Custom Search
Site Home Archive Home FAQ Home How to search the Archive How to Navigate the Archive
Compare FPGA features and resources
Threads starting:
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
In article <Pine.GUL.3.95.971212130608.1533A-100000@po.EECS.Berkeley.EDU> Eyal Soha <barrel@po.EECS.Berkeley.EDU> writes: >Furthermore, I've received at least two emails from people looking at my >design that claim switching to bit-serial processing would allow me to >full pipeline my design such that 12 boards would double the entire >throughput of the project (which currently stands at around the power of >14,000 Pentium Pro/200s). I said that, but I meant it as a (unlikely) upper bound if one managed an effective 1ns clock speed, and managed by hook-or-by-crook to pipeline and parallelize everything so as to crank out one result per clock. That would be ideal, and hence is the thing to shoot for -- but I'd never say flat out that it's possible, merely a good goal. :-) A 300 Mhz Pentium II with SDRAM manages not quite 1 million results per second, so that scenario would indeed hypothetically have 12 boards (12 giga results/sec) equal to 12-16,000 Pentiums. I was a bit over- enthusiastic at the thought of that. But even at 10-100 million results per second, you'd have a nice fast board. Doug -- Doug Merritt doug@netcom.com Professional Wild-eyed Visionary Member, Crusaders for a Better Tomorrow Fight Spam! Join CAUCE (free) at www.cauce.org Unicode Novis Cypherpunks Gutenberg Wavelets Conlang Logli Alife Anthro Computational linguistics Fundamental physics Cogsci Egyptology GA TLAsArticle: 8426
In article <34931A20.44247290@CatenaryScientific.com>, Chuck Parsons <chuck@CatenaryScientific.com> writes > > >John Woodgate wrote: > >> I won't argue with your reasoning: questions of probability often have >> counter-intuitive answers. However, looking at the subject another way, >> the resistor/integrator device seems to be able to extract noise energy, >> and that means thermal energy, from the resistor and store a net amount >> of it in the capacitor. What happens then to the temperature of the >> resistor? What is wrong with this crazy result? > > Well lets not get sidetracked on that. John, I think you may be trying to >exhaust >me ;-) No, Im not. But I'm quite willing to abandon this thread. IT seems to have become just a 2-way discussion. -- Regards, John Woodgate, Phone +44 (0)1268 747839 Fax +44 (0)1268 777124. OOO - Own Opinions Only. It is useless to threaten a strong man - he will ignore you. It is dangerous to threaten a weak man - he will kill you if he can.Article: 8427
muzo wrote: > > hi, > I have a design which uses %55 of a 10K100 and within this design I have 12 8 bit > registers which need to be selectively loaded into another register, IOW, I have > an 8 bit bus from which 12 8 bit registers hang. Because 10K doesn't have > internal tristates, I have a big mux which selects the register. Currently this > mux is the bottleneck in my design which limits my speed to 11 MHz in a -3 part. > Any ideas how to increase the speed in this design? The design is in Verilog. > > Are there any ways to get more performance by using the floor plan tool ? I am > already using FAST synthesis and maximum speed in the compiler options. > > thanks > > muzo > > WDM & NT Kernel Driver Development Consulting <muzok@pacbell.net> Hi, muzo: I meet a similar problem, but I use 8K. My method is that you can assign the function which should be connected to the same LAB. In your design, you can split the 12 8 bit registers to 12 groups, mux to 12 groups, and the destination register to 12groups. Then you can use floorplan to drag the same group to the LAB as near as possibly. Regards. Ss.Article: 8428
In article <3491bc91.80619134@news.walltech.com>, muzok@pacbell.net (muzo) wrote: >hi, >I have a design which uses %55 of a 10K100 and within this design I have 12 8 bit >registers which need to be selectively loaded into another register, IOW, I have >an 8 bit bus from which 12 8 bit registers hang. Because 10K doesn't have >internal tristates, I have a big mux which selects the register. Currently this >mux is the bottleneck in my design which limits my speed to 11 MHz in a -3 part. >Any ideas how to increase the speed in this design? The design is in Verilog. > >Are there any ways to get more performance by using the floor plan tool ? I am >already using FAST synthesis and maximum speed in the compiler options. > >thanks > >muzo > >WDM & NT Kernel Driver Development Consulting <muzok@pacbell.net> Most probably your Flex10K100 device has sufficient pins (12x8=96) to create an EXTERNAL tri-state bus, provided that your board isn't laid-out yet, and that you have sufficient (internal) OE to control the 12 tri-state selection lines. In case the device doesn't support 12 OEs, you may consider having a 16-bit wide tristate bus with a 2x8 -> 1x8 mux at the receiving side, and implementing only 6 OEs. Another option is to use an internal EAB as a 12 x 8-bit wide internal RAM. Alex Koegel DSPC Inc.Article: 8429
Carmen Baena Oliva wrote in article <348E7CE1.EF786F5B@cnm.us.es>...
