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
Hooman Dadrassan wrote: > > Hi, > > my problem is that I am running out of pins on my xilinx 4010 fpga > and I want to > use the mode pin M1 as output and I beleive I should be able to do that > based on > xilinx databook indicating M1 as O (output) after configuration. > I have tried pin LOC but I get error message while mapping in the M1.4 > tool. > Does anybody know how I can do that. > > Thanks, > Hooman Use MD1 component ( in the library ) instead of OPAD. ( Pins M0 and M2 , could also be uses as inputs : component MD0 and MD2 ) sorry for my english.Article: 12801
i don't think i'm making myself clear here. HJ's point is that you should select the largest cap because, to paraphrase: "...then the effective series inductance, the single specification that defines the performance of these parts at high frequencies, is the same for both parts". *but* this isn't relevant - we need performance at the frequencies of interest, not "at high frequencies". the required specification is actually the product of the nominal capacitor value, and the sum of the inductances involved - dielectric, construction, packaging, trace - ie. the specification that controls the SRF. as a designer, your control over inductance is limited. an 0805 with a good layout may give you 5nH; an 0508 with a excellent layout may give you 2nH. but your choice of capacitor value ranges over a factor of about 100 - you get much more control by selecting capacitance. obviously, you have to control inductance. but you must also select capacitance to ensure that you have a low impedance at the required frequency. there's no point using a 100n cap if it turns out that it has a 10ohm impedance at your important frequency. in short, it's *not* adequate to select the highest available capacitance for a given package and layout, as HJ states - you have to select the capacitor appropriately. as for direct via connections, rather than explicit trace connections, my point is that you don't know (or i don't, anyway) what actually happens on the path from a capacitor terminal, through a via, to a plane which is shared by lots of other logic, through another via, to a power pin. it would probably take a supercomputer to model this lot. it doesn't seem to me that you can simply model this as two capacitors in parallel. so, why not take the easy route, and connect your chip capacitor directly to the required pin? ok, it adds some inductance, but you can arrange for this to have a minimal effect at the frequencies of interest. evanArticle: 12802
On 29 Oct 1998 02:39:08 GMT, "Mankit Wong" <lsckwong@netvigator.com> wrote: >Hi there, > >Could anyone tell me where I can get 8051 VHDL Model for free in the >Internet? > >Thanks, >Kit http://www.ee.umr.edu/~mrmayer/8051/ seems to be popular. you may also find some more on the VHDL faq: http://vhdl.org/comp.lang.vhdl/ evanArticle: 12803
On Fri, 30 Oct 1998 12:11:04 -0500, "Alan R. Sieving" <asieving@mediaone.net> wrote: >Table of Impedance (Ohms) at Freq (MHz) > >Part 10MHz 100 300 1000MHz > >A(102) 14 1.16 1.4 6 >B(103) 1.5 0.525 1.9 6 >C(104) 0.11 0.617 1.9 6 >D(104) 0.13 0.369 1.1 3.8 >E(684) 0.03 0.376 1.1 3.8 interesting results. i haven't tried spicap, but these are presumably figures at the cap itself, and you don't get an option to add in external inductances - is this right? if this is the case, then it would be more informative to get in-system figures. AVX's inductance figure for an 0805 is 1.4nH, and i'm guessing that the 0612 is about 500pH. i'm also guessing that a good layout, with direct plane connections, would give you an additional 2nH, giving a total of 2.5nH for the 0612. the high-frequency impedance goes as 2.pi.f.L, so it would increase by a factor of maybe 5 for a given high frequency. what's more interesting, though, is that the impedance curve would rise more quickly from the SRF point, and so the high value caps would look a lot less attractive. has anyone got the time/inclination to plot this? evanArticle: 12804
On Fri, 30 Oct 1998 15:21:23 +0000, Jonathan Bromley <jsebromley@brookes.ac.uk> wrote: <snipped> agreed, but if i can just clarify what i was originally suggesting - point (8) of my first message included a via in the track from cap gnd to device gnd; i suggested limited vias on pwr only, for noise suppression reasons on high speed devices. you only had the summary in the message you replied to. i personally haven't clocked an FPGA above 67MHz, and my devices currently have tracks and vias on both pwr and gnd. second - my 'frequency of interest' is simply the clock speed. a lot of internal charging and discharging is going on at the clock rate, and i'm just generalising the bulk bypassing argument. agreed, the current transients will have frequency components at much higher rates, whatever their source, but you've got to start somewhere. evanArticle: 12805
As mentioned before, use CORDIC. A cordic engine is about the same complexity as a multiplier of the same width, and gives very good accuracy. An additional benefit is that it simultaneously generates both sine and cosine. Juergen Kahrs wrote: > Yves Vandervennet wrote: > > > does anybody know how to digitally realize a sine generator > > other than sampling a sine period and storing it in a ROM ? > > We have to integrate it in an FPGA. If anybody knows book references > > on this subject, we would be happy for a very long time. > > Remember the sine is the solution of a simple differential > equation. Turn the differential equation into a difference > equation and you get > > x2 = (2 - (2*pi*fsin/fsam) ** 2) x1 - x0 > > A -- -Ray Andraka, P.E. President, the Andraka Consulting Group, Inc. 401/884-7930 Fax 401/884-7950 email randraka@ids.net http://users.ids.net/~randrakaArticle: 12806
