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 125625

Article: 125625
Subject: Re: FFT for an arbitrary number of points (not power of 2)
From: John McCaskill <jhmccaskill@gmail.com>
Date: Tue, 30 Oct 2007 15:08:18 -0000
Links: << >>  << T >>  << A >>
On Oct 30, 12:23 am, Patrick Dubois <prdub...@gmail.com> wrote:
> Hi,
>
> I'd like to ask experts here about ideas to perform a FFT on an
> arbitrary number of points (for real data).
>
> The cores usually found for an FPGA implementation only permit FFTs on
> a number of points that is a power of 2. In our particular case
> however, we need to be able to do an FFT on a vector of say, 1025
> points. Our current algorithm is to zero-pad this vector to 2048
> points.
>
> The problem with this approach is that we nearly double the number of
> points for downstream processing. This is very problematic at high
> datarates.
>
> We have a few ideas but since this seems like a common problem, I'd
> like to ask people here for tips.
>
> Thanks!
>
> Patrick




Hello Patrick,


The FFT algorithm is efficient because it uses a divide and conquer
approach to process the data in subsets.  The data is processed by
multiple "butterfly" processing kernels.  Power of two data sets are
easy to deal with, because you can just use power of two butterflies
and the addressing is regular.

You can still use the FFT algorithm on non power of two data sets, but
if you do not want to pad the data up to the next power of two, you
will need non power of two butterflies.

For your case, 1025 factors to 5*5*41 so you would need to have
radix-5 and radix-41 butterflies if you really want to have a 1025
point FFT. If you just want to not have to pad all the way to 2048 but
don't care about padding a bit you could find a nicer size. For
example 1152 would be 2^7 * 3^2 and would only need radix-2 and
radix-3 butterflies.

If you want an FFT that is not a power of two in size, you will need a
non power of two butterfly in the mix.

Regards,

John McCaskill
www.fastertechnology.com


Article: 125626
Subject: Re: FPGA vs ASIC
From: mk <kal*@dspia.*comdelete>
Date: Tue, 30 Oct 2007 08:42:11 -0700
Links: << >>  << T >>  << A >>
On Tue, 30 Oct 2007 07:10:49 -0700, Andy <jonesandy@comcast.net>
wrote:

>On Oct 29, 12:15 pm, mk <kal*@dspia.*comdelete> wrote:
>> On Mon, 29 Oct 2007 17:09:15 +0000, Philip Potter
>>
>> <p...@see.sig.invalid> wrote:
>> >IIRC Xilinx Virtex-5 is on a 45nm process.
>>
>> No, Virtex-5 is 65nm. No foundry is selling 45nm production wafers
>> yet.http://www.xilinx.com/products/silicon_solutions/fpgas/virtex/virtex5...
>
>No FPGA foundry... Intel is selling 45nm quad core processors now.

Intel is not a foundry ie it doesn't sell foundry services where
people can send them a design to be manufactured in their fabrication
facilities for a price. TSMC, UMC, Chartered and SMIC are examples of
foundries and none of them is selling production 45nm wafers at this
time.

Article: 125627
Subject: Re: FFT for an arbitrary number of points (not power of 2)
From: dbd <dbd@ieee.org>
Date: Tue, 30 Oct 2007 09:03:41 -0700
Links: << >>  << T >>  << A >>
On Oct 30, 6:38 am, Patrick Dubois <prdub...@gmail.com> wrote:
> On 30 oct, 09:23, John_H <newsgr...@johnhandwork.com> wrote:
>
>
>
> > I'd suggest that an arbitrary number - selected at runtime rather than
> > by design - isn't practical but designing for 1025 points might be.
>
> > I forget if it was LeCroy or Tektronix, but about a decade ago I read
> > the whitepaper on their scope's FFT capability.  Scopes also have the
> > "limitation" of non-2^n size samples.  Rather than decompose their FFT
> > into 2-element butterflies, they decomposed their 2/5/10 multiple
> > waveform samples into 2-element and 5-element FFT nodules.  If the idea
> > behind the FFT is to reuse the sin/cos values for a given phase delta,
> > decomposing into a non-2^n system can work well.  What obviously won't
> > work well for this approach is any size with large prime numbers.
>
> > Your 1025 point example calls for 5x5x41 which wouldn't pring so much in
> > acceleration.
>
> > Sorry I don't have an url for the whitepaper - it's been a decade.
>
> > - John_H
>
> Thanks for the idea. But ideally I'd like to limit myself to power of
> 2 FFTs so that I can use the Xilinx core. The number of points needs
> to be flexible from 64 to 32k. Decomposition into smaller chunks seems
> like a promising avenue. Although if one number is decomposible into
> two power of 2s (e.g. 8192, which is divisable by 64 and 128), then
> that number itself will be a power of 2.
>
> I'd be quite happy with an algorithm that increases the number of
> points possible (compared to only power of 2s). Being able to do
> _every_ possible number of points is probably too strict a
> requirement.
>
> Patrick

