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
Rodney Myrvaagnes wrote: > > Are there any FPGAs or CPLDs with Schmitt-trigger inputs that could be driven by a slowly varying voltage (i.e. 10s of Hz)? > Xilinx parts have 100 - 200 mV hysteresis on their inputs. For an app. note describing use of positive feedback for user controlled hysteresis (applicable to most PLDs) check out: http://www.xilinx.com/xcell/xl19/xl19-34.pdf (You will need an Acrobat .PDF viewer) regards, tomArticle: 6251
It's easiest to "roll your own": Go into and then out of the programmable device without a signal inversion. Connect a high-value resistor R from the chip output to the chip input. Drive that chip input through a low value resistor r. If the output is complementary ( CMOS ), then the hysteresis is supply voltage times ratio of r over R. Example: R=10 kilohm, r = 1 kilohm, hysteresis 500 mV. This, of course, assumes a drive impedance considerably lower than r. The only bad side is that it occupies two pins, and the trip points are dictated by the device input threshold. There was a more elaborate design, published in the 1986 (!) Xilinx data book, but I consider it unnecessarily complicated. Peter Alfke, Xilinx ApplicationsArticle: 6252
TestArticle: 6254
ANNOUNCEMENT: The prices for the APS-X84 FPGA kits were scheduled to be increased, but APS is pleased to announce that we are able to extend these prices for the next 60 days. This means you get a complete XILINX FPGA development system (non VHDL) kit for as low as $650.00. A VHDL kit can be purchased for as low as $1200.00. These are serious tools offered at very low prices. The APS-X84 FPGA kits for XILINX FPGAs include the XILINX Foundation series package and contains Schematic Capture (which can take in Viewlogic macros), XACT router tools, XCHECKER cables, XABEL compiler and post route simulator. The VHDL packages come with VHDL editor, VHDL WIZARD, and VHDL multimedia tutorial. All kits also includes the multipurpose APS-X84 PC ISA board which comes with an 84 pin PLCC FPGA chip installed and VHDL and C examples. All kits come with one year XILINX maintenance on the XILINX tools. For more information see the following URLs: http://www.erols.com/aaps http://www.erols.com/aaps/prices.htmlArticle: 6255
Arrigo Benedetti wrote: > > I have some problems implementing a design with a MUX having a three > state output with Synopsys FPGA Compiler 3.5a targeted to a Xilinx > 4010E part. The MUX is implemented by the following fragment of VHDL > code: > > ybus0 <= ("000000" & yy) when (s(0) = '1') else "ZZZZZZZZZZZZZZZZ"; > ybus0 <= ("000000" & ff) when (s(1) = '1') else "ZZZZZZZZZZZZZZZZ"; > ybus0 <= ("0000000000000000") when (s(2) = '1') else "ZZZZZZZZZZZZZZZZ"; > ybus0 <= ("00000000000000" & cmd) when (s(3) = '1') else "ZZZZZZZZZZZZZZZZ"; > The tool is probably trying to build multiple tri state drivers, one followed by another. I would rearrange the code so that it matches the H/W structure you have in mind, maybe use a intermediate signal 'ybus0p' which feeds the tristate driver. =-- --= Ross Swanson swanson@est07.md.essd.northgrum.com swan000@erols.com =-- --=Article: 6256
On Fri, 02 May 1997 18:20:16 -0700, Peter Alfke <peter@xilinx.com> wrote: > >It's easiest to "roll your own": snip >The only bad side is that it occupies two pins, and the trip points are >dictated by the device input threshold. Thanks, I knew I could do it that way, but I was hoping to save the pins. Rodney Myrvaagnes Associate Editor, Electronic Products rodneym@ibm.net 516-227-1434 Fax 516-227-1444 When possible, sailing J36 Gjo/aArticle: 6257
Ross Swanson wrote: > > Arrigo Benedetti wrote: > > > > I have some problems implementing a design with a MUX having a three > > state output with Synopsys FPGA Compiler 3.5a targeted to a Xilinx > > 4010E part. The MUX is implemented by the following fragment of VHDL > > code: > > > > ybus0 <= ("000000" & yy) when (s(0) = '1') else "ZZZZZZZZZZZZZZZZ"; > > ybus0 <= ("000000" & ff) when (s(1) = '1') else "ZZZZZZZZZZZZZZZZ"; > > ybus0 <= ("0000000000000000") when (s(2) = '1') else "ZZZZZZZZZZZZZZZZ"; > > ybus0 <= ("00000000000000" & cmd) when (s(3) = '1') else "ZZZZZZZZZZZZZZZZ"; > > > > The tool is probably trying to build multiple tri state drivers, > one followed by another. I would rearrange the code so that it > matches the H/W structure you have in mind, maybe use a intermediate > signal 'ybus0p' which feeds the tristate driver. > > =-- --= > Ross Swanson > swanson@est07.md.essd.northgrum.com > swan000@erols.com > =-- --= Try to rewrite it like this: ybus0 <= ("000000" & yy) when (s(0) = '1') else ("000000" & ff) when (s(1) = '1') else ("0000000000000000") when (s(2) = '1') else ("00000000000000" & cmd) when (s(3) = '1') else "ZZZZZZZZZZZZZZZZ"; -- ------------------------------------------------- Thomas Johansson, assistant Electronics Systems, LiTH e-mail: thomasj@isy.liu.seArticle: 6258
