize that it's not much harder to design a 16-bit CPU than an 8-bit
one.

The question is whether or not it will still fit on one FPGA, or how many
more FPGAs it requires.

If the 8-bit CPU fits on a single FPGA and the 16-bit one doesn't, then
that's a serious problem, because we will have to wire wrap the FPGAs
together, and wire wrapping a 16-bit bus sounds like more work than it needs
to be.

-Tom

rickman <spamgoeshere4@yahoo.com> wrote in message
news:39B9CB69.47BD79C2@yahoo.com...
> Aside from the larger number of gates required and the slower paths for
> carry propagation, and signal routing, you should not have any more
> difficulty in designing a 16 bit CPU compared to an 8 bit one. The logic
> is identical and the circuitry is the same except that the paths have to
> be 16 bits wide.
>
> In what ways do you expect a 16 bit CPU to be harder to build than an 8
> bit one? Of course if you make your instructions 16 bits wide and make
> them more complex, the 16 bit machine will be more difficult to design.
> But if you still use 8 bit instructions and the same level of complexity
> in the instruction encoding you will see little difference.
>
>
> Tom Kerrigan wrote:
> >
> > I'm in a senior project class at the University of Colorado. My group
has
> > opted to design a general purpose CPU, and we have easy access to 4005s
(and
> > some 4010s) and the associated Xilinx tools.
> >
> > My professor told us that our CPU should be 8-bit, with ~20
instructions,
> > and that such a design would take 3 or 4 FPGAs.
> >
> > With that in mind, I just visited this web page, and was shocked:
> > http://www.io.com/~guccione/HW_list.html
> >
> > One guy says he fit a 32-bit RISC design (among other things) on a
single
> > 4010. Another guy says he made a 16-bit RISC design that only takes up
75%
> > of a 4005.
> >
> > I'm worried about doing a 16-bit design because my professor thinks the
> > project would be doomed to fail. But now I'm not so sure anymore. Does
> > anyone have any comments or suggestions about this stuff?
> >
> > Please let me know (via e-mail) if you post a reply. Sometimes my news
> > server is flakey.
> >
> > Thanks,
> > Tom Kerrigan
>
> --
>
> Rick Collins
>
> rick.collins@XYarius.com
>
> Ignore the reply address. To email me use the above address with the XY
> removed.
>
>
>
> Arius - A Signal Processing Solutions Company
> Specializing in DSP and FPGA design
>
> Arius
> 4 King Ave
> Frederick, MD 21701-3110
> 301-682-7772 Voice
> 301-682-7666 FAX
>
> Internet URL http://www.arius.com


Article: 25386
Subject: Re: How many 4005s (4010s) does it take to make a general purpose CPU?
From: "Jan Gray" <jsgray@acm.org>
Date: Sat, 09 Sep 2000 06:32:25 GMT
Links: << >>  << T >>  << A >>
"Tom Kerrigan" <Thomas.Kerrigan@Colorado.EDU> wrote in message
news:8pcads$ime$1@peabody.colorado.edu...
> I'm in a senior project class at the University of Colorado. My group has
> opted to design a general purpose CPU, and we have easy access to 4005s
(and
> some 4010s) and the associated Xilinx tools.
>
> My professor told us that our CPU should be 8-bit, with ~20 instructions,
> and that such a design would take 3 or 4 FPGAs.
>
> With that in mind, I just visited this web page, and was shocked:
> http://www.io.com/~guccione/HW_list.html
>
> One guy says he fit a 32-bit RISC design (among other things) on a single
> 4010. Another guy says he made a 16-bit RISC design that only takes up 75%
> of a 4005.

Hi.  I'm that guy with the 32-bit RISC -- in a little more than half of an
XC4010 (see www3.sympatico.ca/jsgray/homebrew.htm).  The 16-bit R16/RISC4005
was the trailblazing work of Philip Freidin.