Since you have used zero-filling, you don't seem to need a transform
the exact size of the data. There is another technique complementary
to zero-filling, sometimes called data folding. Since the DFT is a
circular convolution, when your data set is larger than the transform
size you can continue to 'wrap' the data in a circular manner and sum
the overlapped samples. You then perform the smaller sized FFT. You
might consider data folding up to sqrt 2 times the smaller power of
two FFT size and zero-filling up to the larger transform size above
that.

Chapter 8 (by fred harris) of 'Handbook of Digital Signal Processing
Engineering Applications' edited by Douglas Eliot discusses this
approach.

With either zero-fill or data folding it is useful to remember that
the window size that determines bin response shapes is the data size
not the transform size.

Dale B. Dalrymple
http://dbdimages.com
http://stores.lulu.com/dbd


Article: 125628
Subject: Re: Xilinx, MIG, UCF: timing constraints for DDR2 DRAM
From: "Simon Heinzle" <sheinzle@inf.ethz.ch>
Date: Tue, 30 Oct 2007 17:41:48 +0100
Links: << >>  << T >>  << A >>
Hi all,

I found the problem: the "Retiming" option in Synplify Pro caused the 
address signals to be retimed and they could therefore not be packed into an 
IOB flip flop. This caused a much higher delay than in the other output 
signals.

Thanks for your help,
Simon

"Simon Heinzle" <sheinzle@inf.ethz.ch> wrote in message 
news:47261129$1@news1-rz-ap.ethz.ch...
> Hi FPGA Group!
>
> I'm struggling to get a fast speed (~ 200 MHz) for the DDR2 DRAM 
> interface, generated with the Xilinx Memory Interface Generator. The 
> complete system consists of a PCI interface, an I/O DMA buffer, a burst 
> module bursting from DMA buffer to the DDR2 DRAM interface.
>
> What is the best way to define setup/hold times for the I/O pads (UCF)? 
> (the RAM interface consists of a bi-dir data bus DQ, some output signals 
> e.g. A and the DRAM clocks CK)
> 1. Using the OFFSET = OUT 5 ns AFTER "SYS_CLK_P"? Unfortunately, this does 
> only work using the Input Clock Pin, but it should probably better be in 
> reference to the DRAM clocks CK)
> 2. Using TIMESPEC "TS_DDR_OUT" = FROM FFS TO "DDR_OUT" 5 ns; ? Is probably 
> better, but I'm not exactly sure.
>
> Furthermore, how would you tackle the problem if the timing at the pads 
> cannot be met?
>
> Thanks in advance for helpful answers and pointers in the right direction!
>
> Best regards,
> Simon
>
>
> 



Article: 125629
Subject: Re: FFT for an arbitrary number of points (not power of 2)
From: Patrick Dubois <prdubois@gmail.com>
Date: Tue, 30 Oct 2007 16:50:24 -0000
Links: << >>  << T >>  << A >>
On 30 oct, 12:03, dbd <d...@ieee.org> wrote:

> Since you have used zero-filling, you don't seem to need a transform
> the exact size of the data.

Correct. Although I want to minimize data growth as much as possible
(which is the drawback of zero-filling).

> There is another technique complementary
> to zero-filling, sometimes called data folding. Since the DFT is a
> circular convolution, when your data set is larger than the transform
> size you can continue to 'wrap' the data in a circular manner and sum
> the overlapped samples. You then perform the smaller sized FFT. You
> might consider data folding up to sqrt 2 times the smaller power of
> two FFT size and zero-filling up to the larger transform size above
> that.
>
> Chapter 8 (by fred harris) of 'Handbook of Digital Signal Processing
> Engineering Applications' edited by Douglas Eliot discusses this
> approach.