For those of you who got a chance to taste Hollywood Blonde and liked it and would like to have the beer delivered in their neiborhood, please email jim@greatbeerco.com with the name address and phone number of where you buy beer. Thanks Steve Casselman, President Virtual Computer Corporation and Associate Brewmaster The Great Beer CompanyArticle: 6259
Scott Thomas wrote: > > On Fri, 02 May 1997 12:20:21 -0700, Ed Barrett > <ed.barrett@worldnet.att.net> wrote: > > > > >As for speed, the 2032 offers a true 5.0 nsec tpd and 180 Mhz operation. > >Also, for CPLD architectures tpd is fixed. The Lattice 5.0 nsec is a > >fixed guaranteed worst case delay. > > Isn't this 5.0 nsec speed only for dedicated inputs (two on the 2032) > and no more than 4 product terms? Doesn't the timing change if I/O > pins are used as inputs, more prouct terms are used, if routing pools > are used? These may be fixed, guaranteed worst case delays, but don't > these cause the tPD to exceed 5.0 nsec? > > The point Steve was making is the equivalent Vantis device > (MACH111SP-5) doens't exceed 5.0 nsec.--i.e., Vantis' worst case is > Lattice's best case. > > Scott Thomas > Vantis Wrong again. The 5.0 nsec tpd iScott Thomas wrote: > > On Fri, 02 May 1997 12:20:21 -0700, Ed Barrett > <ed.barrett@worldnet.att.net> wrote: > > > > >As for speed, the 2032 offers a true 5.0 nsec tpd and 180 Mhz operation. > >Also, for CPLD architectures tpd is fixed. The Lattice 5.0 nsec is a > >fixed guaranteed worst case delay. > > Isn't this 5.0 nsec speed only for dedicated inputs (two on the 2032) > and no more than 4 product terms? Doesn't the timing change if I/O > pins are used as inputs, more prouct terms are used, if routing pools > are used? These may be fixed, guaranteed worst case delays, but don't > these cause the tPD to exceed 5.0 nsec? > > The point Steve was making is the equivalent Vantis device > (MACH111SP-5) doens't exceed 5.0 nsec.--i.e., Vantis' worst case is > Lattice's best case. > > Scott Thomas > Vantis -- Wrong again. The 5.0 nsec tpd is any I/O pin to any other I/O pin just like you would expect. No magic No hidden delay. EdArticle: 6260
Does anybody know what are the advantages/disadvantages for chips that were designed for target process CMOS or BiCMOS. Design engineers widely understand that CMOS consumes less power while BiCMOS eats lots of power but the trade off is that you get extremely high unity-gain bandwidths. Are there any convincing advantages/disadvantages on CMOS/BiCMOS? From a design point of view, what platform should a designer target his/her design between CMOS and BiCMOS to get the most cost effective chip out of the two processes. ******************************************************************** Please visit : http://www.jaring.my/mimos/bi/rd/icdc/icdcprof.html *******************************************************************Article: 6261
ADVERTISE TO OVER 100 MILLION PEOPLE FOR FREE!!! NOW YOU CAN ADVERTISE FREE AND GET DRAMATICALLY MORE RESPONSES THAN ADVERTISING ON NATIONAL TELEVISION With E-mail your potential customer is forced to read at least the headline of your letter. Unlike TV, magazines, newspaper ads, and direct mail where more than 99.99% of the recipients will skip right over your advertisement. TURNS YOUR COMPUTER INTO A CASH REGISTER: Conservative estimates indicate well over 100 MILLION people will have E-mail accounts in the next year! E-Mail Blaster turns your computer into a cash register by giving you FREE, immediate access to all of them. Do you think some of the more than 100 million people with E-mail addresses would be interested in your products or services? Even if you don't have your own products or services you can still make money buy promoting our computer software!!! You can make money with our new software called E-MAIL BLASTER Here's how: * By Using E-Mail To Sell Your Products & Services * By Using E-Mail To Sell our Computer Software such as: Email Blaster, Web Walker, Email Lock Box etc. * By Renting or Selling Your E-mail Addresses * By Becoming A Distributor for E-Mail BLASTER MUCH FASTER: With bulk E-mail