>I'm trying to obtain a combinational multiplier using a Xilinx FPGA.
>Can anybody give me some references about good structures for the
>multiplier?
>
>Thanks in advance.
>Carmen
>
Carmen,
For the fastest - most efficient array multipliers check out the DSP
optimised AT40K architecture. http://www.atmel.com
Martin Mason
Article: 8430G. Herrmannsfeldt <gah@u.washington.edu> wrote in article <66shkq$16uc$1@nntp6.u.washington.edu>... > I am working on a design for an XC4013 with 30 16bit adders, plus > lots of latches, in an XC4013. It should be able to go up to about > 30MHz with the 4013-2, but I am wondering about the dynamic power. > If on the average a signal changes every other clock cycle, > how much power should I expect out of this? > > I don't want a big heatsink and fan for each chip! > > I tried to find this in "the programmable logic data book," but > I didn't find it. I only need rough numbers right now, but I don't > even have that, There's an application note covering the estimation of dynamic power on Xilinx's web site. Go to the XC4000 series page and it should be down the bottom somewhere. Hope this helps, ErikArticle: 8431
I sent Carmen the following reply, but got no answer: Carmen, here are some questions: What array size? What speed is required? Can you pipeline ? Are both factors variables, or is one of them a constant? ( Makes a huge difference) Positive or 2-s complement numbers? If you e-mail me that info, I can point you to a good solution. Peter Alfke, XilinxArticle: 8432
"Peter Fenn" <PeteFenn@iafrica.com> wrote:
:JTAG configuration of Xilinx XC4000E FPGAs?
:
:Urgent! I need to do JTAG configuration of a single FPGA (Xilinx XC4013E).
:Can anyone advise/supply 3rd-party software to do this?
:
:Xilinx documentation details that it's indeed possible to configure a XC4000
:FPGA via its JTAG interface, but Xilinx does not extend their Xchecker
:software to support of this. Any suggestions ?
:
There's a design (and software) to do this for a XC4005, on my
website. URL below, then the "Free Stuff" pointer.
Should work on the XC4KE also.