This spring Circuit Cellar magazine ran my three article series
demonstrating how to implement a 16-bit pipelined RISC and system-on-a-chip
(video) in an XC4005XL.  The sources are available, in Foundation schematics
and synthesizable Verilog, C compiler too, see www.fpgacpu.org/xsoc/cc.html
and www.fpgacpu.org/xsoc/index.html.  There's also a mailing list,
www.egroups.com/group/fpga-cpu/info.html, where we supposedly discuss FPGA
CPU and SoC design, and also links to other FPGA CPUs at
www.fpgacpu.org/links.html.

A simple non-pipelined 8-bit processor will make a splendid senior project,
and you shouldn't have too much trouble fitting one into an XC4005 or
XC4010, if you apply Occam's Razor.  See for example the 4- and 8-bit
processor designs in the last chapter of Dave Vanden Bout's Practical Xilinx
Designer Lab Book (part of the Xilinx Student Ed. product).

Some thoughts on the application of FPGA CPUs to teaching are at
www.fpgapcu.org/teaching.html.  I'd love to hear feedback from educators.

Jan Gray
Gray Research LLC



Article: 25387
Subject: Re: How many 4005s (4010s) does it take to make a general purpose CPU?
From: nweaver@boom.CS.Berkeley.EDU (Nicholas C. Weaver)
Date: 9 Sep 2000 07:00:02 GMT
Links: << >>  << T >>  << A >>

In article <8pcads$ime$1@peabody.colorado.edu>,
Tom Kerrigan <Thomas.Kerrigan@Colorado.EDU> wrote:
>I'm in a senior project class at the University of Colorado. My group has
>opted to design a general purpose CPU, and we have easy access to 4005s (and
>some 4010s) and the associated Xilinx tools.
>
>My professor told us that our CPU should be 8-bit, with ~20 instructions,
>and that such a design would take 3 or 4 FPGAs.

	Nah, an 8 bit, 20 instruction CPU should easily fit in a
4005.  There really isn't much there, if you are cleaver.  Heck,
designing a SIMPLE N-bit CPU is pretty trivial, it was the final exam
question in our digital design class, and an easy final it was, too.

	But some hints:

	Make the design multicycle, don't bother with pipelining.
Implement a single wide mux using the tristate lines, to basically
handle routing all the data around.  It's suprising how much of a CPU
design, especially a simple CPU design, is muxes, muxes and more
muxes.  FPGAs are GOOD at muxes, if you read the data sheet to know
which style to use when.  

	A one-hot state machine for your control logic, makes it
simple to design and debug, and pretty efficient on a Xilinx to boot,
and there should be NO problem at all.

	However, it may be reasonable to do a 16 bit processor in that
space, I'd advise against it.  The key to making a class project work
well is KISS:  Keep it Simple and Stupid.  Do the minimum spec
required to get a good grade.

	Only AFTER it is done, should you try playing with the design
to get more performance.  I'd suggest pipelining, it's the easiest and
can easily get you a 3-5x performance increase with only a minor area
cost.

	Oh, just make sure, these are 4000Es right?  It is easier with
the Es to do the register file, because of the synchronous memories
(knowing universitys, it could be very well classic 4000s instead of
the Es).
-- 
Nicholas C. Weaver                                 nweaver@cs.berkeley.edu
Article: 25388
Subject: Re: How many 4005s (4010s) does it take to make a general purpose CPU?
From: "Jan Gray" <jsgray@acm.org>
Date: Sat, 09 Sep 2000 07:10:04 GMT
Links: << >>  << T >>  << A >>
"Tom Kerrigan" <Thomas.Kerrigan@Colorado.EDU> wrote in message
news:8pcigr$n7m$1@peabody.colorado.edu...
> I realize that it's not much harder to design a 16-bit CPU than an 8-bit
> one.
>
> The question is whether or not it will still fit on one FPGA, or how many
> more FPGAs it requires.
>
> If the 8-bit CPU fits on a single FPGA and the 16-bit one doesn't, then
> that's a serious problem, because we will have to wire wrap the FPGAs
> together, and wire wrapping a 16-bit bus sounds like more work than it
needs
> to be.