you get responses back in 1 to 4 days instead of waiting weeks or months like with television or print advertising. You can begin filling orders the same day you send E-mail. FREE ADVERTISING WORTH MILLIONS: It costs millions of dollars to mail out millions of letters using the United States Postal Service! Now you can send your ad to MILLIONS of people for FREE using E-Mail BLASTER E-MAIL BLASTER IS ACTUALLY 3 PROGRAMS IN 1: E-Mail Blaster is amazing new technology and the most advanced bulk E-mail software on the market today. (1) E-Mail BLASTER automatically extracts E-mail addresses from America Online, Compuserve, Prodigy, and the other online services, the Internet, web pages, news groups, and hundreds of other sources. (2) E-Mail BLASTER automatically stores the E-mail addresses in your computer along with keywords that describe hobbies and interests to created highly targeted E-mail addresses. (3) E-Mail BLASTER sends your marketing letter or ad to MILLIONS of targeted E-mail addresses at the push of a single button! HOW DOES E-MAIL BLASTER WORK? E-Mail Blaster has dozens of special filters to automatically harvest E-mail addresses from major online services such as: America OnLine, Compuserve,Delphi, Genie, and of course the internet. AOL AUTO EXTRACTOR: E-Mail Blastert has many special features like the AOL Auto Extractor that let's you use a regular dial-up Internet connection to collect names and addresses from America Online. Collect America Online addresses without being a member. Allows you to download names and addresses from America Online without paying membership fees or online charges. Auto Extractor for all of the online services and Special E-mail database search engines will also be available at no additional charge. SIMPLE TO USE: E-Mail Blaster does all the work for you. Unlike other programs where you must cut & paste addresses, E-Mail Blaster does it for you automatically. And, unlike other E-mail programs, with E-Mail Blaster there is no need to purchase any additional software SPEED: Your time is important. E-MAILBlaster is DRAMATICALLY FASTER at extracting and sending E-mails than all the other programs on the market that we tested it against. NO SPECIAL EQUIPMENT NEEDED: E-Mail Blastert runs on a PC with Windows 3.xx Windows 95, Windows NT, a 386 or faster processor, 4 MB RAM, and a hard drive with at least 10 MB free and your own SLIP or PPP connetion. No other software needed. Also runs on a Macintosh with SoftWindows installed. ****************************************************************** This message was sent using E-Mail Blaster. E-Mail Blaster is The World's Leading Bulk E-mail software. With E-Mail Blaster you can advertise to over 100 million people for FREE! Cost: We are giving away Email Blaster your only cost is $29.90 License fee. THAT'S IT!!! Our company is in the software business of developing cutting edge products, so we know if you have the opportunity to use our software FREE you will be our future customers. We know we need you to make our software business a true success! ************************************************************************************************************** To take advantage of this opportunity, please fill out the order form below and fax to 1-609-933-1499. Don't miss out on this opportunity! Sales Rep:(Your Name) Name: _____________________________________________________ Address:____________________________________________________ City:___________________________________________ State:___________________________ Zip:_____________________________ Telephone::_____________________________________________ E-Mail Address:__________________________________________ #of Copies $29.90 each ______________ If you would like to run Email Blaster on multiple computers or if you would like to resell Email Blaster for up to $499. you may purchase additional copies of Email Blaster for only $29.90. Retails for $499. This price is only available at the time of this order! So, Take advantage now and reserve your additional copies now! 1 Electronic Help System only $19.95 ________ Unlimited Email Addresses (25 Million & Growing) only $29.95 _________ **************************************************************************** Complete Email Blaster Package only (Retails for $79.80) $49.95 _______ 1 Copy of Email Blaster 1 Electronic Help System: Email Blaster Electronic Help System will provide you with step by step directions on how to use Email Blaster and help with setting up