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
I have seen such a system implemented. As you mentioned in your posting, a Xilinx FPGA device can perform a readback operation. With a small amount of extra logic, the FPGA device can perform a self-check. It reads back its bitstream and compares it against the values stored in the configuration PROM. A second PROM is required to hold the mask values, as the readback bitstream contains the values of the internal logic block and I/O block flip-flops (these would not be constant). If the readback and the configuration values don't match, then the FPGA causes a re-boot. Of course, this doesn't take into account bit-upsets in the configuration PROM. The Xilinx software can create a PROM file with the MakeMask option in Makebits. The details are a bit sketchy (I saw this about five years ago) but it has been done successfully. E-mail me directly if you have more question. As a side note, SRAM cells are used not only in FPGAs but also in most CPLD devices to hold configuration bits that require decreased capacitive loading. These SRAM cells are not like those used in SRAM memory devices. In SRAMs, an extremely-high impedance resistor to VCC and ground permits faster write access and the expense of VCC stability. In most FPGAs/CPLDs, there is lower-impedance resistor to VCC and ground the provides increased cell stability. The write performance is not important in FPGAs/CPLDs. The difference is about a million-to-one improvement in stability. Steven Knapp E-mail: optmagic@ix.netcom.com Programmable Logic Jump Station: http://www.netcom.com/~optmagic Hans Tiggeler <ees1ht@ee.surrey.ac.uk> wrote in article <5bkroa$ru4@info-server.surrey.ac.uk>... > Just out of interest has anybody implemented a continuous checking watchdog on > a Xilinx FPGA? > > Xilinx provides a configuration read back option, using the same device or > some external CRC checker the system can continuously be checked for > configuration errors due to SEU. The error latency will of course depends on > the configuration size and read back speed (1 or 8MHz?). To improve on this > the user can for example add redundant logic and boundary scan techniques. > Total dose tolerance can be improved by different packaging techniques such as > RADPACK (tm) and SEL can be addressed with a fast electronic fuse. > > The unknown of course in this system is, can you damage a Xilinx device by > randomly changing configuration bits?Article: 5051
Hello everybody, Does anyone know of a meta-assembler which might be cheaper/newer than the one offered by Hilevel Technology Inc.? I want to generate programs for an ASIC embedded sequencer with a custom instruction set. For those who aren't familiar with a meta-assembler, it's a software product which can be used to define instruction sets for custom sequencers. Years ago, I used it for 2900 family sequencers, now I would like to use something like it in an ASIC design for an embedded sequencer, similar to a co-processor. The one from Hilevel (HALE) is old, runs on DOS and they quote $2300. I think I can do better than this and hope somebody can help me. Please reply directly to tloftus@hns.com because I don't get to these newsgroups very often. Tom -- Thomas J. Loftus | Electrical Design Automation Group phone: (301) 548-1916 | Hughes Network Systems email: tloftus@hns.com | 11717 Exploration Lane FAX: (301) 212-2099 | Germantown, MD 20876 All statements reflect my personal views and not necessarily that of HNS.Article: 5052
In article <01bc03d5$5f2f8b40$6e0db780@Rich>, rich.katz@gsfc.nasa.gov says... > >designs of this type have been done before and we called it a poor man's >edac Why? voting systems are used in a lot of other engineering disciplines Sometimes the ability of handling multiple bit errors is more desirable than the two extra memory chips (or add extra logic). Ever tried to correct two errors using a modified hamming code? (and it was done when they had 1 kbit memories <- not a typo). didn't >even need an fpga, the 54ls253 makes a great voter w/ only one half and by >adding an inverter the other half can function as an error indicator. and >the state machine wasn't tough and the read cycle had to be transformed >into a read-modify-write; this was important for consistent software >timing/verification in the presence of errors. My point exactly,the beauty of this system is it's simplicity, perhaps we wouldn't have so many failures if we used more "poor man's" techniques. The FPGA was chosen to include glue and error counting logic. > >rk > >p.s. and i did a 16-bit flow through edac in a 1020 which had the ability >(when using x8 ram chips) to swap out failed chips in the real memory and You sounds like a very good engineer, Hans.Article: 5053
Hans Tiggeler <ees1ht@ee.surrey.ac.uk> wrote in article <5bnsbh$7et@info-server.surrey.ac.uk>... > In article <01bc03d5$5f2f8b40$6e0db780@Rich>, rich.katz@gsfc.nasa.gov says... > > > >designs of this type have been done before and we called it a poor man's > >edac > Why? voting systems are used in a lot of other engineering disciplines in these 'prehistoric' days you couldn't go to the store and get an edac chip and fpga's weren't invented. it used dumb, simple brute force techniques. to get speed, you had to do an asic ($, time) and a discrete edac implementation was too slow. tmr is very fast, a single gate delay (54ls253). and tmr systems are used all the time, and very handy for making soft fpga memory elements effectively hard. these is reasonable for fpga's, where there are plenty of gates. implementing this at the discrete level, i.e., for seu-protection of 54lsxx circuits, was kind of bulky and using a newer technology (rad-hard sandia cmos) was the way we went. rkArticle: 5054
Greetings... our customer generates JEDEC files for boards we build for him, and we do the programming of Lattice in-system-programmable parts on the board. I've got an older version of the code in .JED format, and a newer version he just created from scratch after losing the original source logic code (.LIF or .LDF files). The newer version naturally doesn't work, and we don't know exactly what's in the older one. Is there a way to reverse-compile the working .JED file to something humanly-understandable that we can tweak, so he can re-compile it again? I'm not too keen on the idea of taking the fuse map and writing out on a sheet of paper to see what fuses are blown and open, and then figuring out the logic from that. The part in question is the ispLSI1032... fortunately we haven't had this problem (yet) with the much larger ispLSI3256 he's also using on the board. Help? Tom Moeller Encore ElectronicsArticle: 5055
How do you set the swap space in Xilinx ver 5.1.0 on DOS 6.22 ? Many thanks in advance. JoeArticle: 5056