It is difficult to generalize.  You haven't stated whether you are building
a simple multi-cycle CPU, or a one-cycle CPU, or a pipelined CPU, etc.
Assuming a simple design, you should not have too much trouble building a
16-bit CPU in a 4010.  It may well fit in a 4005 but you may have to put
more effort into designing the processor to take advantage of the available
device features.  For example, R16, j32, xr16, etc. all use TBUFs (on-chip
3-state buffers), instead of logic, to implement a wide, many-input result
multiplexer, in order to conserve gates.

How can you tell if a design is feasible before implementing it and
discovering it doesn't fit?  Part of the "art" of FPGA design is to prepare
a sketch of a datapath and/or control logic, and then estimate approximately
how many LUTs (or CLBs) the sketch will require. This estimate can be used
to determine which size of part will be required and/or to determine that
the current design won't fit in the specified part, indicating necessary
rework of the implementation sketch or of the design requirements.

So sketch your proposed 16-bit datapath on paper.  Add up resources
required: every bit of adder, or logic, or mux, or RAM, is a LUT (lookup
table), and every bit of every register is a flip-flop.  Now you can quickly
derive an estimate of resources required and/or feasibility.  In your case,

  probably-infeasible if max(#LUTs, #flops) > f*14*14*2 (device=XC4005)   or
> f*20*20*2 (XC4010).

Here let f=0.75 or so (you don't want to fill the device past 75% if this is
you first project, and leave yourself some margin of error).

Jan Gray
Gray Research LLC



Article: 25389
Subject: Re: timing simulation vs functional one
From: Pinhas <bknpk@hotmail.com>
Date: Sat, 9 Sep 2000 00:47:01 -0700
Links: << >>  << T >>  << A >>
The problem may occur even if you don&#8217;t have those 0.01 ns delays. For flip-flops sharing the same clock domain the clock must come not only in the same time but also in the same delta tick.
For example FFA uses 32MHz and its output drives FFB using 16MHz derived from the 32MHz.
If 32MHz is delayed even by delta ticks, the circuit won&#8217;t work well on functional only simulation (it will however work in functional + timing).
Here is a case where I needed to solve a problem. The designer used clock enable derived from a DLL. The clock enable was in the same delta tick as the clock and flip-flop did not see it on time.
The fix was to add a buffer (delta delay in functional). The buffer was named carefully, because it was needed when the design was converted to ASIC (to fix hold problems for the same very reason).
As for the UNSIM library. I think it is a little scary to touch the library. I would recommend to have a generic that bypasses all XILINX buffer on the clock during functional simulation.

Article: 25390
Subject: Re: VirtexE availability?
From: "Mark Harvey" <mark.harvey@iol.it>
Date: Sat, 09 Sep 2000 11:29:14 GMT
Links: << >>  << T >>  << A >>

S. Ramirez <sramirez@deleet.cfl.rr.com> wrote in message
news:Judu5.13363$98.1346052@typhoon.tampabay.rr.com...
> Andy,
>      Earlier, I posted a message to the newsgroup that explained how the
> disties work and what their motivations are.  A large amount of their time
> is spent "registering" parts.

Because that's the only way that they get lower pricing from Xilinx - it's
what
Xilinx want them to do.


>      You are correct in saying that they will jump all over you asking
those
> silly questions, because they are trying to build a case for registering
the
> design socket with, of all people, the Xilinx manufacturer's rep (Xilinx
> rep)!
>      What you should do is get in touch with the Xilinx rep yourself.
This
> is the person that the disty will go to to get availability and price
> information, so why not go there yourself?  With Virtex-E parts,
> availability and other information are scarce, and the disties are forced
to
> go to the Xilinx rep for the answers.
>Tell the rep that you do not want any
> disties in on your design.  Explain to him/her that you are not interested
> in a bunch of people bothering you "on this one."  They will keep the
> information to themselves and not involve the disties.


If the rep in my area did this, the disties run would screaming to Xilinx,
no rep should ever do this, after all the rep can't sell the parts, only the
disties.


>      The exception to the above is if you need parts that are well
> established.  In this case, the disties may have parts on their shelves.
If
> they do, great, but if they don't, they will have to go to the rep again.

Why, he certainly doesn't have any on HIS shelf!