your provider. For convenience this is built right into Email Blaster. Unlimited Email Addresses: 25 Million and Growning You can download these addresses from our website at your convenience. Now you have all the email addressses you could ever want or need! ********************************************************************************* If you would like to pay by check, please make check payable to: One-One Marketing mail to: PO Box 276 Bellmawr, NJ 08099 or Fax to 1-609-933-1499 along with this order form. If you are paying with a Fax Check please include an additional $3.00 for check processing fees. Credit Card: _______________________________________________ Visa, Master Card and AMEX. EXP. Date: _____________________ Total to be charged: $________________________ Signature:___________________________________________________________Article: 6262
Tim Forcer <tmf@ecs.soton.ac.uk.nospam> wrote: >Frank Dresig wrote: >>For those of you designing with Lattice (is)pLSI devices >>it might be interesting to check out our newly announced >>Lattice Resynthesis server. >>..cut.. it accepts a (is)pLSI design and resynthesiszes it to >>improve speed, resource usage and fittability. >>..cut.. The service is free, so at least it might >>be worth a try. >>..cut.. > >Tim Forcer <tmf@ecs.soton.ac.uk.nospam> wrote: > >..cut.. > >What is the position on Intellectual Property? Specifically, would >ISDATA hold any rights in the resynthesised design? > The answer to this one should be simple: Just running tools to process a design does not give the tool manufacturer any rights in the design. > >What security is there for designs submitted for resynthesis? > Here I think it is mainly a problem of what you can gain vs what the risk is. By chance, I've just seen a LAF pass the server coming in with 49 GLBs (not fitting a 1048) and returning with 48 GLBs (fitting a 1048). I hope this is a little help to the designer provided that all other possible obstacles (timing etc) can be removed for this design. So this might be an example of what you can get. The idea to offer such kind of service was born at the overcrowded "CAE on the Web" session at DAC 95. There a statement was made that the future of CAE usage could be to use the most approriate software for a given problem via the net, instead of installing lots of packages locally. This approach is fascinating, especially for the synthesizer / fitter or synthesis / P&R combination. As all synthesizers work on a heuristic basis, the very same design that costs you days (and nights) to find a fitting Lattice or routable XILINX solution, may work immediately when simply run through a different synthesizer using different heuristics. By the way, of course we think that our heuristics are somewhat better than the other guys heuristics :-) So this is a typical case where "CAE on the web" could be benefical. And the risk? Of course, all design data travelling the net are vulnerable. Of couse, we could not adhere to the text of our Web-Page "all design data are processed automatically and deleted after processing, with the exeption of some figures for our statistics". I have only basic knowledge of the legal side, but I assume, not deleting the design data sent to us under such an assumption, would be a breach of the law. In conclusion: For such a free service, every designer has to evaluate the gain / risk situation based on the security needs for the design. If "CAE on the web" becomes a reality in the future, with defined provider / customer relations, it should be possible to solve the security problem on a contract basis similar to an NDA. -- Albrecht Ditzinger ISDATA Karlsruhe WWW: http://www.isdata.de A very old bavarian saying says: Saving on design tools means paying a fortune for devices.Article: 6263
Vitit Kantabutra wrote: > Likewise, I don't know whether Don Husby's implementation was an > optimal one. But I am thankful that an implementation exists. > I never meant to imply that Don Husby endorses the algorithm. > However, I still think that our algorithm might be faster than the > traditional bit-by-bit ones, at least for larger word lengths. > > It is very important to note that our algorithm only requires ONE > full-length subtraction to retire 2-3 operand bits, plus a 2-bit > comparison and a little more simple logic. Admittedly, my > experience in only in full-custom design, not FPGA's. (And in > fact, I didn't think of FPGA's at all when I wrote that paper. > I didn't think about FPGA's until Don Husby