-- Dave Brooks <http://www.iinet.net.au/~daveb>
PGP public key: finger daveb@opera.iinet.net.au
servers daveb@iinet.net.au
fingerprint 20 8F 95 22 96 D6 1C 0B 3D 4D C3 D4 50 A1 C4 34
What's all this? see http://www.iinet.net.au/~daveb/crypto.html
Article: 8433Acromag Web Surfer wrote: > After a number of customer complaints reporting product malfunctions, > we > investigated and learned that the Xilinx FPGA will not reliable > configure > on power up if the power supply has a slower then 4ms (typically) rise > time > from 3 volts to 4.75 volts. Please send me more details, because there is no real requirement that Vcc should rise that fast. In a daisy-chain of several FPGAs there is a requirement that the master not get ahead of the slaves, which is most easily achieved by interconnecting the INIT lines. Without that interconnection, it is most effective to have a fast Vcc risetime at least a monotonic one.Using the active Low INIT also as the active Low RESET ( I must assume you picked that polarity, and you used Xilinx SPROMs, ) is the proper way to control configuration. There is no combined SPROM and watchdog-timer device. The technological requirements are vey different for these two functions, and both functions are under price- and performance pressure. It is very unlikely that they will ever be combined. Send me details and I promise a fast response, and I can almost guarantee a solution. Most configuration problems are lurking elsewhere on the pc board. Peter Alfke, Xilinx ApplicationsArticle: 8434
> G. Herrmannsfeldt <gah@u.washington.edu> wrote in article > <66shkq$16uc$1@nntp6.u.washington.edu>... > > I am working on a design for an XC4013 with 30 16bit adders, plus > > lots of latches, in an XC4013. It should be able to go up to about > > 30MHz with the 4013-2, but I am wondering about the dynamic power. > > If on the average a signal changes every other clock cycle, > > how much power should I expect out of this? > > > > > I tried to find this in "the programmable logic data book," but > > I didn't find it. Look at page 13-12 It gives you the power consumption in mW per million transitions of: CLBs with minimal or maximal routing, Outputs Longlines and Clocks. The values for clock power are extremely ( ridiculously, and it's my fault !) worst-worst-case, assuming that 100% of all on-chip flip-flops are connected to that clock, which rarely is the case. As a first approximation, derate the clock power by the relative number of flip-flops connected to that clock. You mention latches, I hope you mean flip-flops, since latches are painful to implement and cause terrible simulation problems. Once you have a feel for the power, look up the thermal impedance values for different packages and different airflows ( page 10-8 ), and I am sure that you don't need a fancy heatsink. By the way, XC4013XL at 3.3 V consumes only half the power, and is faster. Peter Alfke, Xilinx ApplicationsArticle: 8435
If you really need an array multiplier, look into the Baugh-Wooley
architecture. It is a good fit with the type of adders that Xilinx has
built-in to the columns of their FPGAs.
Carmen Baena Oliva wrote:
> I'm trying to obtain a combinational multiplier using a Xilinx FPGA.
> Can anybody give me some references about good structures for the
> multiplier?
>
> Thanks in advance.
> Carmen
--
=====================================================================
William Lenihan lenihan3we@earthlink.net
"The greatest barrier to communication is the delusion that
it has already occurred." -- Peter Cummings
=====================================================================
Article: 8436
MA-Boston metro area - several companies including 2 startups (stock/equity incl.)
********FULL RELOCATION provided to qualified candidates!***********
COMPANIES WILL MAKE IT WORTH YOUR WHILE TO CONSIDER RELOCATION IF QUALIFIED!
(see details below - apply at bottom of page....)
****LEAD ASIC VERIFICATION ENGINEER
****METHODOLOGY DESIGN ENGINEER -ASIC DESIGN TOOLS
****COMMS ASIC/BOARD LEVEL DESIGN PROJ LEAD/MGR
****JUNIOR AND SENIOR COMMS ASIC HARDWARE ENGINEERS (x3)
****ASIC/CHIPSET/BOARD-LEVEL HARDWARE ENGINEER
****GRAPHICS ASIC HARDWARE ENGINEER
Most of the following skills are key to these positions:
ASIC tools:
Synopsys, Mentor, Cadence, HPEEsof, VHDL, C/C++, event driven and cycle based simulation, floorplanning, test insertion and ATPG, hardware modeling, ASIC emulation, Spice, harmonic balance, high frequency structure simulation.
ASIC technologies
(CMOS, BiCMOS, Bipolar, GaAs, and multi-chip modules) and digital board families (VME, PCI, x86, 680xx), as well as digital signal integrity and EMI design and analysis techniques.
_______________________
LEAD ASIC VERIFICATION ENGINEER
Job Description:
ASIC design verification for entrepreneurial group designing innovative, new semiconductor test equipment products. Will be responsible for identifying and/or designing and implementing tools and methodologies for ASIC design to ensure high quality designs.
Prefer:
BS EE/CS or related with a minimum of 10 years related experience.Strong knowledge of ASIC design tools including Verilog, VHDL, Synopsys.
