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
> sbaker@best.com writes: I apologize for not identifying myself as program chairman for the PLDCon'96 in the referenced message. Stan Baker > > > > > > > >>>>Article: 3151
Bob Elkind (eteam@aracnet.com) wrote: : Brad Kelley wrote: : > : > I am currently doing a design in an Actel FPGA (Act1 or Act2) that I would : > like to convert to an ASIC to reduce cost, current consumption, package size, : > all the good reasons. It's been a while since we last surveyed vendors, so : > we'll be out doing a current one. Anyone done a conversion lately? Problems, : > suggestions, lessons learned, vendor problems/successes, any tips would be : > appreciated. : > : > Current project is low-volume (fixed order for 500 units, 1 FPGA/ASIC per), so : > zero NRE or low NRE is required. Have some one-year old info that at least : > one vendor would convert for zero NRE with 500-piece order, so I think this : > might be possible even with small volume. : If you stand back, you may conclude that you will be paying for the NRE : costs; either through up-front charges, or through higher per-unit production : cost, or a combination of the two. : Masks cost money, and the ASIC vendor certainly hasn't forgotten that. : The real (rather than accounting magic) savings are probably found in : the following areas: : 1. Find a vendor who has a genuinely lower cost of doing business than : the others. Sometimes this implies finding a vendor with its own captive : fab facility, their own mask shop, etc. (snip) : 2. Find a vendor that requires as little time/effort on your part as possible. : This saves you NRE costs that you would otherwise bear in terms of development : time and effort. There is a third way that a low-volume asic route can save money - by putting two or more designs on the same mask set and sharing the cost between the customers. This is the method used by some of the university prototyping services for instance. It's not as simple as "shared between 2 customers = half the cost", but it is possible to reduce the overall cost. You have to juggle factors like: 1) The fab probably has a minimum and maximum mask size that it can use. So, if one customers requirements don't reach the minimum size there is a benefit from combining with another customer. 2) Mask cost is area dependent - the bigger the mask the more it costs. If you put two designs on the same mask set then the masks will be bigger and therefore more expensive. 2a) There is a constant element in the mask size - test transistors and alignment structures that have to be present on all designs. If 2 designs are put on the same masks then they can share these common elements, so that the size of the composite mask is less that the sum of the sizes of the separate ones. 3) The fab cycle time is probably optimised for processing a minimum number of wafers to a particular design at a time. Suppose this is 20 wafers. If a customer comes along and wants 500 units of a design that you estimate will yield 50 die per wafer then there's a problem - 50 * 20 = 1000. If there are two such customers though then they can share the same production run, each getting 25 die per wafer, but still using the fab at its most efficient loading. Anthony Stansfield ais@pact.srf.ac.ukArticle: 3152
During a recent test of one of our customers systems, we had a failure at upper temperatures. After some debug with the Probe, the part seems to have slow rise times, producing delays on the order of 4-5x what a good part exhibits. Has anyone else experienced any symptons like these? I have verified that the same design works properly on a different card and have verified Vcc and ground continuity to the part. Device is an ACT1020 2.0u intended for a space application.Article: 3153
Hello, world! This posting seeks feedback on the library of parametrized modules (LPM), AHDL design style and logic synthesis options within MAX+plusII version 6.01. It addresses an audience with some deeper insight in AHDL / MAX+plusII. Within a larger design project, I had to design a 16-bit binary up-/down counter with synchronous and asynchronous reset for an Altera MAX-7000 CPLD. Looking at the internal structure of the MAX7000 CPLDs, it was easy to do an optimal hand mapping of the desired function. The new value to be stored for a single counter bit in a LCs flipflop can completely be determined with the LCs logic resources according to: c[i] = c[i] XOR ( sync_reset & c[i] # !sync_reset & incr & c[i-1] & c[i-2] & ... & c[0] # !sync_reset & !incr & decr & !c[i-1] & ... & !c[0] ); or when using an explicit enable and a signal selecting the count direction: c[i] = c[i] XOR ( sync_reset & c[i] # !sync_reset & enable & dir_up & c[i-1] & ... & c[0] # !sync_reset & enable & !dir_up & !c[i-1] & ... & !c[0] ); The question arouse: "How do I tell MAX+plusII to generate these logic functions?" Since we just got our hands on MAX+plusII version 6.01, I decided to give a try to the new LPM-modules and instantiated an 8-bit LPM_COUNTER with the following source code: FUNCTION LPM_COUNTER (data[LPM_WIDTH-1..0], clock, clk_en, cnt_en, updown, sclr, sset, sconst, sload, aclr, aset, aconst, aload) WITH (LPM_WIDTH, LPM_REPRESENTATION, LPM_MODULUS, LPM_AVALUE, LPM_SVALUE) RETURNS (q[LPM_WIDTH-1..0]); SUBDESIGN cnt8_lpm ( /reset, clk, sync_reset, enable, incr: INPUT; value[7..0]: OUTPUT; ) VARIABLE counter: LPM_COUNTER WITH (LPM_WIDTH = 8); BEGIN counter.