and a lot of other > FPGA people wrote me email in response to my earliest Usenet News > posting a few weeks ago.) > > However, it does appear, intuitively speaking, that at least for a > large enough word length (maybe more than 8 bits?), a circuit using > our algorithm could be faster than one that needs 2 full-length > subtractions to retire 2 bits. I agree. For long word widths, there will eventually be some advantage for Kantabutra's algorithm. My guess is that crossover point is greater than 32 bits. Here's an explanation: FPGA's are based on lookup tables, so implementing some complex functions is no more costly than implementing simple functions. A 4-bit adder takes about the same number of resources and time as a 2-way 4-bit multiplexer. Also, the routing between logic blocks is much more dominant in an FPGA than a custom chip. The time to route signals between two logic blocks is typically 1/2 to 2 times the actual logic propagation delay: e.g. A logic block may take 3ns while the time to route signals beween logic blocks may take 1.5 to 6 ns. Carry paths have been optimized to eliminate this routing delay between blocks used as arithmetic units. Kantabutra's algorithm requires at least a 3-way word-width multiplexer plus one level of logic to control the multiplexer. To implement this for a 32-bit data path requires 9 ORCA PFUs (or 33 Xilinx CLBs) and two levels of logic. To implement a 32-bit sub-mux requires 8 PFU (or 16 Xilinx CLBs) and about the same amount of time as two levels of logic. ------- Included below is a C-language simulation of both algorithms. It was written using Borland C. ------- #include <stdio.h> #define Iter 9 /* Number of result bits (iterations) Must be < 23 */ #define Filt 16 /* Print any error >= 2^Filt */ void PrintB(char *S, long I, int Size) // Print Bits diagnostic { printf("%s",S); while (Size--) { if (Size%4==3) putchar(' '); if (I & (1L<<Size)) putchar('1'); else putchar('0'); } } long Div_S(long R, long D) // Simple Division Algorithm { // R is a 15-bit positive integer // D is an 8-bit positive integer > 128 // Return 16.16 integer.fraction format int I; // Number of bits in result int N=24-Iter; // Normalization count long T; // Temporary long Q=0; // Quotient D <<= 7L; // Normalize to 16.16 int.fraction format for (I= Iter; I>0; --I) { Q <<= 1; T = R-D; if ((T&0x10000L)==0) { Q |= 1; R = T; } R <<= 1; } Q <<= N; return(Q); } long Div_K(long R, long D) // Kantabutra's Division Algorithm { // R is a 16-bit signed integer // D is an 8-bit positive integer // Return 16.16 integer.fraction format long D2,D3; // 2*D and 3*D int I=Iter; // Number of bits in result int N=23-Iter; // Normalization count int Case; // Case is 0 to 3 as described below int T; // Temporary int Dbit; // Used in case determination long Q=0; // Quotient D <<= 8L; // Normalize to 16.16 int.fraction format Dbit=((D&0x4000)!=0); // Get 2nd MSB D2= D+D; D3= D+D+D; R |= -(R&0x8000L); // Extend sign of R R<<=1L; // convert to 16.16 if (R>0 && R>D || R<0 && -R>=D) // Ensure R<=D { R /=2; ++N; } while (I>0) // Main Loop { T=(R>>13L) & 15; // Get 4 bits of R x.xxx // Case determination. This can be done in a single 5-input LUT. if (T==0 || T==15) Case= 0; // R < 1/8 else if (T==1 || T==14) Case= 1; // R < 1/4 else if (T<3 || T>12) Case= 2; // 2R <= D else if ((T==3 || T==12) && Dbit) Case= 2; // 2R <= D else Case= 3; // 2R > D if (Case==1) // Handle cases { R <<= 1L; Q <<= 1L; I-=1; // Case 1, shift a zero into Q continue; } R <<= 2L; Q <<= 2L; I-=2; // Case 0,2,3 shift 2 bits if (Case==0) continue; // Case 0: shift 2 zeros if (Case==3) D=D3; else D=D2; // Choose D if ((T&8)==0) // Get sign of R { R-=D; Q += Case; } // If sign is positive else { R+=D; Q -= Case; } // If negative } Q <<= (N+I); return(Q); } main() { long R,D; long F,Q,X; int E=0; int Log; long Histo[32]; for (Log=32; Log--;) Histo[Log]=0; printf("Testing division algorithm with %d iterations\n",Iter); for (R=16384L; R<32768L; ++R) for (D=128; D<256; ++D) // Test all combinations { F= (R*65536L)/D; // Compute quotient using cpu Q= Div_K(R,D); // Compute using division algorithm X= (F-Q); if (X<0) X= -X; // Find ABS(F-Q) for (Log=0; X; X>>=1L) ++Log; // Find Log2(X) ++Histo[Log]; // Enter into histogram if ((Log>=Filt) && (E < 10)) // If error bigger than Filter, print { ++E; printf("%5ld / %5ld\n",R,D); PrintB(" Expect: ",F,32); printf(" =%10.5f\n",F/65536.0); PrintB(" Got: ",Q,32); printf(" =%10.5f\n",Q/65536.0); } } for (Log= 32; Log--;) printf("%2d %8ld\n",Log,Histo[Log]); printf("Press return to exit."); getchar(); return(0); } -------------------==== Posted via Deja News ====----------------------- http://www.dejanews.com/ Search, Read, Post to UsenetArticle: 6264