Salaries from $60-110K + benefits +FULL RELO
_______________________
METHODOLOGY DESIGN ENGINEER -ASIC DESIGN TOOLS
Job Description:
Will join team responsible for developing new leading edge, state-of-the-art design tools and methodologies which will reduce design/development time and ensure high quality designs. Will be responsible for developing ASIC design tools and methodologies to be used by ASIC design groups; will work with design teams to examine new design paradigms in order to improve ASIC and FPGA design process and results.
Prefer:
BS/MS EE or related education or equivalent experience, plus 1-5+ years related design or design methodology experience required. Familiarity with Verilog, VHDL and/or Synopsys required.
Salaries from $45-80K + benefits +FULL RELO
_______________________
COMMS ASIC/BOARD LEVEL DESIGN PROJ LEAD/MGR
Job Description:
Lead team of Design Engineers assigned the responsibility of providing ASIC and high speed digital designs in Communications Test Equipment.
Actively work with R&D Lab Engineers, Project Managers, and Section Managers to develop and execute a long term Product/Productivity strategy for both ASICs (Digital, Analog, and mixed mode) and Digital PC board designs and design.
Prefer:
*Bachelor's degree or equivalent in Electrical Engineering, Computer
Science or related scientific disciplines.
*Minimum of three years of related (ASIC or high speed digital design) design experience, including working familiarity with requirements capture, behavioral (HDL) modeling, simulation, design rule checks, synthesis, design for test, foundry submission and sign off, semiconductor technology lifecycles, and PC Board lifecycles.
Salaries from $85-130K + benefits +FULL RELO
_______________________
JUNIOR AND SENIOR COMMS ASIC HARDWARE ENGINEERS (x3)
Job Description:
3x ASIC Design Engineers needed to develop ASICs for next-generation products, a Layer 3, IP/IPX Ethernet Switch.
Prefer:
2-7+ years experience ASIC design using VHDL and Verilog coding, including simulation, synthesis, scan insertion, I/O and test benches. You should be familiar with Local Area Networking. You should understand the architecture/implementation of Ethernet switching products in ASIC.
Salaries from $65-115K + benefits +FULL RELO
_______________________
ASIC/CHIPSET/BOARD-LEVEL HARDWARE ENGINEER
Execute the development of the next generation of embedded processor with integrated peripherals for Handheld PCs. Prefer prior experience in design of ASICs for peripheral or accessory products.
Prefer:
BSEE and a minimum 4 years experience in digital systems design and ASIC implementation and systems architecture, with experience using Verilog and Synopsys. Must have implemented and tested ASICs focusing on highly integrated low power, low cost ASIC designs.
Salary from $75-100K + benefits +FULL RELO
_______________________
GRAPHICS ASIC HARDWARE ENGINEER (ENTRY-Senior)
Job Description:
Primary responsibilities include implementation and verification of advanced graphic/ video ASIC designs, as well as validating designs within a system model, creating test benches, and supporting testability using ASIC design and verification tools such as Verilog, Synopsys, C-link test benches, static timing analyziers and ATPG tools.
Prefer:
BSEE or approved equivalent, 1-3 years experience using modern ASIC tools, like Verilog, VHDL, Synopsys, and/or C/C++. Prefer a recent history of developing system validation test benches, modeling ASIC interfaces and interfaces to external components, as well as experience using Sunrise ATPG and Quad Motive. Any knowledge of graphics, video, audio or DSP technology is a big plus!