(clock, cnt_en, updown, sclr, aclr) = (clk, enable, dir_up, sync_reset, /reset); value[] = counter.q[]; END; The results were disappointing: 17 LCs in a EPM7032 ! So I tried some different synthesis styles: FAST, NORMAL, WYSIWYG. The results changed, but were still far from the optimum value: 8 LCs. What are the LPMs good for, when such a simple design as a counter is synthesized with such an overhead? "Ok, it's not that simple with the LPMs. Then let's turn back to good old AHDL high level constructs." Next, I tried the following straightforward source code: SUBDESIGN cnt8 ( /reset, clk, sync_reset, incr, decr: INPUT; value[7..0]: OUTPUT; ) VARIABLE c[7..0]: DFF; BEGIN c[].(clrn, clk) = (/reset, clk); IF sync_reset THEN c[] = 0; ELSIF incr THEN c[] = c[] +1; ELSIF decr THEN c[] = c[] -1; ELSE c[] = c[]; END IF; value[] = c[]; END; The results, as reported in the .rpt file, where quite good: 8 LCs as expected, but - what the heck? - 4 sharable expanders? What are those used for. So I took a closer look to the report file and found the following equation for bit 5: value5 = TFFE( _EQ006, GLOBAL( clk), GLOBAL( /reset), VCC, VCC); _EQ006 = decr & !incr & !sync_reset & !value0 & !value1 & !value2 & !value3 & !value4 & !value5 # incr & !sync_reset & value0 & value1 & value2 & value3 & value4 & !value5 # decr & !incr & !value0 & !value1 & !value2 & !value3 & !value4 & value5 # incr & value0 & value1 & value2 & value3 & value4 & value5 # sync_reset & value5; The last term I had expected. But what has happened to other terms? The second, forth and fifth term can obviously combined giving # !sync_reset & incr & value0 & ... & value4 and the first, third and fifth together yield # !sync_reset & !incr & decr & !value0 & ... !value4 . For the higher order bits (bits 6 and 7), DFFEs and sharable expanders (for value0 & ... & value5 and !value0 & ... & !value5) were used. Then I guessed: "Ok, maybe I have to use some other synthesis options to make the compiler to produce the optimum result." So I tried FAST, NORMAL, WYSIWYG, switching of this and that advanced option, turning XOR synthesis on and off, but I didn't arrive at the optimal design. I finally typed in the logic equation for each bit and compiled with WYSIWYG - and got the optimum design at last. This hand-made counter is reported to run at 62 MHz and uses 8 LCs (16 LCs for the 16-bit version). Finally I tried the old-style macrofunction 8COUNT and this also result in an 8 LCs mapping with no expanders. That's the end of this (lengthy) story (thanks for following to this point). My points are: o Has anybody allready made some experiences with the new LPMs? Is my experience typical or am I doing anything wrong with using the module? o Can someone explain what is happening to the AHDL description? What synthesis options (if any) are responsible for the inefficiencies in the mapping? Why does the compiler fail to reduce the logic in _EQ006 above? o What "rule" makes MAX+plusII to change from TFFEs to DFFEs for the two most significant counter bits? Any ideas, comments and design style suggestions appreciated. Andre'. -- --------------------------------------------------------------------------- Andre' Klindworth Universitaet Hamburg, FB Informatik klindwor@informatik.uni-hamburg.de Vogt-Koelln-Str.30, D-22527 Hamburg http://tech-www.informatik.uni-hamburg.de/Personal/klindwor/Klindworth.htmlArticle: 3154
Hi Vijay, the VHDL predefined Attributes 'VAL and 'POS should work: variable t : time; variable i : integer; t := TIME'VAL(i); i := TIME'POS(t); WolfgangArticle: 3155
Vijay A Nebhrajani (vijayn@cdac.ernet.in) wrote: : Hi, everybody! : Anyone know of a function on Synopsys to convert time to integer and vice versa? : How could such a function be written, if it were not available readymade? need help : VERY urgently. : Thanks in advance. Email direct to vijayn@cdac.ernet.in : -- vijay. How about: signal_of_type_time <= signal_of_type_integer * 1 ns; signal_of_type_integer <= signal_of_type_time / 1 ns; Note that the space between `1' and `ns' is important!! AndrewArticle: 3156