Hello, I am a computer science major at Williams College. I am working on a computational number theory problem for one of my professors. I have implemented it in C and optimized it to death and am now considering building a special purpose computer to speed it up. So, along with everyone else, I'm trying to find the best balance between a limited budget and speed. I have a design for discrete chips that should run about 5 times as fast as a P-120. The cost is $300-$400, which isn't bad, but the circuit involves nearly 200 chips, with about 3000 pins. That's a lot of wiring. The algorithm involves: Two parallel 128-bit additions of internal-only data Storage for the result (256 bits) Two 48-bit conditional inversions using XOR gates Two priority encodings across 48 bits Some counters, flip-flops and simple control circuitry Around 50 bits of output It's not a particularly complicated circuit, just large. I've been looking at Motorola's MPA1036 chips -- I think that two of these would do the job, and leave plenty of room for neat parallel addition algorithms and such. I think I would be able to borrow an EPROM programmer to store the configuration. But I'm pretty new to digital hardware in general, and especially to FPGA's. So here are my questions: Are there faster/cheaper/better chips than the MPA series for this type of computation? I have nothing available to me except computers -- no commercial design software. Reprogrammability is a plus, since I'm new and the chances I'll get everything right the first time are slim. Is there a way to import a schematic from Chipmunk's diglog into a layout program such as Motorola's? Does anybody have an estimate of how long a 128-bit ripple-carry addition would take? (i.e., is it worth it to use FPGA's?) Any answers, advice, or general comments would be useful. Thanks very much. Ben Chaffin 98bcc@williams.eduArticle: 5057
Hello, I am a computer science major at Williams College. I am working on a computational number theory problem for one of my professors. I have implemented it in C and optimized it to death and am now considering building a special purpose computer to speed it up. So, along with everyone else, I'm trying to find the best balance between a limited budget and speed. I have a design for discrete chips that should run about 5 times as fast as a P-120. The cost is $300-$400, which isn't bad, but the circuit involves nearly 200 chips, with about 3000 pins. That's a lot of wiring. The algorithm involves: Two parallel 128-bit additions of internal-only data Storage for the result (256 bits) Two 48-bit conditional inversions using XOR gates Two priority encodings across 48 bits Some counters, flip-flops and simple control circuitry Around 50 bits of output It's not a particularly complicated circuit, just large. I've been looking at Motorola's MPA1036 chips -- I think that two of these would do the job, and leave plenty of room for neat parallel addition algorithms and such. I think I would be able to borrow an EPROM programmer to store the configuration. But I'm pretty new to digital hardware in general, and especially to FPGA's. So here are my questions: Are there faster/cheaper/better chips than the MPA series for this type of computation? I have nothing available to me except computers -- no commercial design software. Reprogrammability is a plus, since I'm new and the chances I'll get everything right the first time are slim. Is there a way to import a schematic from Chipmunk's diglog into a layout program such as Motorola's? Does anybody have an estimate of how long a 128-bit ripple-carry addition would take? (i.e., is it worth it to use FPGA's?) Any answers, advice, or general comments would be useful. Thanks very much. Ben Chaffin 98bcc@williams.eduArticle: 5058
Ben Chaffin wrote: > > Hello, > I am a computer science major at Williams College. I am working > on a computational number theory problem for one of my professors. I have > implemented it in C and optimized it to death and am now considering > building a special purpose computer to speed it up. So, along with > everyone else, I'm trying to find the best balance between a limited > budget and speed. I have a design for discrete chips that should run > about 5 times as fast as a P-120. The cost is $300-$400, which isn't bad, > but the circuit involves nearly 200 chips, with about 3000 pins. That's a > lot of wiring. The algorithm involves: > > Two parallel 128-bit additions of internal-only data > Storage for the result (256 bits) > Two 48-bit conditional inversions using XOR gates > Two priority encodings across 48 bits > Some counters, flip-flops and simple control circuitry > Around 50 bits of output > > It's not a particularly complicated circuit, just large. I've > been looking at Motorola's MPA1036 chips -- I think that two of these > would do the job, and leave plenty of room for neat parallel addition > algorithms and such. I think I would be able to borrow an EPROM > programmer to store the configuration. But I'm pretty new to digital > hardware in general, and especially to FPGA's. So here are my questions: > > Are there faster/cheaper/better chips than the MPA series for > this type of computation? I have nothing available to me except computers > -- no commercial design software. Reprogrammability is a plus, since I'm > new and the chances I'll get everything right the first time are slim. > > Is there a way to import a schematic from Chipmunk's diglog into > a layout program such as Motorola's? > > Does anybody have an estimate of how long a 128-bit ripple-carry > addition would take? (i.e., is it worth it to use FPGA's?) > > Any answers, advice, or general comments would be useful. Thanks > very much. > > Ben Chaffin > 98bcc@williams.edu Many of the other vendors sell their software at prices starting around $1,000. If you are not intending to do many designs, it is probably not cost effective for you to go and buy the software. However, since you are essentially doing the work for an educational institution, you may qualify for one of the special educational programs many of the vendors have. As you may already be aware, a ripple adder is the slowest implementation. There are a number of carry speed up schemes that can help (at the cost of more cells used). For smaller adders, the amount of improved performance is generally eclipsed by the longer delays in each adder bit for FPGAs. However for big adders, there is plenty to gain by doing some form of radix addition with carry between the digits. Where you are apparently doing this as a numerical processor that doesn't really have any timing constraints to the outside world, I would heavily pipeline the design to pick up the throughput (that includes internally