Asger Sporring wrote: > > I'm doing my master thesis on EvolHard, and am going to use the > Khepera robot for some practical testing. > > But I need to place a FPGA chip (right now I'm looking into using the > Xilinx chips(like Splash 2), but this is not fixed) on the robot, for > it to be of any use. > > Does anybody have any experience with this (using, I assume, the > general IO turret). > Embedded Solutions Ltd. markets a modular board package known as the Accelerator System. Modules are equipped with a mezzanine daughter board I/O connector. This system is planned to be supported by a commercial release of the Handel-C programming language, which enables a programmer to directly target the FPGA, without recourse to a HDL, in a similar fashion to a classical microprocessor cross-compiler. http://www.embedded-solutions.ltd.uk ******* Crossposted: comp.robotics.research (moderated) ******* Summary: Academic, government & industry research in robotics. Archives and information: http://www.robot.ireq.ca/CRR Charter: ftp://ftp.robot.ireq.ca/pub/crr/Charter Meta-discussions/information: crr-request@robot.ireq.caArticle: 6265
Arrigo Benedetti wrote: > > Dear all, > > I have a question about the instantiation of the STARTUP block in a > design implemented in Synopsys ver. 3.5a. > What it is not clear to me is the example in the Xilinx Synopsys > Interface for FPGAs ``Interface/Tutorial Guide'' presented in chapter 7, > fig. 7-1. What I don't understand is why the STARTUP component is not > instantiated in the count8 architecture: in this way in fact the GSR net is > totally useless. If you look at the count8_vss.vhd generated by xnf2vss > (a copy of it is in $XACT/tutorial/synopsys/vss/xc4000) you will see that > each flip-flop is connected both to GSR and CLR. > In other words my question is: is it possible to exploit the global GSR net > and still perform a timing simulation ? > I've not been able to try instantiating the STARTUP block and run xnf2vss > only beacause at my site the Xact core tools license has not been > installed yet. > > Thanks in advance > > -Arrigo > -- > Arrigo Benedetti e-mail: arrigo@vision.caltech.edu > Caltech, MS 136-93 phone: (818) 395-3695 > Pasadena, CA 91125 fax: (818) 795-8649 Yes, you are right. For BEHAVIORAL simulation the GSR-signal is useless, BUT if you are going to simulate the backannotated file (count8_vss.vhd) you still need the GSR-signal, because there IS a startup-phase in your design and so the flipflops will have a default-state even without any signals on the reset-pin (remember clock-devider-ff, they do NOT need an reset-signal for working). What XILINX did (maybe) is a timing-simulation of their behavioral-model. So my advice for you: - write your 'normal' VHDL-file (without any GSR, STARTUP ..) - create a higher level of hierarchy and instantiate your modul & GSR & STARTUP.. - create a testbench (one for both : behavioral AND timing simulation) - all things will work pretty fine...(hopefully) Hope that helps, HagenArticle: 6266
Fellow comrades, I need and covet your assistance. I'm looking for design, specification, reference material (etc.) on USB, PCI, and ISA interfaces. Any/all pointers will be most warmly appreciated. Thanx! -- Bob Elkind **************************************************************** Bob Elkind mailto:eteam@aracnet.com 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: 6267
-- ----------------------------------------------------------------------- John Cavazos cavazos@cs.umass.edu Department of Computer Science (413) 545-0256 University of Massachusetts at Amherst Fax: (413) 545-1249 Amherst, Ma. 01003 http://www.cs.umass.edu/~cavazosArticle: 6268