Interesting. Thanks for the reference, I'll try to find a copy of the
book.

Patrick



Article: 125630
Subject: Re: X3100A design with Synplify 8.8 and foundation 1.5 possible?
From: Walter Dvorak <use-reply-to@invalid.invalid>
Date: Tue, 30 Oct 2007 19:18:12 +0000 (UTC)
Links: << >>  << T >>  << A >>
Antti <Antti.Lukats@googlemail.com> wrote:
> foundation 1.5 mapper terminates on windows XP with GPF fault
[...]
> or any other recommendations how to create bit file from VHDL for
> XC3100A FPGA

	why not setup up a vmware sandbox with win95 (or 
something like that) and running the old tools in this box? 

	(and it is much easier to backup the vmware image)

WD
-- 

Article: 125631
Subject: Re: FPGA vs ASIC
From: glen herrmannsfeldt <gah@ugcs.caltech.edu>
Date: Tue, 30 Oct 2007 12:59:17 -0800
Links: << >>  << T >>  << A >>
Andrew FPGA wrote:
(snip)

> Also, they compared the FPGA to a standard cell asic. Presumably with
> full custom the difference would be even greater. (and the NRE FPGA
> advantage would be even greater too of course.)

Is Sea-of-gates still available?  I used to  have the descriptions
of the libraries for them.  For SOG, as I understand it, only one custom 
mask is needed, over what it pretty much an array of transistors.
The timing isn't quite as good as standard cell, as the transistor
size can't be varied as much.  It should be a lot cheaper, though.

-- glen


Article: 125632
Subject: Re: X3100A design with Synplify 8.8 and foundation 1.5 possible?
From: Antti <Antti.Lukats@googlemail.com>
Date: Tue, 30 Oct 2007 21:21:44 -0000
Links: << >>  << T >>  << A >>
On 30 Okt., 20:18, Walter Dvorak <use-reply...@invalid.invalid> wrote:
> Antti <Antti.Luk...@googlemail.com> wrote:
> > foundation 1.5 mapper terminates on windows XP with GPF fault
> [...]
> > or any other recommendations how to create bit file from VHDL for
> > XC3100A FPGA
>
>         why not setup up a vmware sandbox with win95 (or
> something like that) and running the old tools in this box?
>
>         (and it is much easier to backup the vmware image)
>
> WD
> --
yes can do that - my question was how likely is F 1.5 to work at all
on 95 or NT
ok, i also checked the M 1.5 with sch, it also GPF on map so its not
the synplify edif issue

weird I also installed F 3.1 that according to xilinx should also
support 3100A but it doesnt
so i still may need to get M 1.5 working somehow

Antti







Article: 125633
Subject: Re: Power supply filter capacitors
From: glen herrmannsfeldt <gah@ugcs.caltech.edu>
Date: Tue, 30 Oct 2007 13:21:53 -0800
Links: << >>  << T >>  << A >>
Nevo wrote:

(snip)

> This is definitely on my list of improvements to make in the future.  I'm 
> going to build the first unit with the brute force approach and add in 
> refinements.  (Christmas 2008 should be killer!) :)

> Not only will this change reduce total power through the circuit, it'll 
> allow me to use a smaller transformer, which is one of the costliest parts 
> in the design. 

I once thought about doing an array of christmas lights multiplexed
so that I could turn each one on and off under computer control.
It needs a lot of diodes to do that, though now with LED christmas
lights those wouldn't be needed.

-- glen


Article: 125634
Subject: Re: Is it possible to check how cache memories are mapped to FPGA block rams?
From: Eric Smith <eric@brouhaha.com>
Date: Tue, 30 Oct 2007 14:22:41 -0700
Links: << >>  << T >>  << A >>
Wei Wang wrote:
> Eric, I appreciate your willingness to dig further to help, but my
> question was how I could check the block ram mapping of my design
> which I thought it was quite generic, and I would expect answers, such

Without looking at your HDL code, I don't think we have any clue how
you've attached the blockrams to your CPU, so I'm not sure how we
could tell you how to check the mapping.

> BTW, I
> suppose most of us in this group do not work for ourselves, only lazy
> university students would post their entire project and let somebody
> else do the work for them.