Salaries from $45-115K + benefits +FULL RELO
_______________________
For immediate consideration email resume to: evanst@sourcesvc.com
Tim Evans
Associate Director
Source Engineering
Voice: 508-366-2600 x3009
FAX: 508-898-0115
email: evanst@sourcesvc.com
http:\\www.experienceondemand.com
Article: 8437Hello, my company is looking for a sub-contractor in an upcoming project. What we are looking for is a electronics company with experience in space projects, which can design several boards for a signal processing device on a satellite. The company should be experienced in the following subjects: - Design of memory boards and extensions - FPGA design - Actel parts - Dealing with requirements for space missions (reliability, environment of launch and orbit, etc.) The company should preferably be located in Europe. Please mail me, if you know a company that fits our requirements. Regards, Sven Muencheberg __ __ -------------------------------------------------------- | |/ / Kayser-Threde GmbH ms@kayser-threde.de | ====< Wolfratshauser Str. 48 Tel.: +49 89/724 95-121 |__|\__\ 81379 Munich, Germany Fax: +49 89/724 95-104Article: 8438
Bill Lenihan <lenihan3we@earthlink.net> wrote in article <348CE6E4.76124EF@earthlink.net>... > Does anyone know of any cases, benchmarks, studies, etc., demonstrating > where the envelope is for FPGA designs done on state-of-the-art PCs vs. > state-of-the-art workstations? Has anyone had a case where a design > couldn't compile on a PC, but could when ported to a workstation? (or > vice-versa?) > > Example: At time "A", Xilinx will strongly recommend that customers > design their biggest chips on workstations, not PCs. Yet, at time "A + 1 > year" they give an official sanction that these chips can be designed on > PCs of X-memory, Y-hard disk space, and Z-CPU speed ...... but PCs with > XYZ specs were not extraordinary at time "A". Do FPGA companies just get > skittish with the PC and lean toward the conservative workstation when > it comes to their newer (and untested in the arena of real designs) > FPGAs? Is there much, or any, risk with sticking to a total-PC design > flow, even when using premium FPGAs? Bill, For me it has always been a matter of tools. The PC tools were always more mature, and out first. Bug fixes were faster.... etc. At least this has been true for Xilinx and Viewlogic.. Austin Franklin darkroom@ix.netcom.comArticle: 8439
Sorry the first posting was uncomplete! Hi everybody, we are currently thinking about a project where we need at least 16 24bit counters in one device? These counters (approx 2MHz) have to be latched syncronously (?) to registers. These registers must be accessed from a microcontroller bus (C167). Unfortunatly, this will be our first project with FPGAs, so we would appreciate some help. The most important step seems to be the choice for a certain device. After looking into some data books I recognized that you have to read every page of the description to make sure not to choose a device that will be unsuitable for the purpose. I discoved the FLEX 8000 family that seems to have most required features. Is there anything I have missed in the description of the device? Is there a better device on the market? Any help appreciated best regards, Michael not THAT one :-) This is posted from my company accountArticle: 8440
Stuart Clubb <s_clubb@die.spammer.netcomuk.co.uk> wrote in article <3492956b.24349988@nntp.netcruiser>... > On Fri, 12 Dec 1997 22:43:10 GMT, muzok@pacbell.net (muzo) wrote: > > >I have a design which uses %55 of a 10K100 and within this design I have 12 8 bit > >registers which need to be selectively loaded into another register, IOW, I have > >an 8 bit bus from which 12 8 bit registers hang. Because 10K doesn't have > >internal tristates, I have a big mux which selects the register. Currently this > >mux is the bottleneck in my design which limits my speed to 11 MHz in a -3 part. > > What speed do you need? > > >Any ideas how to increase the speed in this design? The design is in Verilog. > > So am I right in saying that a 12 to 1 multiplexer, register to > register (hence, including routing) takes 90 ns? As Altera claims a 16 > to 1 mux takes 9.4 ns in a -3 speed grade there are three > possibilities: > 1 > You need a new synthesis tool (which are you using?) > 2 > There's more than the mux in your critical path. (how many layers of > logic and routing in the path?) > 3 > Altera isn't up to it, has a "printing error" in it's data sheets, and > you need a different FPGA. > > 1 & 2 seem more probable. > > Stuart > -- > For Email remove "die.spammer." from the address Hi all, My intent is not to start an Altera vs. Xilinx war but my experience is similar to the original poster so I think its worth posting. I hope you will find it professional and informative. First let me say that if you are getting 11MHz, then there is likely more than one problem and you can probably improve the design at the HDL and layout level (also check fanout) to get the speed up. That said, I had real trouble getting the Altera 10K100 to operate at the speed I needed it (50MHz). I had been told by Altera that 66MHz would be possible but after working with the part for a while, we had to back off our expectations to 50MHz. After weeks of working closely with a very knowledgeable Altera FAE, we could not get the part to run beyond 40 to 42 MHz. We made many modifications and kept all the logic levels to 3 or less. After a while, it sunk in that the 10K100 architecture is just not up to the task. The main problem is that the row to row delay is FAR too prohibative. You can work all you want to partition the design so as to minimize the row to row communication, but in a large design (why else would you use a 10K100), the rows can never be totally independent. If you're going to burn 10 to 15 ns just to go from one row to another, then you're dead. You can't even have one level of logic if the level is on a different row from one of the inputs and the output. I actually had "latch to latch" delays (from input latch to latch in another row) be greater than 20 ns! This is so incredible from a part that is touted as being 66MHz capable, I'll say it again -- I had latch to latch (no logic in between) delays greater than 20 ns (almost 25 ns). And I'm not talking about going ALL the way across the part. Another problem -- the input setup, clock to out and output enable times of a 10K100 I/O block are bad enough so that 2 10K100 could not communicate with each other at 50MHz and meet setup and hold times. To make a long story short, I switched to a Xilinx 4085 -1 and the existing VHDL ran at 50MHz no problem. Another advantage that the 4085 has over the 10K100 is you can use both an I/O Block Input and Output latch for a bi-di signal. Altera only has one latch in the IOB so you can either latch the input or the output, but not both (in the IOB). Another advantage if you use the on chip RAM, is that Xilinx ram is dual ported and the Altera ram is not so you can't build a fifo using Altera ram if you need to be able to read and write the fifo on the same clock. Since making the switch to the 4085, I've added much logic (so much that the design probably would not fit in a 10K100 anymore) and the part still runs at 50 MHz. My opinion is that the 10K100 is ok for 33MHz type designs -- maybe 40 MHz if you work very hard. Beyond that, use Xilinx. Of course, you get what you pay for -- the 4085 is more expensive than the 10K100. -- Rich Iachetta IBM Corporation iachetta@us.ibm.comArticle: 8441
>
> After reading almost 95% of the mails on the MTBF,
> i know everything but could not find a good article
> of how to simulate a FF and extract the MTBF out of it.
>
There are actually THREE problems:
(1) Figuring out what *your* reliability goals
really are. Unlike what appears in this
newsgroup, your goals are very likely
going to include an aggregate probability
of failure of a large number of flipflops,
not P(fail) of a single flipflop.
Something like:
Our beloved customer American Airlines
has a room containing 500 systems, and
each system contains 32 of our chips,
and each of our chips contains one
flipflop that could go metastable.
What is the required MTBF of each individual
flipflop, to guarantee that there is a
probability of > 75% that ABSOLUTELY NONE
of the (500*32*1) = 1600 flipflops at
American Airlines, ever has a single
metastability failure, within three years
of continuous operation.
(2) Finding the metastability parameters TAU and
T0 of your particular flipflop, including
parasitic resistance and capacitance effects.
(3) Writing an equation for (Pfail,aggregate)
in terms of TAU, T0, (the number of flipflops
in your aggregate), (the clock frequency),
(the data frequency), (the lifetime of
the mission-critical system, e.g. 3 years),
and so forth.
Item (2) is the easiest of the three. The Ph.D. thesis
by Portmann (cited earlier on this group) shows how.
So does the IEEE J.SSC paper by Hohl. Sorry I forget
the exact issue# and page #s, you can look it up,
there aren't too many papers by authors of that name.
-M.J.