Let me try this again... CALL FOR PAPERS Computing Surveys Student Tutorial Paper Contest Computing Surveys is sponsoring an annual competition for the best student tutorial paper. The paper should be in the form of a substantive tutorial of 5000 to 12000 words that examines a topic of computer science at a level understandable by senior undergraduate students. It should conform to the standards of Surveys tutorials, combining tutorial clarity, historical and scholarly perspective, and technical interest. Submissions will initially be judged by a panel of students who will recommend between five and ten papers to a panel selected by the Editors of Computing Surveys. The panel will select the best paper from among these for publication in Surveys and select runners-up for honorable mention. Student ACM chapters as well as undergraduate computer science departments should actively promote this competition as a way for students to develop their writing and tutorial skills. Papers should have been written in the previous two years and should not have been previously published in commercial publications. However, papers previously published in the ACM student mazazine Crossroads or in other student publications will be accepted. The paper should be accompanied by a letter from a faculty sponsor that certifies it as the work of the author or authors and describes the circumstances under which it was written. Authors should normally be undergraduates or graduate students with less than two years of graduate school and should normally be members of the ACM, but exceptions will be considered based on letters from the faculty sponsor. International participation is encouraged, as in the student programming contest. Information on becoming a Student ACM member can be found at http://www.acm.org/membership/. Further information about the contest can be found at http://osl-www.cs.umass.edu/~cavazos/Contest/. Updated information will be posted on this page periodically. Submission for the 1997 competition are due by June 30, 1997. Students will make a recommendation to the Editors of Surveys by August 31, and the winner will be announced by October 31 and published in an issue of Surveys no later than the summer of 1998. It is desirable though not required that contestants indicate their intention to submit, with a provisional title. Submissions can be sent electronically to cavazos@cs.umass.edu, or submitted in hard copy form to: John Cavazos Dept of Computer Science University of Massachusetts at Amherst Amherst, Mass, 01003Article: 6269
>Does anybody know what are the advantages/disadvantages for chips that >were designed for target process CMOS or BiCMOS. >Design engineers widely understand that CMOS consumes less power while >BiCMOS eats lots of power but the trade off is that you get extremely >high unity-gain bandwidths. >Are there any convincing advantages/disadvantages on CMOS/BiCMOS? >From a design point of view, what platform should a designer target >his/her design between CMOS and BiCMOS to get the most cost effective >chip out of the two processes. CMOS advantages: low static power dissipation, most advanced feature size (if you want to push it), largest number of foundry sources, simpler process architecture usually means lower wafer cost. A very small geometry CMOS process may be faster in analog applications than a less advanced bipolar or BiCMOS process (see RF work being done in CMOS-only processes - Berkely?) CMOS disadvantages: No bipolars, some circuits not practicable or more difficult to implement BiCMOS advantages: richer device set. Bipolars are easier to make high frequency analog circuits, precision references, low-offset amplifiers with. And you can always constrain yourself to only use the CMOS if you're trying to keep static power down. CMOS circuits at speed may have more total losses (capacitance is higher) than bipolars. May allow more compact circuits for some types. BiCMOS disadvantages: More complex process, higher wafer cost, probably less advanced lithography (process probably took longer to get up and running since there were more device types to global-optimize). Since you can use the CMOS in either case, designers would generally opt for the greater up-front design freedom. Meanwhile whoever's holding the money bag will press for straight CMOS implementations. If you know the circuit functional content you can make a more considered choice. There is no generalization that works - else why would there be options? -- ########################################################################## #Irresponsible rantings of the author alone. Any resemblance to persons # #living or dead then yer bummin. May cause drowsiness. Alcohol may inten-# #sify this effect. Pay no attention to the man behind the curtain. Billy!#Article: 6270