pipelining the 128 bit adders). If the process is one that can be performed by splitting into identical parallel paths, you can do it using a bunch of identical parallel bit-serial processors (see my papers on my website). For a 128 bit straight ripple adder your looking at somewhere around 150-200ns (rough estimate, really depends on speed grade, architecture, etc.). If you do it with bit serial hardware, the bit rate can often be better than 100MHz. So for a 128 bit add, it'll take somewhere around 1-2us with less than 1/100th of the logic otherwise required. Generally speaking, you can expect performance improvements of anywhere from 10x to 1000x over software by using an FPGA. As far as better/faster etc., I find look-up table type devices a little easier for the uninitiated. They generally do not have as many functionality and routing restrictions at the cell level. The fine grain architectures can often be pushed a little faster, but you have to tailor the design to the architecture to gain that advantage (not likely to happen on your first FPGA design). Of course, this is not much of an issue if the vendor has decent libraries and you are not shooting for high performance (relative to maximum clock speeds) or high density. For the average user, it really comes down to personal taste (and usually your employer's preference), and what you have experience with. I am not familiar with chipmunk software. Odds are you won't be able to translate between it and any of the FPGA fitters unless it outputs in a format compatible with the more standard capture tools and it can read the library for the target device. Perhaps a part like the Lattice Semiconductor ISP parts will do the trick. I think Lattice may distribute a free basic entry tool for some of their CPLDs. Those parts are EEPROM based, are programmed in circuit via a 5 wire interface, and the programmer is any PC. They are more limited in the number of registers than an FPGA, but if you do it bit serially, there should be no problem fitting. Some of the vendors you may want to look at are: Actel, Altera, AMD, Atmel, Crosspoint, Cypress, Lattice, Lucent, QuickLogic, National Semiconductor, Xilinx, and Zycad (sorry if I missed anyone). Actel, Altera and Xilinx are the 600 lb gorillas in the industry. Motorola is a newcomer, so you'll find the experience level is less. Hope this helps a little. -Ray Andraka, P.E. Chairman, the Andraka Consulting Group 401/884-7930 FAX 401/884-7950 mailto:randraka@ids.net http://www.ids.net/~randraka/ The Andraka Consulting Group is a digital hardware design firm specializing in high performance DSP designs in FPGAs. Expertise includes reconfigurable computers, computer arithmetic, and maximum performance/density FPGA design.Article: 5059
David Baker wrote: > > Available now. Richard Watts Associates RAW8052 core targeted > to Altera FLEX10K FPGA. > > Occupies approx 1/3 of the logic of a 10K100 part. > > Netlist (VHDL, Verilog) or VHDL source available for seamless > migration to ASIC. > > To obtain a free evaluation version check our page at > www.evolution.co.uk or contact richardw@evolution.co.uk. How much is it, and how fast does it run ? -- =============== Serious Design Tools for uC and PLD ================ = Optimising Structured Text compilers for ALL 80X51 variants = Reusable object modules, for i2c, SPI and PLDwire bus interfaces = Safe, Readable & Fast code - Step up from Assembler and C = Emulators / Programmers for ATMEL 89C1051, 2051, 89C51 89S8252 89C55 = HW : IceP2051, Emul517C, TRICE-31X, PC CANport = for more info, Email : DesignTools@xtra.co.nz Subject : c51ToolsArticle: 5060
Thomas C. Jones wrote: > > Does anyone know of a pin-compatible replacement to the PEEL22CV10A > offering greater routing capacity and resources? > > -Tom > tjones@aspect.com Look at the ATMEL ATV750BL, ( Dip24) this has 20 FF's, 10 buried, and Toggle options. All FF's have ASYNC clocks, and OE terms. This is the smartest 24 pin PAL we have found, and we have fitted i2c slave designs into ir. -- =============== Serious Design Tools for PLD and uC ================ = HDL PLD design libraries, Source & Sim - HDLapIO and HDL_Pro = Chip Vendor independant designs & tools = for more info, Email : DesignTools@xtra.co.nz Subject : HDLapPacksArticle: 5061
David Erstad <erstad@ssec.honeywell.com> writes: > I think most people and designing and flying space systems would be very > uncomfortable with the concept that a part would latch up and it would > be "OK". Clementine did, IIRC. (The rad-hard 1802 watched the current drawn by the R3000, and if ti started to rise, reset the processor.) And the reason for the final, ugh, "off-nominal" behaviour had nothing to do with this. JanArticle: 5062
Ben Chaffin wrote: > > Hello, > I am a computer science major at Williams College. I am working > on a computational number theory problem for one of my professors. I have > implemented it in C and optimized it to death and am now considering > building a special purpose computer to speed it up. So, along with > everyone else, I'm trying to find the best balance between a limited > budget and speed. I have a design for discrete chips that should run > about 5 times as fast as a P-120. The cost is $300-$400, which isn't bad, > but the circuit involves nearly 200 chips, with about 3000 pins. That's a > lot of wiring. The algorithm involves: > > Two parallel 128-bit additions of internal-only data > Storage for the result (256 bits) > Two 48-bit conditional inversions using XOR gates > Two priority encodings across 48 bits > Some counters, flip-flops and simple control circuitry > Around 50 bits of output > > It's not a particularly complicated circuit, just large. I've > been looking at Motorola's MPA1036 chips -- I think that two of these > would do the job, and leave plenty of room for neat parallel addition > algorithms and such. I think I would be able to borrow an EPROM > programmer to store the configuration. But I'm pretty new to digital > hardware in general, and especially to FPGA's. So here are my questions: > > Are there faster/cheaper/better chips than the MPA series for > this type of computation? I have nothing available to me except computers > -- no commercial design software. Reprogrammability is a plus, since I'm > new and the chances I'll get everything right the first time are slim. > > Is there a way to import a schematic from Chipmunk's diglog into > a layout program such as Motorola's? > > Does anybody have an estimate of how long a 128-bit ripple-carry > addition would take? (i.e., is it worth it to use FPGA's?) > > Any answers, advice, or general comments would be useful. Thanks > very much. > > Ben Chaffin > 98bcc@williams.edu Ben, Check out the XILINX kits at http://www.erols.com/aaps