Article: 8442One raw of a Flex 10k100 contains enough LCELL for a 12*8 multiplexer. Have you tried to specify a raw for this multiplexer? You can do this before compilation by selecting it in the graphic editor and do "assign pin/raw/...". Select an empty raw. You can also do this after compilation with the floor plan editor but you have at least 12*8 LCELL to move with your mouse! If you can put all the LCELLs of the multiplexer inside one raw, you should be able to reach 35-40MHz with a 10K100-3. I also recommend you to apply Daniel Lang's method. Regards, Emmanuel ----------------------------- Design Engineer, SchlumbergerArticle: 8443
>Please send me more details, because there is no real requirement that >Vcc should rise that fast. The rise needs to be monotonic. This implies it must be very noise-free. Peter. Return address is invalid to help stop junk mail. E-mail replies to z80@digiXYZserve.com but remove the XYZ.Article: 8444
In article <01bd097f$6ebe3e20$6d733509@iachetta>, "Richard Iachetta" <iachetta@us.ibm.com> writes: >My opinion is that the 10K100 is ok for >33MHz type designs -- maybe 40 MHz if you work very hard. >Beyond that, use >Xilinx. Of course, you get what you pay for >-- the 4085 is more expensive >than the 10K100. > >-- >Rich Iachetta >IBM Corporation >iachetta@us.ibm.com I have been programming -3 Altera FLEX10K's for the past 1 1/2 years or so, and I would have to say that if you're trying to do anything the least bit complicated 30-40 MHz is about the best you can expect, anything faster (especially on the higher gate count chips) will require significant fine tuning to work. That said, I have found that the new (and very expensive) -2 and -1 speed grades are able to do 50 MHz applications without significant fine tuning. --Jeff BerryhillArticle: 8445
Michael Schumacher wrote: > Sorry the first posting was uncomplete! > > Hi everybody, we are currently thinking about a project where we need > at least 16 24bit counters in one device? These counters (approx 2MHz) > > have to be latched syncronously (?) to registers. These registers > must be accessed from a microcontroller bus (C167). > Hi, Michael, You obviously need a programmable device with at least 17 x 24 flip-flops, or do you need 32 x 24 flip-flops ? In either case, you need an FPGA, since CPLDs don't have enough flip-flops. Let's first describe a straightforward design: The smallest member of the Xilinx XC4000 family to use for that design is the XC4008E which has an 18x18 array of logic blocks, and each of your counters fits into 13 of the 18 blocks per column. So you have 30% of the chip left over for control etc. You might also find the horizontal Longlines convenient to collect the counter values and route them to their destination. The XC4010E has a 20x20 array, again with two flip-flops per block. If you really need to store each counter individually, you need a larger device. The 5-V XC4005E, XC4010E, XC4013E and XC4020E also come in 3.3-V versions called XC4005XL etc. Now fro the more elegant or sneaky design: Since you mentioned the very slow speed of 2 MHz, there is a way to do the counting in the distributed RAM on the XC4000E devices, by using the RAMs as shift registers plus a serial incrementer. That would reduce the incremental cost per counter to two blocks,(one block as a 32-bit RAM, the other one as the adder/carry logic.) The common overhead would be less than a dozen blocks, and the whole design would easily fit into half of the smallest device of our family, the XC4003E with its 100 blocks. But it would need a 48 MHz oscillator ( $2.-) and a little bit of creative thinking. If you accept the challenge, I'll help you. Even in German. Peter Alfke, Xilinx ApplicationsArticle: 8446