In general, I agree.  But sometimes it's hard to offer any help without
more detail, and sometimes the easiest way to see the relevant detail
is to look at the HDL.

I'm sorry that I don't have any more specific advice to offer.

Best regards,
Eric


Article: 125635
Subject: Re: Power supply filter capacitors
From: glen herrmannsfeldt <gah@ugcs.caltech.edu>
Date: Tue, 30 Oct 2007 13:22:58 -0800
Links: << >>  << T >>  << A >>
Nevo wrote:

(snip)

> If I were designing the circuit board, I'd put in pads for snubbers. I'm 
> buying from a 'group buy' on a DIY Christmas lights forum and the existing 
> boards don't have snubbers designed in. I'll provide this feedback for next 
> year's designs. :)

How about a URL so that others interested could follow along?

-- glen


Article: 125636
Subject: debugging ppc + mb
From: me_2003@walla.co.il
Date: Tue, 30 Oct 2007 14:33:30 -0700
Links: << >>  << T >>  << A >>
Hi all,
I would like to debug a system containing a microblze and a ppc405.
I'm
using the xmd (gdb) for both of these units. I have a single mdm unit
and a jtagppc (a single jtag interface).
Is there a way to debug both of the processors simultaneously (via
two
GDBs).
Thanks in advance, Mordehay


Article: 125637
Subject: Re: FFT for an arbitrary number of points (not power of 2)
From: Andrew FPGA <andrew.newsgroup@gmail.com>
Date: Tue, 30 Oct 2007 14:45:30 -0700
Links: << >>  << T >>  << A >>
> I'd like to ask experts here about ideas to perform a FFT on an
> arbitrary number of points (for real data).

Hi Patrick,
A popular misconception seems to be that in order to use cooley-turkey
FFT the number of points must be a power of 2. Digital Signal
Processing with Field Programmable Gate Arrays by Uwe Meyer-Baese, for
example, has an example we he shows the Cooley-Tukey FFT for N = 12.
He claims 674 real additions and 432 real multiplications for a 12
point DFT, and for the 12 point FFT 108 real additions and 28 real
multiplications. The signal flow graph he shows, shows in the 1st
stage, 3x four point DFTs, followed by twiddle factors, and then a
final stage of 4x three point DFT's.

He also goes on to present the Good-Thomas FFT algorithm and Winograd
FFT. Again he demonstrates for N=12.

Regards
Andrew




Article: 125638
Subject: Re: FPGA vs ASIC
From: Thomas Stanka <usenet_nospam_valid@stanka-web.de>
Date: Tue, 30 Oct 2007 14:46:40 -0700
Links: << >>  << T >>  << A >>
On 30 Okt., 16:42, mk <kal*@dspia.*comdelete> wrote:
> >No FPGA foundry... Intel is selling 45nm quad core processors now.
>
> Intel is not a foundry ie it doesn't sell foundry services where
> people can send them a design to be manufactured in their fabrication
> facilities for a price. TSMC, UMC, Chartered and SMIC are examples of
> foundries and none of them is selling production 45nm wafers at this
> time.


IBM claims to provide a 45 nm ASIC process(IBM Cu-45HP ASIC). I don't
know, if you could get a production start right now(if your device
would be ready for production), but it is offered to customers on
their homepage.

bye Thomas


Article: 125639
Subject: Re: FPGA vs ASIC
From: Thomas Stanka <usenet_nospam_valid@stanka-web.de>
Date: Tue, 30 Oct 2007 14:50:44 -0700
Links: << >>  << T >>  << A >>
On 27 Okt., 17:08, mk <kal*@dspia.*comdelete> wrote:
> On Sat, 27 Oct 2007 00:39:05 -0700, Thomas Stanka
>
> <usenet_nospam_va...@stanka-web.de> wrote:
> >when it comed to adders because nearly every actual fpga provides fast
> >carry logic which I haven't seen in ASIC so far.
>
> Almost all ASIC libraires I've used has a full adder in it which is
> basically what a fast carry logic is in an FPGA where they have
> hardwired full adders which don't need to be made from luts and don't
> need the programmable interconnect. I'ts very easy to accomplish the
> same in an ASIC.