My news reader program does not include the posting header info. I am trying to fix it. Please let me know if there is a better place to do this.. Robb Cole e-mail rmcole@lfs.loral.comArticle: 3157
I promise.. this is the last one. Robb Cole MS 0302 Loral Federal Systems-Owego 1801 State Route 17C Owego, NY 13827-3994 Voice 607-751-3708 Fax 607-751-6732 e-mail rmcole@lfs.loral.comArticle: 3158
Hi, What is the lowest-priced FPGA? I've used XC4003 for some project before but they are very expensive. Now I'm working on a new project which is very cost-sensitive. We need to select a programable logic device with very low cost and relatively high gate and I/O counts. What are costs for XC2000 family parts? Thanks in advance for any information. -GF-Article: 3159
Yes, use Exemplar Logic Galileo. They syntheisze the complex I/Os contained in the ACT3 family. They even do design rule checks to be sure the designer doesn't overuse the clock resources. MTC and Logic Innovations are the Exemplar representatives in Muenchen. Also contact Exemplar at info@exemplar.com In <316E36A7.2781E494@scn.de> b1052 <b1052@scn.de> writes: > >Hi, >when synthesizing a VHDL design with Synopsys Design Compiler into an >act3 lib and transfering it to the Designer software from Actel I get >errors because non-clock pins use a clock net. The reason is that Design >Compiler puts "complex" cells on clock nets; the also used inverters are >not flagged with errors. >Has somebody an idea how I can avoid this? > >TIA > Guido >-- >Guido Kinast Siemens AG, AUT E721 Fuerth, Germany >b1052@scn.de http://www.scn.de/~b1052 +49 911 750 2720Article: 3160
gf0570 wrote: > > Hi, > What is the lowest-priced FPGA? I've used XC4003 for some project > before but they are very expensive. Now I'm working on a new project > which is very cost-sensitive. We need to select a programable logic > device with very low cost and relatively high gate and I/O counts. > > What are costs for XC2000 family parts? > > Thanks in advance for any information. > > -GF- The XC2000 series are a bit basic - especially if you're using 4000 series now and need similar density devices. A better bet would be the XC5000 series. Your distributor should be able to get data - or check out http://www.xilinx.com -- Regards AndyG ------------------------------------------------------------------------ ** Terry Pratchett's 'Wyrd Sisters' on stage in Leeds - April 25-27 ** check out http://ourworld.compuserve.com/homepages/Whitkirk for details, or e-mail 73064.1273@compuserve.com ------------------------------------------------------------------------Article: 3161
In article <4ktl7j$akc@cc.iu.net>, clifford@bb.iu.net says... > >During a recent test of one of our customers systems, we had a failure at >upper temperatures. After some debug with the Probe, the part seems to have >slow rise times, producing delays on the order of 4-5x what a good part >exhibits. Has anyone else experienced any symptons like these? I have >verified that the same design works properly on a different card and have >verified Vcc and ground continuity to the part. Device is an ACT1020 2.0u >intended for a space application. Interesting, We used a number of ACT1020A on our previous satellite, they all worked fine. I assume you are aware that the 2.0 A version is being discontinued. If you have the money available you can get a (real?) Rad-Hard version from Loral. I belief they make the A1225 and A1280 using Silicon on Insulator. RAD-PACK is an another company who makes rad-hard Actel (and others) FPGA's. They take a standard commercial die and re-packing it using there patented RAD-PACK package. There prices are quite reasonable, you also don't have to order 10000 as a minimum :-) Regards, Hans Surrey Satellite Technology Ltd.Article: 3162
If you are interested in programmable logic hardware or software and are in Germany on May 15th, check out the PLD Developers Forum in Munich. Hosted by an electronics magazine (Design&Elektronik)this will be one of the biggest (if not THE biggest) PLD event in Germany in 96. In contrast to last year presentations will not be by manufactures, but by independent presenters. Should take out the hype. For more information: http://www.magnamedia.de/d&e/forum/logik.html Uwe Kremmin AMDArticle: 3163
Hi, Our research group is looking into purchasing circuits boards that will take one to a few Xilinx 4020 FPGA chips. These Boards are to be used for circuit verifications by downloading design using Xilinx's Xact cable. If the board contains multiple chips, they should all be able to be programmed using a single xact cable link. If you have such kind of boards avaliable for sale or know anybody who supplies them, please mail me at S.C.Lim@lboro.uc.uk. Regards Ryan Lim Seow Chuan ============================================================ Seow-Chuan Lim (Ryan) 1010010001010011101 Research Student 1 1 Loughborough University 0 DIGITAL SYSTEMS 0 Ashby Road, Loughborough. LE11 3TU. U.K. 1 LABORATORY 0 Tel: 0-1509-228104 0 1 email S.C.Lim@Lboro.ac.uk 1001001001011100111 ============================================================Article: 3164