Jeff Berryhill <berryhill@hep.uchicago.edu> wrote in article <EL901u.1D5@midway.uchicago.edu>... > In article <01bd097f$6ebe3e20$6d733509@iachetta>, > "Richard Iachetta" <iachetta@us.ibm.com> writes: > > >My opinion is that the 10K100 is ok for > >33MHz type designs -- maybe 40 MHz if you work very hard. > >Beyond that, use > >Xilinx. Of course, you get what you pay for > >-- the 4085 is more expensive > >than the 10K100. > > > >-- > >Rich Iachetta > >IBM Corporation > >iachetta@us.ibm.com > > I have been programming -3 Altera FLEX10K's for the past 1 1/2 years > or so, and I would have to say that if you're trying to do anything > the least bit complicated 30-40 MHz is about the best you can > expect, anything faster (especially on the higher gate count chips) > will require significant fine tuning to work. > > That said, I have found that the new (and very expensive) -2 and -1 > speed grades are able to do 50 MHz applications without significant > fine tuning. > > --Jeff Berryhill Ok. My experience was this summer (June/July) and we were using the fastest 10k100 vs. the fastest 4085 at the time. Both companies have come out with new stuff since. When you say you've been using FLEX10K's, which ones did you use? The 10k100? The smaller they are, the faster they run. It would take a lot to convincing me that the "new" 10K100's can do 50MHz. If they can, that's great, but having been burned, I'm a little skeptical. -- Rich Iachetta IBM Corporation iachetta@us.ibm.comArticle: 8447
Howdy, Does anyone know of any 5 volt serial PROMs larger than the XC17256D, that can be used with XC4000EX FPGAs? I notice that Xilinx has 512K and 1M bit PROMs in a 3.3V package, but apparently nothing so far in 5V. -- -DuaneArticle: 8448
I take it that you are builting the MUX out of case statements? I did a quick run in GDF and I was able to achieve 50MHz in a 96 to 8 mux with LPM_MUX. Blacking box your mux in verilog and instantiaing it in Max+Plus design file with lpm_mux would be a quick workaround. ying@csua.berkeley.edu In article <3491bc91.80619134@news.walltech.com>, muzo <muzok@pacbell.net> wrote: >hi, >I have a design which uses %55 of a 10K100 and within this design I have 12 8 bit >registers which need to be selectively loaded into another register, IOW, I have >an 8 bit bus from which 12 8 bit registers hang. Because 10K doesn't have >internal tristates, I have a big mux which selects the register. Currently this >mux is the bottleneck in my design which limits my speed to 11 MHz in a -3 part. >Any ideas how to increase the speed in this design? The design is in Verilog. > >Are there any ways to get more performance by using the floor plan tool ? I am >already using FAST synthesis and maximum speed in the compiler options. > >thanks > >muzo > >WDM & NT Kernel Driver Development Consulting <muzok@pacbell.net> -- ----------------------------------- http://www.csua.berkeley.edu/~yingArticle: 8449
In article <670s6d$jv7$1@news.netvision.net.il>,
Alex Koegel <alex.koegel@dspis.co.il> wrote:
>In article <3491bc91.80619134@news.walltech.com>,
> muzok@pacbell.net (muzo) wrote:
>>hi,
>>I have a design which uses %55 of a 10K100 and within this design I have 12 8
>bit
>>registers which need to be selectively loaded into another register, IOW, I
>have
>>an 8 bit bus from which 12 8 bit registers hang. Because 10K doesn't have
>>internal tristates, I have a big mux which selects the register. Currently
>this
>>mux is the bottleneck in my design which limits my speed to 11 MHz in a -3
>part.
>>Any ideas how to increase the speed in this design? The design is in Verilog.
>>
>>Are there any ways to get more performance by using the floor plan tool ? I
>am
>>already using FAST synthesis and maximum speed in the compiler options.
>>
>>thanks
>>
>>muzo
>>
>>WDM & NT Kernel Driver Development Consulting <muzok@pacbell.net>
>
>Another option is to use an internal EAB as a 12 x 8-bit wide internal RAM.
>
>Alex Koegel
>DSPC Inc.
1/ I hvae successfully used this idea for readback of CPU registers. Ie
shadow the registers in the EAB ram.
2/ For fast access to yoru registers, under certain circumstances, you can
just copy them into a shift register, and shift the whole thing out.
3/ One thing to watch out for in the mux's in the 10K family, is that there
is a feature that the LCELL() primitive is used, so the final output
stage will not be merged into a following ff. So in the instance of that
mux set CASCADE_CHAIN = "OFF" (just look at the lpm_mux code to see why
this works). Basically this cuts a full logic level off.
Eric Pearson
--
Eric Pearson -- Focus Systems -- Waterloo, Ontario
ecp@focus-systems.on.ca (519) 746-4918
"We Engineer Innovative Imaging Solutions"
Site Home Archive Home FAQ Home How to search the Archive How to Navigate the Archive
Compare FPGA features and resources
Threads starting:
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