> In either case, you will have the salesperson or the FAE knocking on your
> door or calling you up for the same reason -- to gather information in
order
> to register the part.
>     So my advice is, if you don't want the disties to come give you a 2
hour
> PowerPoint presentation on why they are the chosen ones, to call the
Xilinx
> rep yourself regardless of the part.  Let the rep find the parts for you
at
> the disty warehouse or factory.

What's the point of asking the rep to ask the disty? ..and if the rep sells
direct from factory, the disty should immediately complain to Xilinx.


  That way you will minimize disty PowerPoint
> presentations and obnoxious questions.

Better just to tell the disty to leave his powerpoint slides at home.



At the end of the day, all Xilinx devices sold thru the disties must
undergo the design registration process, it's better to build a
relationship with a disty that you trust, who gives good tech support &
service and let
him/her register all your designs - that way he/she is guaranteed lower
pricing from the rep & has further incentive to keep giving you support.

Use the design registration as a carrot on a stick to wave in front of the
disty.....no support, no registration.

>      By the way, this is true of most semiconductor companies, not just
> Xilinx.
> -Simon Ramirez, Consultant
>  Synchronous Design, Inc.
>
>
>
> "Andy Peters n o a o [.] e d u>" <"apeters <"@> wrote in message
> news:8pbfj5$16qm$1@noao.edu...
> > I really hate to ask silly questions like this, but I hate calling the
> > distros because once you tell them you're thinking about considering
> > thinking about using a part in a design, they jump all over you because
> > they think you're gonna be writing a req for 10,000 parts tomorrow ...
> >
> > I'm looking at doing a VirtexE design.  XCV50E should be big enough.
> > Mainly, I want the LVDS I/O.  Are they available now, or will they be
> > available in the next couple of months?
> >
> > Of course, I could probably do the design with external LVDS parts, but
> > this seems like a "neat" solution.
> >
> > -- andy
> > ----------------------------
> > 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
> >
>
>


Article: 25391
Subject: Re: bga->dip?
From: mhkohne@discordia.org
Date: Sat, 09 Sep 2000 12:44:32 GMT
Links: << >>  << T >>  << A >>
On Wed, 06 Sep 2000 21:17:25 GMT, uwuh@my-deja.com wrote:
> Hi,
> Is there such a thing as a BGA to DIP converter?
> Thanks
> 
> 
> Sent via Deja.com http://www.deja.com/
> Before you buy.
I sincerely doubt it, but you may find something of use from 
Emulation Technology, inc.
(http://www.emulation.com/)
They make a LOT of adapters for almost every type of package to 
almost every type of package. They do make a BGA emulator adaptor 
that solders down in place of the BGA, then the BGA is soldered to 
the top. Out the sides come pins for logic analysis.

Good luck!

Michael Kohne
mhkohne@discordia.org



Article: 25392
Subject: Re: VirtexE availability?
From: rickman <spamgoeshere4@yahoo.com>
Date: Sat, 09 Sep 2000 10:05:00 -0400
Links: << >>  << T >>  << A >>
Mark,

The relationship between disti and manufacturer is not as well balanced
and even handed as you might think. The manufacturer often takes
accounts direct and cuts the disti out entirely. It is all based on
volume and price pressure. If the volume is high enough and especially
if the manufacture feels too much pressure to reduce prices, they will
cut the disti out and pocket the difference themselves. It is seldom
that the distis get an exclusive sales agreement with the large
manufacturers. 

It is also often a *good* idea to speak directly to the manufacturers
rep as they can give you breaks on pricing that the disti can't, even if
you are ultimately buying through the disti. Of course this only works
if you are buying thousands of units. But even at 1000 pieces I have
gotten better pricing than what the disti would quote me. 