Is it only a fulladder gate, or a full adder (for given bitsize)?
The adders I've seen in ASIC libs provide no fast carry compared to
normal cell delay. But that has nothing to say, as I'm not that
experienced in comparing ASIC technologies.

bye Thomas


Article: 125640
Subject: Re: debugging ppc + mb
From: "Vasanth Asokan" <vasanth@xilinx.com>
Date: Tue, 30 Oct 2007 15:08:42 -0700
Links: << >>  << T >>  << A >>

<me_2003@walla.co.il> wrote in message 
news:1193780010.206149.316160@57g2000hsv.googlegroups.com...
> Hi all,
> I would like to debug a system containing a microblze and a ppc405.
> I'm
> using the xmd (gdb) for both of these units. I have a single mdm unit
> and a jtagppc (a single jtag interface).
> Is there a way to debug both of the processors simultaneously (via
> two
> GDBs).
> Thanks in advance, Mordehay
>

Just connect to the two processors in XMD via two separate connect commands. 
XMD will open up a GDBServer port for each processor. Connect a GDB session 
each to the GDBServer ports. You are all set to debug the two processors 
simultaneously.

Be aware of the default system reset performed by XMD upon download of a 
program. You can use the debugconfig command to change this behavior, if it 
is destructive for your simultaneous debug sessions.




Article: 125641
Subject: Re: debugging ppc + mb
From: John Williams <jwilliams@itee.uq.edu.au>
Date: Wed, 31 Oct 2007 08:17:35 +1000
Links: << >>  << T >>  << A >>
Hi,

me_2003@walla.co.il wrote:

> I would like to debug a system containing a microblze and a ppc405.
> I'm
> using the xmd (gdb) for both of these units. I have a single mdm unit
> and a jtagppc (a single jtag interface).
> Is there a way to debug both of the processors simultaneously (via
> two
> GDBs).

You shld be able to do this - using xmd, connect to both CPUs:

% connect mb mdm
% connect ppc hw

you may need to add other options to each connect statement, depending 
on your FPGA and JTAG setup etc.

This sequence would make the MB target 0, and the PPC target 1

xmd should then be listening on two different ports, one for the mb, and 
one for the ppc.

Since xmd tends to allocate ports from 1234 upwards, my guess is that 
the mb will be on port 1234, and the ppc on port 1235 - the actual port 
no's will be printed by xmd after each connection is made.

Then, start each gdb, issue a target command

target remote localhost:1234 (for microblaze)

and

target remote localhost:1235 (for the ppc).

Some minor details may remain for you to work out, but this is an 
overview of the process

Regards,

John

Article: 125642
Subject: Re: FFT for an arbitrary number of points (not power of 2)
From: Ray Andraka <ray@andraka.com>
Date: Tue, 30 Oct 2007 19:36:08 -0400
Links: << >>  << T >>  << A >>
Patrick Dubois wrote:

> Hi,
> 
> I'd like to ask experts here about ideas to perform a FFT on an
> arbitrary number of points (for real data).
> 
> The cores usually found for an FPGA implementation only permit FFTs on
> a number of points that is a power of 2. In our particular case
> however, we need to be able to do an FFT on a vector of say, 1025
> points. Our current algorithm is to zero-pad this vector to 2048
> points.
> 
> The problem with this approach is that we nearly double the number of
> points for downstream processing. This is very problematic at high
> datarates.
> 
> We have a few ideas but since this seems like a common problem, I'd
> like to ask people here for tips.
> 
> Thanks!
> 
> Patrick
> 


As others have mentioned, you need non-power of two FFT kernels to get 
non-power of two transform sizes unless you use zero padding.  A 
possibility would be to have a set of non-power of two kernels you could 
combine using the mixed radix algorithm to get larger FFTs.  If your FFT 
kernel sizes are relatively prime, you get a significant simplification 
because it doesn't need phase rotations between the FFT kernels.  For 
example, a 1540 point FFT can be had by combining 4, 5, 7 and 11 point 
kernels with no intervening rotations.  If you use rotations between the 
FFT kernels, you can mix the kernels without regard to whether or not 
they are relatively prime. There are Winograd kernels for all these 
sizes, and others (Rader or Singleton as I recall) for 3 and 5 point. 
These are actually quite a bit easier to do in hardware than they are 
with a microprocessor, as the reordering is a bit convoluted.  You might 
look at the Smith & Smith book on FFT's 
(http://www.amazon.com/exec/obidos/ASIN/0780310918/andraka), as they 
have a very good coverage of non-Cooley-Tukey FFTs.  Unfortunately, 
arbitrary sizes are a little harder to do if it is to be flexible 
because the addressing and phase rotations are not as well ordered as 
they are for power of 2 sizes.