they are very reasonable and include a decent utility board. The board takes chips which can implement up to 16,000 gates! More powerful boards are also available. I designed the boards and use them for many of my FPGA projects. We are offering substantial discounts on XILINX FOUNDATION software. This package includes VHDL, SCHEMATIC CAPTURE, POST ROUTE simulation and the fitter/router software. The software is packaged extremely well and X84 board comes with C templates which make implementing the designs and real time control as painless as possible, and the board is set up so as to be able to take any FPGA from 2k to 16K gates in an 80 pin PLCC package with all pins going to 20 pin connectors which are set up in the HP logic analyzer format. It really is a helpful tool, and I would reccomend these kits highly to both experienced and unexperiences FPGA developers. The board and C drivers are also available seperatly for $275.00 and come with the C drivers and a 5202 FPGA which is socketed and can be replaced with chips upto 16,000 gates. The C drivers come with download code so that the design can be loaded from the PC or via EPROM or via XILINX download cable if the board is used standalone, outside the PC. Richard Schwarz RichardArticle: 5063
I'm currently working on a project that requires continuous programming and reprogramming of a Xilinx XC4000 FPGA board. Due to some strange setup requirements, we are not going to be able to directly use the xchecker application and cable to program the board. Does anyone have any information regarding the technical details of programming Xilinx boards? Any help would be greatly appreciated, -David Hirschfield +-===========================================================================-+ | --== e-mail ==-- | | _/_/_/ _/_/ _/ _/ _/_/_/ dch+@andrew.cmu.edu | | _/ _/ _/ _/ _/ _/ _/ or | | _/ _/ _/_/_/_/ _/ _/ _/_/ cddch@paleo.giss.nasa.gov | | _/ _/ _/ _/ _/_/ _/ | | _/_/_/ _/ _/ _/ _/_/_/ --== WWW ==-- | | http://www.contrib.andrew.cmu.edu/~dch/ | +-===========================================================================-+Article: 5064
Anyone had any problems using xact v6.0.1 under win3.x when using smartdrive? I've just found that loading smartdrive with the /x option (no write behind cache) fixed some/all of my crashing problems experienced when compiling my source. Stu.Article: 5065
In article <32DE9AC9.79CA@hns.com>, "Thomas J. Loftus" <tloftus@hns.com> writes >Hello everybody, > >Does anyone know of a meta-assembler which might be cheaper/newer than >the one offered by Hilevel Technology Inc.? I want to generate >programs for an ASIC embedded sequencer with a custom instruction set. > >For those who aren't familiar with a meta-assembler, it's a software >product which can be used to define instruction sets for custom >sequencers. Years ago, I used it for 2900 family sequencers, >now I would like to use something like it in an ASIC design for >an embedded sequencer, similar to a co-processor. > >The one from Hilevel (HALE) is old, runs on DOS and they quote $2300. >I think I can do better than this and hope somebody can help me. > >Please reply directly to tloftus@hns.com because I don't get to >these newsgroups very often. You will find a free table-driven cross-assembler on the Simtel archive. Look for TASM301.zip. You can create your own table for any instruction set. There is a similar commercial product called Cross-32, which is quite inexpensive ($500, perhaps). Cross-32 is suitable for 8, 16 and 32-bit processors, TASM might only be 8-bit. Leon -- Leon Heller, G1HSM leon@lfheller.demon.co.uk Tel: +44 (0) 118 947 1424 (home) +44 (0) 1344 385556 (work)Article: 5066
ees1ht@ee.surrey.ac.uk (Hans Tiggeler) wrote: >In article <01bc0161$d44246a0$6e0db780@Rich>, rich.katz@gsfc.nasa.gov says... >> >>In the near term, there will be an increasing number of NASA missions >>flying DRAMs, including some long-term missions like Cassini to Saturn. >> >>The easiest codes to work with are the Hamming codes, with most >>implementations correcting single-bit errors and detecting double-bit >>errors. These codes permit easy correction on the fly and are very simple >>to implement (and if one has time it can be done if FPGAs). >>If not, then Reed-Solomon or some other code would be >>required, which are more difficult to decode. > >Given the price of S/DRAM memories why not use a majority voting system such as TMR (Triple >Modular Redundancy). The system can handle multiple bit upsets and is very easy to implement. I >have currently a system running in an Actel1020 (17% used!). Of course the drawback is that you >have to either triplicate your memory or add a statemachine to perform three memory read cycles >per CPU read cycle. > Or use 3 processors / FPGA's to give you fault tolerance there too. ;-)Article: 5067
Ben Chaffin <98bcc@williams.edu> wrote: >Hello, > I am a computer science major at Williams College. I am working >on a computational number theory problem for one of my professors. I have >implemented it in C and optimized it to death and am now considering >building a special purpose computer to speed it up. So, along with >everyone else, I'm trying to find the best balance between a limited >budget and speed. I have a design for discrete chips that should run >about 5 times as fast as a P-120. The cost is $300-$400, which isn't bad, >but the circuit involves nearly 200 chips, with about 3000 pins. That's a >lot of wiring. The algorithm involves: > > Two parallel 128-bit additions of internal-only data > Storage for the result (256 bits) > Two 48-bit conditional inversions using XOR gates > Two priority encodings across 48 bits > Some counters, flip-flops and simple control circuitry > Around 50 bits of output > > It's not a particularly complicated circuit, just large. I've >been looking at Motorola's MPA1036 chips -- I think that two of these >would do the job, and leave plenty of room for neat parallel addition >algorithms and such. I think I would be able to borrow an EPROM >programmer to store the configuration. But I'm pretty new to digital >hardware in general, and especially to FPGA's. So here are my questions: > > Are there faster/cheaper/better chips than the MPA series for >this type of computation? I have nothing available to me except computers >-- no commercial design software. Reprogrammability is a plus, since I'm >new and the chances I'll get everything right the first time are slim. > > Is there a way to import a schematic from Chipmunk's diglog into >a layout program such as Motorola's? > > Does anybody have an estimate of how long a 128-bit ripple-carry >addition would take? (i.e., is it worth it to use FPGA's?) > > Any answers, advice, or general comments would be useful. Thanks >very much. > > Ben