Mark Harvey wrote:
> 
> S. Ramirez <sramirez@deleet.cfl.rr.com> wrote in message
> news:Judu5.13363$98.1346052@typhoon.tampabay.rr.com...
> > Andy,
> >      Earlier, I posted a message to the newsgroup that explained how the
> > disties work and what their motivations are.  A large amount of their time
> > is spent "registering" parts.
> 
> Because that's the only way that they get lower pricing from Xilinx - it's
> what
> Xilinx want them to do.
> 
> >      You are correct in saying that they will jump all over you asking
> those
> > silly questions, because they are trying to build a case for registering
> the
> > design socket with, of all people, the Xilinx manufacturer's rep (Xilinx
> > rep)!
> >      What you should do is get in touch with the Xilinx rep yourself.
> This
> > is the person that the disty will go to to get availability and price
> > information, so why not go there yourself?  With Virtex-E parts,
> > availability and other information are scarce, and the disties are forced
> to
> > go to the Xilinx rep for the answers.
> >Tell the rep that you do not want any
> > disties in on your design.  Explain to him/her that you are not interested
> > in a bunch of people bothering you "on this one."  They will keep the
> > information to themselves and not involve the disties.
> 
> If the rep in my area did this, the disties run would screaming to Xilinx,
> no rep should ever do this, after all the rep can't sell the parts, only the
> disties.
> 
> >      The exception to the above is if you need parts that are well
> > established.  In this case, the disties may have parts on their shelves.
> If
> > they do, great, but if they don't, they will have to go to the rep again.
> 
> Why, he certainly doesn't have any on HIS shelf!
> 
> > In either case, you will have the salesperson or the FAE knocking on your
> > door or calling you up for the same reason -- to gather information in
> order
> > to register the part.
> >     So my advice is, if you don't want the disties to come give you a 2
> hour
> > PowerPoint presentation on why they are the chosen ones, to call the
> Xilinx
> > rep yourself regardless of the part.  Let the rep find the parts for you
> at
> > the disty warehouse or factory.
> 
> What's the point of asking the rep to ask the disty? ..and if the rep sells
> direct from factory, the disty should immediately complain to Xilinx.
> 
>   That way you will minimize disty PowerPoint
> > presentations and obnoxious questions.
> 
> Better just to tell the disty to leave his powerpoint slides at home.
> 
> At the end of the day, all Xilinx devices sold thru the disties must
> undergo the design registration process, it's better to build a
> relationship with a disty that you trust, who gives good tech support &
> service and let
> him/her register all your designs - that way he/she is guaranteed lower
> pricing from the rep & has further incentive to keep giving you support.
> 
> Use the design registration as a carrot on a stick to wave in front of the
> disty.....no support, no registration.
> 
> >      By the way, this is true of most semiconductor companies, not just
> > Xilinx.
> > -Simon Ramirez, Consultant
> >  Synchronous Design, Inc.
> >
> >
> >
> > "Andy Peters n o a o [.] e d u>" <"apeters <"@> wrote in message
> > news:8pbfj5$16qm$1@noao.edu...
> > > I really hate to ask silly questions like this, but I hate calling the
> > > distros because once you tell them you're thinking about considering
> > > thinking about using a part in a design, they jump all over you because
> > > they think you're gonna be writing a req for 10,000 parts tomorrow ...
> > >
> > > I'm looking at doing a VirtexE design.  XCV50E should be big enough.
> > > Mainly, I want the LVDS I/O.  Are they available now, or will they be
> > > available in the next couple of months?
> > >
> > > Of course, I could probably do the design with external LVDS parts, but
> > > this seems like a "neat" solution.
> > >
> > > -- andy
> > > ----------------------------
> > > 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
> > >
> >
> >

-- 

Rick Collins

rick.collins@XYarius.com

Ignore the reply address. To email me use the above address with the XY
removed.



Arius - A Signal Processing Solutions Company
Specializing in DSP and FPGA design

Arius
4 King Ave
Frederick, MD 21701-3110
301-682-7772 Voice
301-682-7666 FAX

Internet URL http://www.arius.com
Article: 25393
Subject: Re: How many 4005s (4010s) does it take to make a general purpose CPU?
From: Ben Franchuk <bfranchuk@jetnet.ab.ca>
Date: Sat, 09 Sep 2000 15:54:50 +0000
Links: << >>  << T >>  << A >>
"B. Joshua Rosen" wrote:
> 
> Do a DG Nova 16 bit CPU. The Nova was able to fit on a single board in
> 1969, surely it can fit in a 4005. The Nova's instruction set requires
> almost no decoding, it has a fixed width 16 bit instruction and a tiny
> number of instructions. The instruction set is very simple, load, store,
> jump, and a single type of execute instruction called and ALC which
> combines a simple operation (add,sub, mov, neg, and, or, xor, com), with
> a carry in select (0,1,Carry,!Carry), a shift (left 1, right 1, no shift
> and byte swap), and a skip (noskip, always skip, skip on zero, skip on
> not zero, skip on carry 1, skip on carry 0, skip on neg, skip on pos).
> 