i just did (ie copied) the circuit illustrated in [figure 5] of 'Efficient Shift Registers, LSFR Counters, and Long Pseudo-Random Sequence Generators' (p. 4) I used Mentor Graphics for the schematic capture. I had a few problems when simulating (quicksim2) The address sequence seems to be 16 cycles long (which is right), the RAM seems to be loaded with the right values, BUT as soon as I switch load to low one of the following happens, depending on my loaded-in sequence. 1) The output is always 0 2) The output is always 1 3) The output is the same as any of the address bus bits, but shifted and repeats every 16 cycles... Any ideas? Did I do something wrong? (I used 16x1 xc4000 RAMs in place of the 15x1 RAMs displayed in the schematic I always hold WE high Presumably the address-generating FFs feedback to Q1..Q4) *HHEELLP* -- Christos Dimitrakakis --------------------- mailto:mbge4cd1@fs4.eng.man.ac.uk mailto:mbge4cd1@afs.mcc.ac.uk http://www.man.ac.uk/~mbge4cd1Article: 6271
Hi, we are looking for an universal PCI interface for a board design. A possible solution would be a FPGA based PCI bridge. Are there any practical experiences with such PCI interfaces ? Thanks in advance RalfArticle: 6272
> From: jws@billy.mlb.semi.harris.com (James W. Swonger) > Newsgroups: sci.electronics.design,sci.electronics.misc,sci.electronics.cad,sci.electronics.components,sci.electronics.repair,sci.engr.semiconductors,comp.arch.fpga,comp.arch.embedded > Date: 7 May 1997 13:25:37 GMT > Organization: Harris Semiconductor, Melbourne, Florida > > >Does anybody know what are the advantages/disadvantages for chips that > >were designed for target process CMOS or BiCMOS. An important issue which was left out of the previous list is the scalability of the process to smaller feature sizes (and therefore lower voltages). Right now, a .25 micron process really wants to run at about 2.5 volts or less; as we scale to .15 or below, the voltages continue to go down. Unfortunately, the Vbe of bipolar devices does not scale with dimension, but remains a constant (roughly) since it depends on the log of the doping concentrations and the temperature. So, since for most bipolar circuits we need 5Vbe to 3Vbe of power supply voltage, and Vbe is about 0.75 volts, we have a severe conflict. BiCMOS thus has a very limited life span, due to scaling laws. Do your design in pure CMOS and take advantage of the next 15 years of scaling. Note that this is what Intel is doing, as opposed to what they were saying three years ago, when BiCMOS was the "killer technology" which would soon wipe everyone else off the map. Anyone with a half decent process oriented education knew this was BS then, but the trade and financial press thought it was great. Homework: Compare and contrast Exponential.Article: 6273
I wrote this app note, and I never heard of a problem with it. But then we usually don't know how many people implement such app notes.... The address counter is supposed to divide by 15, not by 16, as you wrote. The front page of the app note describes how this 4-bit counter divides by 15, but there are hundreds of other ways to build a 4-bit counter that skips one of its codes. I wrote 15 x 1 into the RAM because that's the way they are being used. Of course, they are really 16 x 1 RAMs, but one location is never used. Four times fifteen is sixty. Plus three flip-flops makes it a 63-bit shift register. That was carefully chosen, since 61, 62, 64, 65, 66, 67 etc -bit LFSR counters require a more complex feedback. Best way to debug is to break the feedback chain and exercise this design as just a 63-bit shift register. If that doesn't work, look at intermediate points and see where the mistake is buried. The divide-by-15 counter has to work, and the 63-bit shift register has to work before you can close the XNOR loop and make it an LFSR. Please e-mail me with the results of your debugging. peter@xilinx.com Peter Alfke, Xilinx ApplicationsArticle: 6274
Will be responsible for analog and digital circuit design for wide-dynamic low noise instruments. Requires a high degree of ability to make tradeoffs between competing goals and customer requirements. These designs have a high degree of interaction between analog and digital circuitry in order to make the desired measurements. The position will utilize a wide range of design skills, from frequency analog design, to nonlinear systems analysis, to DSP (digital signal processing) and real-time firmware design. Requirements: BSEE minimum (MSEE preferred) DSP C and/or C++ Windows NT or Windows 95 experience is desired Superb written and verbal communication skills Lloyd D. Songne, Jr. Regional Recruiting Manager Volt Services Group Email: volt1@ix.netcom.com Visit our interactive employment web site at http://www.volt-tech.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