Article: 125643
Subject: Re: builing a SPI interface in vhdl
From: techG <giuliopulina@gmail.com>
Date: Tue, 30 Oct 2007 17:58:37 -0700
Links: << >>  << T >>  << A >>
On Oct 30, 6:17 am, futz...@gmail.com wrote:
> On Oct 25, 9:57 am, techG <giuliopul...@gmail.com> wrote:
>
> > HI, I'm new in FPGA, I have to build a SPI interface (in VHDL) to let
> > an fpga read and write a flash memory.
> > The fpga is a Xilinx Spartan3E, while the memory is an ST M25P16
> > (serial I/O).
> > Do you know if is there any built vhdl core to start with?
>
> > Thanks in advance
> > Giulio
>
> You can trywww.opencores.org;
> The SPI core interface is quite simple to code up as well.
> Cheers

i just found a project for SPI controller in opencores.org
(spiflashcontroller)
it's not as simple as I expected, but i found it more useful!
thank you all



Article: 125644
Subject: Re: FPGA vs ASIC
From: mk <kal*@dspia.*comdelete>
Date: Wed, 31 Oct 2007 04:04:11 GMT
Links: << >>  << T >>  << A >>
On Tue, 30 Oct 2007 14:46:40 -0700, Thomas Stanka
<usenet_nospam_valid@stanka-web.de> wrote:

>On 30 Okt., 16:42, mk <kal*@dspia.*comdelete> wrote:
>> >No FPGA foundry... Intel is selling 45nm quad core processors now.
>>
>> Intel is not a foundry ie it doesn't sell foundry services where
>> people can send them a design to be manufactured in their fabrication
>> facilities for a price. TSMC, UMC, Chartered and SMIC are examples of
>> foundries and none of them is selling production 45nm wafers at this
>> time.
>
>
>IBM claims to provide a 45 nm ASIC process(IBM Cu-45HP ASIC). I don't
>know, if you could get a production start right now(if your device
>would be ready for production), but it is offered to customers on
>their homepage.

Here is a quote from their press release: "IBM SiGe BiCMOS 6WL design
kits are available now. IBM plans to have SiGe BiCMOS 5PAe design kits
available this summer and first design kits for CMOS 11LP Foundry
products later in 2007. The planned availability date for Cu-45HP ASIC
is in early 2008."

http://www-03.ibm.com/press/us/en/pressrelease/21648.wss

Article: 125645
Subject: Re: FPGA vs ASIC
From: mk <kal*@dspia.*comdelete>
Date: Tue, 30 Oct 2007 21:15:22 -0700
Links: << >>  << T >>  << A >>
On Tue, 30 Oct 2007 14:50:44 -0700, Thomas Stanka
<usenet_nospam_valid@stanka-web.de> wrote:

>On 27 Okt., 17:08, mk <kal*@dspia.*comdelete> wrote:
>> On Sat, 27 Oct 2007 00:39:05 -0700, Thomas Stanka
>>
>> <usenet_nospam_va...@stanka-web.de> wrote:
>> >when it comed to adders because nearly every actual fpga provides fast
>> >carry logic which I haven't seen in ASIC so far.
>>
>> Almost all ASIC libraires I've used has a full adder in it which is
>> basically what a fast carry logic is in an FPGA where they have
>> hardwired full adders which don't need to be made from luts and don't
>> need the programmable interconnect. I'ts very easy to accomplish the
>> same in an ASIC.
>
>Is it only a fulladder gate, or a full adder (for given bitsize)?
>The adders I've seen in ASIC libs provide no fast carry compared to
>normal cell delay. But that has nothing to say, as I'm not that
>experienced in comparing ASIC technologies.

Xilinx fast carry logic path for v5 is described in this document
http://direct.xilinx.com/bvdocs/userguides/ug190.pdf page 193. As you
can see it is not a multi-bit carry lookahead or anything similarly
complicated. It's a hardwired implementation of a ripple carry logic
which can be duplicated using standard cell full adders relatively
easily.