Chaffin > 98bcc@williams.edu > First of all, I'm gonna speak here about 3 vendors only. That's not because I think they are the fastest, cheapest or better than the rest, it's just that these are the ones I use. One of the vendors for which I have no experience is Motorola so I cannot comment there but I do use Actel, Altera and Cypress daily. The biggest factor in the cost of a discrete vs FPGA solution is probably gonna be the tools. The cheapest way into the wonderful world of programmable silicon that I know of is the Cypress Warp VHDL.. Get in touch with a Cypress rep & you should be able to get on a 1 day VHDL training course with a kit to take home for $99. There is no limit to the size of parts you can use with this budget kit, if Cypress make 'em, you can use 'em. There is, however, no schematic entry of any kind, you'll have to rewrite your entire design in VHDL. If you wanna get the job done fast, then (IMHO) you should use Altera. Their Max+Plus 2 system is all you need. The Flex 10K series is probably the right place to start for your design, the compiler will tell you which part you need. The other big advantage of Altera is price (of silicon) their price cutting is quite astonishing. Take a look at their web site / press releases to get an idea. If you do go for Altera don't use their Configurator Proms, use Atmel's E^2 parts. Actel would probably give you the greatest speed of operation of these three vendors. Their ACT3 parts are damn fast but I guess you would be better with the 3200DX family 'cos those have SRAM bits built in. The tools are expensive but for entry you can use OrCad who definitely do give academic discount & give you your best chance of Chipmunk import. In summary, the advantages of these three vendors are: Cypress: Cheap tools Huge range of devices (16V8 PAL to 20K gate FPGA from same tools) Re programmable on some ranges. Altera: Best suite of integrated tools. Cheap silicon Re programmable. Actel: Fastest operation Cheapest schematic entry via OrCad. Few more point spring to mind b4 I sign off. There is some research being done on compiling C directly to FPGA's, I have a link to Oxford Uni in my bookmarks but I havn't checked it out for months. Try: http://www.comlab.ox.ac.uk/oucl/hwcomp . OrCad are about to release a new product called Express & were looking for beta testers a few weeks back. Look for thread 'Consulting Opportunity' in this group or comp.lang.vhdl (Try www.dejanews.com ). Express gives you schematic & vhdl entry but you still need a vendors compiler. The best starting point I have found on the web for info on FPGAs is: Programmable Logic Jump Station (FPGA, CPLD) http://pw1.netcom.com/~optmagic/index.html Hope that lot helps TTFN JulianArticle: 5068
On Sun, 19 Jan 1997 09:30:47 +0000, Leon Heller <leon@lfheller.demon.co.uk> wrote: >> >>For those who aren't familiar with a meta-assembler, it's a software >>product which can be used to define instruction sets for custom >>sequencers. Years ago, I used it for 2900 family sequencers, He needs something that puts out arbitrary word-length microcode. A cross assembler won't do unless his microcode word just happens to be 32 bits, for instance. >You will find a free table-driven cross-assembler on the Simtel archive. >Look for TASM301.zip. You can create your own table for any instruction >set. There is a similar commercial product called Cross-32, which is >quite inexpensive ($500, perhaps). Cross-32 is suitable for 8, 16 and >32-bit processors, TASM might only be 8-bit. -- Rodney Myrvaagnes Associate Editor, Electronic Products rodneym@ibm.net 516-227-1434 Fax 516-227-1444 When possible, sailing J36 Gjo/aArticle: 5069
Viewlogic has a tool called JED2VHDL which generates a structural VHDL model from a JEDEC file. I do not know if it's still available in ViewOffice but used to be available with ViewPLD to allow for simulation of PAL and PLD devices. Encore Electronics (encore@shell.global2000.net) wrote: : Greetings... our customer generates JEDEC files for boards we build for : him, and we do the programming of Lattice in-system-programmable parts on : the board. I've got an older version of the code in .JED format, and a : newer version he just created from scratch after losing the original : source logic code (.LIF or .LDF files). The newer version naturally : doesn't work, and we don't know exactly what's in the older one. : Is there a way to reverse-compile the working .JED file to something : humanly-understandable that we can tweak, so he can re-compile it again? : I'm not too keen on the idea of taking the fuse map and writing out on a : sheet of paper to see what fuses are blown and open, and then figuring out : the logic from that. : The part in question is the ispLSI1032... fortunately we haven't had this : problem (yet) with the much larger ispLSI3256 he's also using on the : board. : Help? : Tom Moeller : Encore ElectronicsArticle: 5070
SUNNYVALE, CA-JANUARY 20, 1997-AMD today announced that Vantis Corporation will be the name of its programmable logic company. Vantis is being formed to better serve the specialized requirements of programmable logic customers. The company designs, develops and markets programmable logic products, taking advantage of AMD's advanced process technology and wafer manufacturing and packaging capabilities. The naming of Vantis Corporation is the latest milestone in a process that began last year when AMD announced the intent to form a programmable logic subsidiary. Since then, Vantis has established a worldwide dedicated sales force and a dedicated team of field applications engineers. "Vantis is a 28 year-old startup backed by a multi-billion dollar parent company with world-class manufacturing capabilities, leading-edge technology and a global scope," said Rich Forte, president of Vantis Corporation. "We are already one of the industry's leading suppliers of programmable logic. By forming Vantis we will be able to focus more on offering superior, cost-effective solutions that allow our customers to get to market before their competition. Vantis is committed to make its customers winners." Vantis also unveiled its new logo and worldwide web site at www.vantis.com. The logo can be viewed and downloaded from the web site. Vantis' new toll-free number is 1-888-826-8472 (VANTIS2). Vantis will continue to utilize AMD's worldwide Distributors. In addition, Vantis will receive administrative support functions from AMD. All customer support processes, such as order entry and technical support to name a few, will remain the same so that customers are ensured the same high-level of service that they receive today. About AMD AMD is a global supplier of integrated circuits for the personal and networked computer and communications markets. AMD produces processors, flash memories, programmable logic devices, and products for communications and networking applications. Founded in 1969 and based in Sunnyvale, CA, AMD had revenues of $2 billion in 1996. (NYSE: AMD). Vantis and the Vantis logo are trademarks and AMD, the AMD logo and MACH are registered trademarks of Advanced Micro Devices, Inc. GENERAL NOTICE: Product names used in this publication are for identification purposes only and may be trademarks of their respective companies.Article: 5071