If you are interested in programmable logic hardware or software and are in Germany on May 15th, check out the PLD Developers Forum in Munich. Hosted by an electronics magazine (Design&Elektronik)this will be one of the biggest (if not THE biggest) PLD event in Germany in 96. In contrast to last year presentations will not be by manufactures, but by independent presenters. Should take out the hype. For more information: http://www.magnamedia.de/d&e/forum/logik.html Uwe Kremmin AMDArticle: 3165
For a given design, what can be done to reduce the power consumption of a Xilinx 3020 device? I think in terms of special routing, unused nets etc. Anyone got any ideas on how much I could expect to reduce the power consumption using "after-design" techniques only? 10%, or? - RolfArticle: 3166
I'm about to do my first expieriencies working with PSPICE (PLSYN) as the programming tool on a MACH445-device. 1. Is this the right newsgroup for me (any othe sugestions) 2. Is anyone here that will know what to watch out for. I have limmited expierience with PLD's. All the best From Kaj Norman NielsenArticle: 3167
In article <4kungr$k6n@news.cc.utah.edu>, gf0570 <fang@signus.utah.edu> wrote: > Hi, > What is the lowest-priced FPGA? I've used XC4003 for some project > before You were not specific regarding pin-count and speed, and I will not quote prices, that's done by our local salesforce,.Nevertheless, here are the Xilinx devices that you might want to look at, in descending price order, starting from the XC40003. Remember, package, speed, and quantity affect the price significantly. XC4003, XC5204, XC3042A, XC3030A, XC5202, XC3020A, XC2018, XC2064. The XC2000 series is a bit old, the XC3000A and especially the XC5200 families offer better value, IMHO. Peter Alfke, Xilinx ApplicationsArticle: 3168
In article <316E36A7.2781E494@scn.de>, b1052 <b1052@scn.de> wrote: >Hi, >when synthesizing a VHDL design with Synopsys Design Compiler into an >act3 lib and transfering it to the Designer software from Actel I get >errors because non-clock pins use a clock net. The reason is that Design >Compiler puts "complex" cells on clock nets; the also used inverters are >not flagged with errors. >Has somebody an idea how I can avoid this? > >TIA > Guido The easiest way to avoid this problem is to not use Actel, and to use the Xilinx XC8100 instead. :-) :-) :-) The XC8100 architecture allows the high drive nets to be used for any pin connection (clock, reset, combinatorial, etc...) unlike the Actel families that require that these nets be connected to clock and reset inputs only. Here are my suggestions to get around this problem in the Actel families: 1) If the clock net must be combined with combinatorial logic Split the clock net into two signals, 1 that drives the clock pins and another that goes to the combinatorial logic. Use an internal clock buffer as the source on the clock drive. You will need to instantiate this in your VHDL 2) If the clock net only goes to clocks and resets but you still get this problem Check your compile script and remember to apply a dont_use attribute to the clock buffers after inserting pads on your clock pins and before inserting on your regular IO. This will prevent Synopsys from adding the nifty high drive input buffers were they're not wanted. Example: set_port_is_pad {clk reset} set_pad_type -exact CLKBUF {clk reset} remove_attribute act3/CLKBUF dont_use insert_pads set_port_is_pad . remove_attribute {clk reset} port_is_pad set_dont_use act3/CLKBUF set_dont_touch act3/CLKBUF insert_pads Regards, Ed -- +----------------------------------------------------------------------+ |Edward McGettigan | Xilinx Inc. "The Programmable Logic Company" sm | |mcgett@xilinx.com | XC8100 FPGA Applications Engineer (408)879-4772 | +----------------------------------------------------------------------+Article: 3169