Article: 125646
Subject: Re: Updating my bookshelf
From: Mike Treseler <mike_treseler@comcast.net>
Date: Tue, 30 Oct 2007 22:52:48 -0700
Links: << >>  << T >>  << A >>
Nicolas Matringe wrote:

> What are the books you wouldn't work without ?

I could work without books,
but not without a simulator.

    -- Mike Treseler

Article: 125647
Subject: Re: FPGA vs ASIC
From: Thomas Stanka <usenet_nospam_valid@stanka-web.de>
Date: Tue, 30 Oct 2007 23:10:39 -0700
Links: << >>  << T >>  << A >>
On 31 Okt., 05:15, mk <kal*@dspia.*comdelete> wrote:
> On Tue, 30 Oct 2007 14:50:44 -0700, Thomas Stanka
>
>
>
> <usenet_nospam_va...@stanka-web.de> wrote:
> >On 27 Okt., 17:08, mk <kal*@dspia.*comdelete> wrote:
> >> On Sat, 27 Oct 2007 00:39:05 -0700, Thomas Stanka
>
> >> <usenet_nospam_va...@stanka-web.de> wrote:
> >> >when it comed to adders because nearly every actual fpga provides fast
> >> >carry logic which I haven't seen in ASIC so far.
>
> >> Almost all ASIC libraires I've used has a full adder in it which is
> >> basically what a fast carry logic is in an FPGA where they have
> >> hardwired full adders which don't need to be made from luts and don't
> >> need the programmable interconnect. I'ts very easy to accomplish the
> >> same in an ASIC.
>
> >Is it only a fulladder gate, or a full adder (for given bitsize)?
> >The adders I've seen in ASIC libs provide no fast carry compared to
> >normal cell delay. But that has nothing to say, as I'm not that
> >experienced in comparing ASIC technologies.
>
> Xilinx fast carry logic path for v5 is described in this documenthttp://direct.xilinx.com/bvdocs/userguides/ug190.pdfpage 193. As you
> can see it is not a multi-bit carry lookahead or anything similarly
> complicated. It's a hardwired implementation of a ripple carry logic
> which can be duplicated using standard cell full adders relatively
> easily.


Thankyou, I have no need to learn the details of the virtex5 carry
logic at the moment.
The point I'm on is that in the fpga technologies I know the carry
chain path has a much faster gate delay compared to the normal gate
delay.
This means you could do 8 to 16 bit adding in a pipeline running
nearly at max technology speed for FF-gate-FF, maybe you need more
time multiplexing the data before or after the adder stage than the
time you need for adding.
In ASIC I've only seen gates with the carry chain having a gate delay
compareable to a gate delay, this means, that a 16 bit ripple adder
won't be able to run anywhere near FF-gate-FF but needs something like
FF-17xgate-FF. In that case you will likely have the adder stage
dominating your pipeline frequency. You could only speed up a ripple
carry adder by placing it tight together, but this has to be done at
any timing critical path anyway.

bye Thomas


Article: 125648
Subject: Re: FPGA vs ASIC
From: Kim Enkovaara <kim.enkovaara@iki.fi>
Date: Wed, 31 Oct 2007 08:49:05 +0200
Links: << >>  << T >>  << A >>
Thomas Stanka wrote:

> Is it only a fulladder gate, or a full adder (for given bitsize)?
> The adders I've seen in ASIC libs provide no fast carry compared to
> normal cell delay. But that has nothing to say, as I'm not that
> experienced in comparing ASIC technologies.

There are also different adder implementations at the synthesis
tool level. For example Design compiler with proper libraries
supports following adders: ripple-carry, carry-look-ahead,
delay-optimized flexible parallel-prefix, brent-kung, conditional-sum
and ripple-carry-select.

--Kim

Article: 125649
Subject: Re: IDE to Flash memory
From: Franck <franck.jullien@gmail.com>
Date: Wed, 31 Oct 2007 06:57:47 -0000
Links: << >>  << T >>  << A >>
Thanks,

I have seen this product but this is not flexible enough for me. This
king of chip is too specific, that why I wanted to do this with a
fpga. I don't know exactly how flash memory works so I'm not able to
know which throughput I could reach with an IDE interface and a flash
memory controller in a fpga....

I thought someone else could have done this before, then I would gain
some time...

Anyway, I'll have to shearch by myself :)





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