[This article is a repost of <01bc0692$176777e0$4e7dbacd@p5-166>, cleaned up and with corrections regarding use of XC4000 H-muxes.] Ben Chaffin <98bcc@williams.edu> wrote in article <32E01194.778E@williams.edu>... > Hello, > I am a computer science major at Williams College. I am working > on a computational number theory problem for one of my professors. I have > implemented it in C and optimized it to death and am now considering > building a special purpose computer to speed it up. So, along with > everyone else, I'm trying to find the best balance between a limited > budget and speed. I have a design for discrete chips that should run > about 5 times as fast as a P-120. The cost is $300-$400, which isn't bad, > but the circuit involves nearly 200 chips, with about 3000 pins. That's a > lot of wiring. The algorithm involves: > > Two parallel 128-bit additions of internal-only data > Storage for the result (256 bits) > Two 48-bit conditional inversions using XOR gates > Two priority encodings across 48 bits > Some counters, flip-flops and simple control circuitry > Around 50 bits of output Hooray for such questions, which help to give us insight into the categorization of problems that are better done on FPGAs than on general purpose computers. In this case, we compare a fast implementation on a general purpose computer with a fast implementation on an FPGA (a XC4010E-2) which seems well suited for the problem. I hope the discussion is of some indirect use to Mr. Chaffin. Let's consider how well this problem could run on a superscalar 64-bit microprocessor. The description above would seem to require 2 adds, 2 adds-with-carry, 2 complements, 2 conditional moves, and 2 instances of a priority encoder. Each of the latter could be implemented as two iterations of binary search followed by a table lookup, e.g. as a shift, branch, shift, branch, load byte (4KB table), add constant. That's 30-40 clocks on a 2-integer issue machine, assuming all four branches are mispredicted with a 5-cycle misprediction penalty (as on a DEC Alpha 21164), and mostly L1 cache hits. With a 3 ns clock, that's about 100 ns per iteration. Now let's consider an FPGA implementation, specifically a simpleminded, non-pipelined, no-sharing, approach. We'll use a Xilinx XC4010E-2 (20x20 CLBs), $110 quantity one according to www.marshall.com. 128-bit adder: Slow approach: 64 CLBs configured as a 128-bit fast ripple carry adder (3+ columns of CLBs). Latency approx. 3+64*.6 =42 ns. Faster approach: 16+2*16+2*16+2*16 CLBs configured as a 32+32+32+32-bit carry select adder. Latency approx. 3+16*.6+5 gate delays+slop=25 ns. [By the way, if a CLB's F-LUTs and fast carry circuits are configured as a 2-bit fast ripple carry adder, can you still use its H-LUT to multiplex one of its F-LUT outputs with another signal, driving *that* onto a X or Y output?] Assuming H-muxes to multiplex the intermediate sums, two such adders require about 240 CLBs, more than half of the chip resources. You can store the sums in the flip-flops in the same CLBs. (Note, we did not discuss the adder input values. We may need to store an additional 256 or 512 bits of input state.) Two 48-bit inversions require a total of 48 CLBs if you can't fold them in elsewhere. Latency: 2-3 ns of logic plus some interconnect delay. 48-bit priority encoder: A 4-bit encoder is 1.5 CLBs. A 16-bit encoder is about 10 CLBs: 4 4-bit encoders (6 CLBs) + about 4 CLBs to combine them. A 48-bit encoder is about 36 CLBs: 3 16-bit encoders (30 CLBs) + about 6 CLBs to combine them. Therefore, two encoders require 72 CLBs. Latency is about 4 gate delays (8 ns) plus some interconnect delay => 15 ns. Altogether, a non-pipelined implementation could complete an iteration in 50 ns. If there is a way to overlap or pipeline the adder and the priority encoder, you might get this down to 30 ns. This is 2-3X faster than the general purpose CPU approach. However, it is also instructive to compare the approximate latencies of the individual operations. Adds: 6 ns (CPU); 25 ns (FPGA). Inversions: 6 ns (CPU); <10 ns (FPGA). Priority encoders: 90 ns (CPU); 15 ns (FPGA). The CPU loses badly because of the expense of the priority encoder implementation. So, one approach to the problem is to find a CPU with a priority encoder (e.g. the Microunity Mediaprocessor), or a not-yet-existent CPU with a little bit of programmable logic in its datapath, with which to implement a priority encoder... This circuit just fits in the 400 CLBs of an XC4010E. Two instances would fit in a XC4020E or XC4025E, more in the promised XC4000EX family parts. So, if this is not an inherently serial problem, doing several iterations in parallel would show a further performance advantage for the FPGA approach. Or, as Ray Andraka suggests in another reply, if your problem tolerates a lot of parallelism, a "massively" instanced bit serial approach will be faster yet. Jan GrayArticle: 5072
Julian Cox (CoxJA@augustsl.demon.co.uk) wrote: : Or use 3 processors / FPGA's to give you fault tolerance there too. : ;-) Of course, this does not always work. Just ask the folks at Stratus that built a fault-tolerant machine out of Intel i860 processors which used a "random" TLB replacement strategy. Hard to keep the processors synchronized at the bus cycle level when the chip behavior is not always deterministic. Worth considering when choosing a chip. [comp.sys.mentor removed from newsgroups] -- ---------------------------------------------------------------- Ewan D. Milne / Computervision Corporation (milne@petra.cv.com)Article: 5073