Dear friends, people always talk about "prototyping ASIC" with FPGA(s). But at least in my case, due to the uncertainty of market, ASIC solution is not possible at the moment. I have to find FPGA device whose gate count is about 20000 (at real utilization) and whose price is relative low to make the final product competitive. Because of the internal architecture of my design, I could not partition it into more than two FPGAs efficiently. Of course, I believe that the higher integration, the lower cost. It is lucky that I design it in VHDL so that I have the freedom in choosing FPGAs. Don't tell me Altera's Flex10K-50. Its high price and its sole huge PGA package (more than 390 pins) stops me. I have tried Actel's DX32140 but its gate count is not enough either. They told me that Actel's backend tool does not support DX32200 yet. I have also tried the Xilinx's XC5200. But the result remained the same. I do not care for the operating speed of the final chip, as long as the chip can hold my design. It is really tough. After the search, I got a strong feeling that the gap (performance and size) between FPGA and ASIC is becoming larger and larger. FPGAs seem can only find market position for "Glue logic integration" and "peripheral function for supporting ASIC". What else FPGA vendors can I tried? Regards, Felix K.C. CHEN -- --------------------------------- Felix, Kuan-chih CHEN (³¯ «a §Ó) Associate Project Manager System Product Division D-Link Co., Hsin-chu, Taiwan Email: flxchen@diig.dlink.com.tw Machines and tools are only as good as the people who use it. ---------------------------------Article: 3170
Felix K.C. CHEN wrote: > I have to find FPGA device whose gate count > is about 20000 (at real utilization) and whose price is relative > low to make the final product competitive. > > Don't tell me Altera's Flex10K-50. Its high price and its sole huge > PGA package (more than 390 pins) stops me. I have tried Actel's > DX32140 but its gate count is not enough either. They told me > that Actel's backend tool does not support DX32200 yet. I have > also tried the Xilinx's XC5200. But the result remained the same. > I do not care for the operating speed of the final chip, as long > as the chip can hold my design. > > What else FPGA vendors can I tried? Felix, Xilinx 4028EX (28K gates, enhanced routing resources, promises 100% gate utilisation) ATT/Lucent 2C26, 2C40 (26K, 40K gates respectively.) The Xilinx device is brand-spanking new, and I'm sure Xilinx will push the gate count higher on successive devices. The 2Cxx parts are in production, and they are generally regarded as a good choice (reliable toolset, utilisation factor, etc.). Their end-user cost is quite reasonable, also. I don't know much about VHDL support for these puppies, though. I've heard 2nd hand that the Quicklogic family has excellent VHDL support, but I think you'll bump against gate count limits in that family. Can someone clarify this notion? Good Luck, Bob Elkind ************************************************************************** Bob Elkind email:eteam@aracnet.com CIS:72022,21 7118 SW Lee Road part-time fax number:503.357.9001 Gaston, OR 97119 cell:503.709.1985 home:503.359.4903 ******** Video processing, R&D, ASIC, FPGA design consulting *************Article: 3171
Rolf V. wrote: > For a given design, what can be done to reduce the power consumption of > a Xilinx 3020 device? I think in terms of special routing, unused nets > etc. > > Anyone got any ideas on how much I could expect to reduce the power > consumption using "after-design" techniques only? 10%, or?-- Suggestions: 1. Reduce the clock rate 2. Use the 3100a family (5V), they are smaller geometry and lower power. 3. Use the 3100L family (3.3V). 4. If die temperature is the root concern, rather than current consumption, then try a different package with better thermal characteristics (e.g. the metal quad flatpak). Or try a heat sink on a non-plastic package. There are some design hacking techniques that will theoretically reduce power consumption, but they are very time-intensive, and you won't get dramatic results (short of complete design re-structuring). Bob Elkind ************************************************************************** Bob Elkind email:eteam@aracnet.com CIS:72022,21 7118 SW Lee Road part-time fax number:503.357.9001 Gaston, OR 97119 cell:503.709.1985 home:503.359.4903 ******** Video processing, R&D, ASIC, FPGA design consulting *************Article: 3172
-- Harald Bratko alias hbratko@sbox.tu-graz.ac.atArticle: 3173
In article <4l45ct$7q6@news.cis.nctu.edu.tw>, flxchen@diig.dlink.com.tw says... > snip > >What else FPGA vendors can I tried? > >Regards, > >Felix K.C. CHEN You might want to have a look at AT&T's 40000 gates ATT2C40. http://www.att.com/press/0295/950227.mea.html Hans.Article: 3174
If you are interested in programmable logic hardware or software and are in Germany on May 15th, check out the PLD Developers Forum in Munich. Hosted by an electronics magazine (Design&Elektronik)this will be one of the biggest (if not THE biggest) PLD event in Germany in 96. In contrast to last year presentations will not be by manufactures, but by independent presenters. Should take out the hype. For more information: http://www.magnamedia.de/d&e/forum/logik.html Uwe Kremmin AMD
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