Here is some photos of the Nova Computers ,and more information on the
instruction set.
http://www.ultranet.com/~crfriend/museum/
Ben.
PS.I wonder how many 8008's does it take to replace the 586?

-- 
"We do not inherit our time on this planet from our parents...
 We borrow it from our children."
"Luna family of Octal Computers" http://www.jetnet.ab.ca/users/bfranchuk
Article: 25394
Subject: Re: 3.3/2.5 voltage regulators
From: eml@riverside-machines.com.NOSPAM
Date: Sat, 09 Sep 2000 16:23:25 GMT
Links: << >>  << T >>  << A >>
On Fri, 08 Sep 2000 13:52:00 -0700, Austin Lesea
<austin.lesea@xilinx.com> wrote:

>Spartan2 transistors are smaller than Virtex transistors, and there are
>fewer of them, and the engineers (just went a'visiting) improved the design.

Can you tell us why there are fewer transistors in Spartan2? Is this
just a blockRAM size difference? I thought the only significant
difference was the geometry.

Evan
Article: 25395
Subject: Re: IEEE 754 Floating point VHDL functions / MATH package
From: eml@riverside-machines.com.NOSPAM
Date: Sat, 09 Sep 2000 16:23:58 GMT
Links: << >>  << T >>  << A >>
On Fri, 08 Sep 2000 19:02:27 +0200, "Jaap H. Mol" <jh_mol@wxs.nl>
wrote:

>
>Hi,
>
>I'm looking for VHDL (conversions) functions supporting the IEEE 754
>floating point standard. To be more specific, I want to convert
>variables 
>of type "real" to the IEEE 754 floating point format (single precision,
>32-bit), and vice versa.

VHDL (in common with other languages) doesn't define the
bit-representation of a float; all you know is that you have at least
6 decimal digits of precision, and the range is in the bounds -1.0E38
to +1.0E38. The actual representation will the same as the one used by
the processor, so it's likely to be 754 anyway. Even if you knew what
the representation was, you'd have a problem getting to the bits to do
anything with them. 

>In addition, I would greatly appreciate anyone who could direct me to
>the (lastest version of) the (proposed?) VHDL MATH package.

The package has been around for a few years now - it's not free, and
you have to get it from the IEEE.

Evan
Article: 25396
Subject: Tutorial for ABEl-HDL
From: Michael Daldrup <michael@mdaldrup.de>
Date: Sat, 09 Sep 2000 19:18:26 +0200
Links: << >>  << T >>  << A >>
Hi out there!

Can anybody tell me where I can find
a GOOD Abel-tutorial?

Thanks
  Michael


Article: 25397
Subject: Re: pcilogic celss
From: Duane <junkmail@junkmail.com>
Date: Sat, 09 Sep 2000 14:50:54 -0700
Links: << >>  << T >>  << A >>
ed.moore@snellwilcox.com wrote:
> 
> Does anyone have any information on the PCILOGIC cells in Xilinx Virtex
> devices ?.
> 
> From viewing the layout in FPGA Viewer (I would call it FPGA Editor if
> it didn't crash 90% of the time I try to edit anything) they seem to
> take in IRDY and TRDY from dedicated PCIIOB cells, plus a couple of
> other signals, and generate a clock enable signal for use by local IOBs.
> 
> Anyone know anything more ?

The easy way to find out more about the block is just instantiate it in
a design. It is named "pcilogic" and has the inputs IRDY, TRDY, I1, I2,
and I3, and the output PCI_CE, all of which can be determined from "FPGA
Viewer". Run map and par, then ngdanno and ngd2vhdl. The result is a
VHDL file that includes a VHDL model of that logic block.