------------------------------------------------------------------------------- Advance Program 1997 ACM/SIGDA Fifth International Symposium on Field-Programmable Gate Arrays (FPGA'97) Sponsored by ACM SIGDA, with support from Altera, Xilinx, and Actel Monterey Beach Hotel, Monterey, California February 9-11, 1997 (Web page: http://www.ece.nwu.edu/~hauck/fpga97) ------------------------------------------------------------------------------ Welcome to the 1997 ACM/SIGDA International Symposium on Field-Programmable Gate Arrays (FPGA'97). This annual symposium is the premier forum for presentation of advances in all areas related to FPGA technology, and also provides a relaxed atmosphere for exchanging ideas and stimulating discussions for future research and development in this exciting new field. This year's symposium sees a strong increase of interest in FPGA technology, with over 20% increase in paper submissions. The technical program consists of 20 regular papers, 35 poster papers, an evening panel, and an invited session. The technical papers present the latest results on advances in FPGA architectures, new CAD algorithms and tools for FPGA designs, and novel applications of FPGAs. The Monday evening panel will debate whether reconfigurable computing is commercially viable. The invited session on Tuesday morning addresses the challenges for architecture development, CAD tools, and circuit design of one million-gate FPGAs and beyond. We hope that you find the symposium informative, stimulating, and enjoyable. Carl Ebeling, General Chair Jason Cong, Technical Program Chair ------------------------------------------------------------------------------ SYMPOSIUM PROGRAM Sunday February 9, 1997 6:00pm Registration 7:00pm Welcoming Reception, Monterey Beach Hotel, Monterey Monday February 10, 1997 7:30am Continental Breakfast/Registration 8:20am Welcome and Opening Remarks Session 1: FPGA Architectures Session Chair: Rob Rutenbar, Carnegie Mellon Univ. Time: 8:30 - 9:30am 1.1 "Architecture Issues and Solutions for a High-Capacity FPGA", S. Trimberger, K. Duong, B. Conn, Xilinx, Inc. 1.2 "Memory-to-Memory Connection Structures in FPGAs with Embedded Memory Arrays", Steven J.E. Wilton, J. Rose, Z.G. Vranesic, University of Toronto 1.3 "Laser Correcting Defects to Create Transparent Routing for Large Area FPGAs", G.H. Chapman, B. Bufort, Simon Fraser University Poster Session 1: Analysis and Design of New FPGA Architectures Session Chair: Tim Southgate, Altera, Inc. Time: 9:30 - 10:30am (including coffee break) Session 2: FPGA Partitioning and Synthesis Session Chair: Richard Rudell, Synopsys, Inc. Time: 10:30 - 11:30am 2.1 "I/O and Performance Tradeoffs with the FunctionBus during Multi-FPGA Partitioning", F. Vahid, University of California, Riverside 2.2 "Partially-Dependent Functional Decomposition with Applications in FPGA Synthesis and Mapping", J. Cong, Y. Hwang, Univ. of California, Los Angeles 2.3 "General Modeling and Technology-Mapping Technique for LUT-based FPGAs", A. Chowdhary, J.P. Hayes, University of Michigan Poster Session 2: Logic Optimization for FPGAs Session Chair: Martine Schlag, Univ. of California, Santa Cruz Time: 11:30 - 12noon Lunch: noon - 1:30pm Session 3: Rapid Prototyping and Emulation Session Chair: Carl Ebeling, Univ. of Washington Time: 1:30 - 2:30pm 3.1 "The Transmogrifier-2: A 1 Million Gate Rapid Prototyping System", D.M. Lewis, D.R. Galloway, M. V. Ierssel, J. Rose, P. Chow, University of Toronto 3.2 "Signal Processing at 250 MHz using High-Performance Pipelined FPGA's", Brian Von Herzen, Rapid Prototypes, Inc. 3.3 "Module Generation of Complex Macros for Logic-Emulation Applications", Wen-Jong Fang, Allen C.H. Wu, Duan-Ping Chen, Tsinghua University Poster Session 3: Novel FPGA Applications Session Chair: Brad Hutchings, Brigham Young Univ. Time: 2:30 - 3:30pm (including coffee break) Session 4: Reconfigurable Computing Session Chair: Jonathan Rose, Univ. of Toronto Time: 3:30 - 4:30pm 4.1 "Wormhole Run-time Reconfiguration", R. Bittner, P. Athanas, Virginia Polytechnic Institute 4.2 "Improving Computational Efficiency Through Run-Time Constant Propagation", M.J. Wirthlin, B.L. Hutchings, Brigham Young University 4.3 "YARDS: FPGA/MPU Hybrid Architecture for Telecommunication Data Processing", A. Tsutsui, T. Miyazaki, NTT Optical Network System Lab. Poster Session 4: Reconfigurable Systems Session Chair: Scott Hauck, Northwestern Univ. Time: 4:30 - 5:30pm Dinner: 6:00 - 7:30pm Evening Panel: Is reconfigurable computing commercially viable? Moderator: Herman Schmit, Carnegie Mellon Univ. Time: 7:30 - 9:00pm Panelists: Steve Casselman: President, Virtual Computer Corp. Daryl Eigen: President, Metalithic Systems, Inc. Robert Parker: Deputy Director, ITO, DARPA Peter Athanas: Assistant Professor, Virginia Polytechnic Institute Robert Colwell: Pentium Pro Architecture Manager, Intel Corp. In this panel session, we will try to address the questions of whether there will be a mass-market for FPGA-based computing solutions. Are there large sets of applications whose performance requirements far exceed that offered by microprocessors but which are only occasionally executed? Where are these applications? Does the ability to reconfigure during execution change the cost and performance benefits of reconfigurable hardware significantly? What are the key challenges to making reconfigurable computing a reality, and what can PLD manufacturers, system houses, government, and academia do to overcome these obstacles? Session 5: FPGA Floorplanning and Routing Session Chair: Dwight Hill, Synopsys, Inc. Time: 8:30 - 9:30am 5.1 "Synthesis and Floorplanning for Large Hierarchical FPGAs", H. Krupnova, C. Rabedaoro, G. Saucier, Institut National Polytechnique de Grenoble/CSI 5.2 "Performance Driven Floorplanning for FPGA Based Designs", J. Shi, Dinesh Bhatia, University of Cincinnati 5.3 "FPGA Routing and Routability Estimation Via Boolean Satisfiability", R.G. Wood, R.A. Rutenbar, Carnegie Mellon University Poster Session 5: High level Synthesis and Module Generation for FPGAs Session Chair: Martin Wong, Univ. of Texas at Austin Time: 9:30 - 10:30am (including coffee break) Session 6 (Invited): Challenges for 1 Million-Gate FPGAs and Beyond Session Chair: Jason Cong, Univ. of California, Los Angeles Time: 10:30am - noon Process technology advances tell us that the one million gate FPGA will soon be here, and larger devices shortly after that. Current architectures will not extend easily to this scale because of process characteristics and because new opportunities are presented by the increase in available transistors. In addition, such large FPGAs will also present significant challenges to the computer-aided design tools and methods. Two invited papers address these issues. 6.1 "Architectural and Physical Design Challenges for One Million Gate FPGAs and Beyond", Jonathan Rose, University of Toronto, Dwight Hill, Synopsys, Inc. 6.2. "Challenges in CAD for the One Million-Plus Gate FPGA", Kurt Keutzer, Synopsys, Inc. Lunch: noon - 1:30pm Session 7: Studies of New FPGA Architectures Session Chair: Steve Trimberger, Xilinx, Inc. Time: 1:30 - 2:30pm 7.1 "A CMOS Continuous-time Field Programmable Analog Array", C.A. Looby, C. Lyden, National Microelectronics Research Center 7.2 "Combinational Logic on Dynamically Reconfigurable FPGAs", D. Chang, M. Marek-Sadowska, Univ. of California, Santa Barbara 7.3 "Generation of Synthetic Sequential Benchmark Circuits", M. Hutton, J. Rose, D. Corneil, University of Toronto Poster Session 6: FPGA Testing Session Chair: Sinan Kaptanoglu, Actel, Inc. Time: 2:30 - 3:30pm (including coffee break) Session 8: Novel Design and Applications Session Chair: Pak Chan, Univ. of California, Santa Cruz Time: 3:30 - 4:10pm 8.1 "Synchronous Up/Down Binary Counter for LUT FPGAs with Counting Frequency Independent of Counter Size", A.F. Tenca, M. D. Ercegovac, Univ. of California, Los Angeles 8.2 "A FPGA-based Implementation of a Fault Tolerant Neural Architecture for Photon Identification" M. Alderight, E.L. Gummati, V. Piuri, G.R. Sechi, Consiglio Nazionale delle Ricerche, Universita degli Studi di Milano, Politecnico di Milano 4:30pm Symposium Ends. ------------------------------------------------------------------------------- Organizing Committee: General Chair: Carl Ebeling, University of Washington Program Chair: Jason Cong, UCLA Publicity Chair: Scott Hauck, Northwestern University Finance Chair: Jonathan Rose, University of Toronto Local Chair: Pak Chan, UC Santa Cruz Program Committee: Michael Butts, Quickturn Pak Chan, UCSC Jason Cong, UCLA Carl Ebeling, U. Washington Masahiro Fujita, Fujitsu Labs Scott Hauck, Northwestern Univ. Dwight Hill, Synopsys Brad Hutchings, BYU Sinan Kaptanoglu, Actel David Lewis, U. Toronto Jonathan Rose, U. Toronto Richard Rudell, Synopsys Rob Rutenbar, CMU Gabriele Saucier, Imag Martine Schlag, UCSC Tim Southgate, Altera Steve Trimberger, Xilinx Martin Wong, UT Austin Nam-Sung Woo, Lucent Technologies ------------------------------------------------------------------------------- Hotel Information FPGA'97 will be held at the Monterey Beach Hotel, 2600 Sand Dunes Dr., Monterey, CA 93940 USA. The phone number for room reservations is 1-800-242-8627 (from USA or Canada) or +1-408-394-3321 (fax: +1-408-393-1912). Reservations must be made before January 10, 1997. Identify yourself with the group: ACM/FPGA'97 to receive the special rates of US$75 single/double for Gardenside and US$105 single/double for Oceanside (additional person in the room is $10), plus applicable state and local taxes. Reservations may be canceled or modified up to 72 hours prior to arrival without a penalty. If the cancellation is made within 72 hours of arrival, or you fail to show up, first nights room and tax will be charged. If a modification is made within 72 hours of arrival (i.e., postpones arrival or departs earlier than reserved) the actual nights of your stay will be charged at the quoted rack rate for the room occupied. Check-in time is 4:00 pm, and check-out time is 12:00 noon. Directions by car: From San Jose (1.5 hours) or San Francisco Airport (2.5 hours) take Hwy 101 South to Hwy 156 West to Hwy 1 South. From Hwy 1 South, take Seaside/Del Rey Oaks exit. The hotel is at this exit on the ocean side. You can also fly directly to Monterey Airport, which is served by United, American and other airlines with at least 8 flights per day. Monterey Area The Monterey Peninsula is famous for its many attractions and recreational activities, such as John Steinbeck's famous Cannery Row and the Monterey Bay Aquarium. Also, play one of 19 championship golf courses. Charter fishing is available right at Firsherman's Wharf. Monterey is renowned worldwide for its spectacular coastline, including Big Sur and the Seventeen Mile Drive. Recreational activities, shopping opportunities and restaurants abound. ------------------------------------------------------------------------------- Registration Information: The Symposium registration fee includes a copy of the symposium proceedings, a reception on Sunday evening, February 9, coffee breaks, lunch on both days, and dinner on Monday evening, February 10. First Name:_____________________Last Name:_________________________________ Title/Job Function:________________________________________________________ Company/Institution:_______________________________________________________ Address:___________________________________________________________________ City:___________________________State:_____________________________________ Postal Code:____________________Country:___________________________________ E-mail:_________________________ACM Member #:______________________________ Phone:__________________________Fax:_______________________________________ Circle Fee Before January 22, 1997 After January 22, 1997 ACM/SIGDA Member US$300 US$370 *Non-Member US$400 US$470 Student US$ 90 (does not include reception or banquet, available for US$15 and US$55 respectively) Guest Reception Tickets: # Tickets _____x US$15 = ______ Guest Banquet Tickets: # Tickets _____x US$55 = _______ Total Fees: _________________ (Make checks payable to ACM/FPGA'97) Payment included (circle one): American Express MasterCard Visa Check Credit Card # :_______________________ Expiration Date:________ Signature:______________________________________________________ Send Registration, including payment in full, to: FPGA'97, Meeting Hall, Inc., 571 Dunbar Hill Rd., Hamden, CT 06514 USA Phone/fax: +1 203 287 9555 For registration information contact Debbie Hall via e-mail at halldeb@aol.com. Cancellations must be in writing and received by Meeting Hall, Inc. before January 22, 1997. +-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-+ | Scott A. Hauck, Assistant Professor | | Dept. of ECE Voice: (847) 467-1849 | | Northwestern University FAX: (847) 467-4144 | | 2145 Sheridan Road Email: hauck@ece.nwu.edu | | Evanston, IL 60208 WWW: http://www.ece.nwu.edu/~hauck | +-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-+Article: 5074
>>>>> Peter Alfke <peter@xilinx.com> writes: > Xilinx XC3000 and XC31000 have an on-chip single-stage inverter, > intended for being wrapped around a 1 MHz to 40 MHz crystal, and > there are thousands of designs using this feature. More recent > Xilinx architectures don't have this on-chip circuitry for the > following reasons: [marketing handwaving deleted] Unfortunately, one important application for the XTL stage was as a clock recovery VCXOs, which isn't available off the shelf (or for "around one dollar" either). I agree that oscillator design may be beyond the capability of some digital designers, but there are certainly plenty of competent datacomm engineers out here. > There have been no complaints from our customer base when we > deleted the crystal oscillator option from the newer > architectures... Ummm... -- Roger Williams finger me for my PGP public key Coelacanth Engineering consulting & turnkey product development Middleborough, MA wireless * DSP-based instrumentation * ATE tel +1 508 947-8049 * fax +1 508 947-9118 * http://www.coelacanth.com/
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