It is a very simple combinatorial logic block, where I1 is an enable for
IRDY, I3 is an enable for TRDY, and I2 is a direct enable.

--
My real email is akamail.com@dclark (or something like that).
Article: 25398
Subject: Re: How do I mix vhdl and verilog source files in Synplify?
From: rkadam@my-deja.com
Date: Sat, 09 Sep 2000 22:32:56 GMT
Links: << >>  << T >>  << A >>
HI!
I never tried mixing verilog and vhdl modules in synplify.
But i can tell you a workaround that will surely work.
Synthesize your verilog module (assuming your VHDL module
instanciates all the verilog modules) first, create a black
box for all the verilog instances, compile the VHDL module
and write out the EDIF file, So now you are left with two
EDIF files, one generated from Verilog and other from VHDL
synthesis.
Your place and route tool will take care of the rest.

Hope this helps.
If you require any further help please mail me at
rkadam@asic.qntm.com

Rajkumar...



In article <39B8F7FA.A8727396@emw.ericsson.se>,
  Thomas Karlsson <thomas.karlsson@emw.ericsson.se> wrote:
> Hi all,
>
> I trying to synthesize an old design written in verilog (which is not
my
> cup of tea, I use VHDL). This design instantiate some module, but I
want
> to use a new version of this module, written in VHDL. How do I do
this?
> I am using Synplify 6.0, which should support mixed language source
> files, but it complains that the referenced module can not be found.
> I know that the module name is not declared in any
verilog file
anymore,
> but when the vhdl file with the corresponding entity name is compiled
> into the library work, I supposed that Synplify should be able to plug
> that compiled entity name into the place where the module name is
> referenced, but I can not figure out how to do it. Maybe I trying the
> impossible here?
>
> Thanks for any help.
>
> Thomas
>


Sent via Deja.com http://www.deja.com/
Before you buy.
Article: 25399
Subject: Re: Mealy vs Moore FSM model
From: rajkumar@gdatech.com
Date: Sat, 09 Sep 2000 22:47:33 GMT
Links: << >>  << T >>  << A >>
Hi!
I would just like to add my personal comments which i
have gained through designing many state machines.
First of all when designing a real ASIC, mostly you
end up with a mixture of Moore and Mealy, I mean sometimes
your output depends on state only sometimes it depends on
State and the present input.
I have rarely found Pure ( Text book ) state machines
implemented in reality, except for small state machines.
Now whether you should implement the state machine using
one process or multiple processes, I think the best way
is to follow the Hardware description ( Diagram ) of the
State machine, in which we end up with three processes,
Ist process -> P.S <= NS;
IInd process -> Next State assignment
IIIrd process -> Output assignment.

You can actually end up with four processes if you want to
register the output, Other way of implementing is to have
a single process for output and next state assignment.

And use the third process for Registered outputs.

I follow this as it is close to the hardware definition of
the state machine and it is easier to understand.


Rajkumar...



In article <G0H2wq.6sB@world.std.com>,
  jhallen@world.std.com (Joseph H Allen) wrote:
> In article <39b607f5.7405564@news.dial.pipex.com>,
>  <eml@riverside-machines.com.NOSPAM> wrote:
>
> >Isn't this a classic Verilog race condition? What happens if the 2
> >always blocks are executed in a different order?
>
> Oops, you're right, it is.  You can use = for temporary variables,
since
> execution within an always block is sequential, but you had better not
use
> it for generating Mealy state signals unless it's in a seperate
> combinatorial always block.
>
> Verilog just doesn't work the way I want it to :-)
>
> --
> /*  jhallen@world.std.com (192.74.137.5) */               /* Joseph H.
Allen */
> int
a[1817];main(z,p,q,r){for(p=80;q+p-80;p-=2*a[p])for(z=9;z--;)q=3&(r=time
(0)
>
+r*57)/7,q=q?q-1?q-2?1-p%79?-1:0:p%79-77?1:0:p<1659?79:0:p>158?-79:0,q?!
a[p+q*2
> ]?a[p+=a[p+=q]=q]=q:0:0;for(;q++-1817;)printf(q%79?"%c":"%c\n","
#"[!a[q-1]]);}
>


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



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