Hullo, I have a question about the use of the M2,M1,M0 pins of XC4013XL FPGAs. What are the correct assignement in the following situations: - configuration with an SPROM (must be "000") ? - configuration with F1.5 "JTAG Programmer" and serial Xchecker cable (the manual of JTAG Programmer says "000") ??? - configuration with F1.5 "Hardware Debugger" and serial Xchecker cable (must be "111") ? Is it correct to connect /INIT pin of XC4kXL devices like it is described on the next scenarios ? - when configuring on Asynchronous Peripheral Mode, /INIT is connected to Vcc with a 4.7K - when configuring with "JTAG Programmer", /INIT is connected to GND with a 1K - when configuring with an SPROM, /INIT is connected to Vcc with a 22K The last question: Can we program a daisy chain with 4 XC9500 CPLDs and 4 XC4013XL FPGAs with serial Xchecker cable ? Thank you. ----------------------------------------------------- Antonio J A Esteves Dep Informatica, Universidade do Minho Braga, Portugal E-mail: esteves@di.uminho.pt Web: http://www.di.uminho.pt/~esteves/ -----------------------------------------------------Article: 12807
ems@riverside-machines.com.NOSPAM wrote: <amongst other things> > so, why not take the easy route, and connect your chip > capacitor directly to the required pin? ok, it adds some inductance, > but you can arrange for this to have a minimal effect at the > frequencies of interest. Sorry, I don't think this will do (as an argument - I concede it often works OK in practice). Lots of problems. 1) What on earth (pun intended) do you mean by "frequency of interest"? Decoupling on MOS LSI devices is primarily there to supply the current transients that occur on switching edges. These transients are _narrow_ (like you would not believe) and their narrowness is _not_ a function of pulse repetition rate. Ergo, spectrum of current through capacitor is exceedingly broad (out to 1GHz in fairly ordinary modern logic families) although of course it will have big peaks at multiples of your pulse repetition rate. At the highest of these frequencies it is unquestionably the equivalent series inductance of the various current paths which dominates. At these high frerquencies, tiny capacitances will do the job of decoupling. 2) Traces from decoupling cap direct to device pin are often a rather bad idea IMHO. If the decoupling is there to support switching transient currents that occur entirely _on-chip_ then you are in good shape - because you have soaked up all the HF components of the device's supply current. The (presumably rather inductive) path from the device/cap combination to the nearest really solid ground point will be carrying only slowly varying currents so its inductance doesn't hurt you. BUT if, as is usually the case, some significant edge-related transients flow into and out of the device's signal pins then those same currents must flow from the supply into one or other of the device's power pins. If it's the ground pin, then no amount of local decoupling will do. The current transient, flowing in the path from chip+cap to "real" ground, will move the chip's ground away from "real" ground and this dropped voltage will appear in series with any inputs to the chip driven from devices whose output reference is the aforementioned "real" ground. This is, more or less, the famous "ground bounce" problem. You could hypothesise an ideal IC which had many nanofarads of supply bypass actually built on the chip (not very practical!) and it would still go nowhere towards fixing this problem. The sane solutions to this problem are: a) make all logic signals on the board DIFFERENTIAL so that small movements of one device ground relative to another are insignificant - this is exactly how traditional ECL worked b) minimise output slew rates so that off-chip capacitances are charged as slowly as possible by changing signals - this is the main motivation for going to low-swing logic systems, because you can't increase the rise/fall time beyond, say, 20% of a system clock period c) aim for the lowest possible inductance between chip ground and the system ground reference Of these, only (c) is realistic for most people. All the usual rules of thumb - ground planes, thick ground traces, small SMT chip packages - are important. NOTE NOTE NOTE - bypass caps don't come into it, not even a little bit! Where the bypass caps do come into their own, of course, is in helping to keep the Vcc power line well behaved under these conditions. It seems to me that we have two problems to solve, and that the solutions are almost conflicting. We must supply chip internal transient currents, which means caps as close to the chip as possible - Evan's solution. These currents can be huge on big VLSI devices with lots of internal nodes changing state simultaneously and fast, which is why Intel recommend 24 separate 10nf caps for bypassing Pentium supplies - many parallel current paths so that their inductances hit you less. Then we have to source the currents that are needed to charge off-chip nodes as they change state. The caps that source this current _absolutely should not_ have their ground ends linked directly to chip ground pins, because that means the currents flowing in them will also be flowing in the inductance of the chip's ground track and will shift the chip's local ground undesirably. The cap _grounds_ should be returned to the ground plane by independent, short traces; the cap _supply sides_ should be linked to their corresponding chip power pin by the shortest possible traces. Of course, using the distributed C of closely adjacent power and ground planes is a thoroughly good thing, as others have pointed out. Moral: There is no substitute for knowing WHERE YOUR TRANSIENT CURRENTS ARE FLOWING, and remembering that VOLTAGE SHIFTS ON INDIVIDUAL DEVICE GROUNDS IS A BAD THING. Finally, of course, the total parallel capacitance between power and ground on the board should be big enough to soak up any slower current pulses, so an earlier and rather apologetic suggestion about 10uF near the power inlet connector is not to be sneezed at. But on a big and busy board you may well find that the parallel sum of all the distributed bypass caps is enough to spare you the trouble of an explicit bulk bypass cap. I'm sure I've said stuff here that has been better stated elsewhere, but I felt it was no bad thing to bring the discussion back to basics - in my experience this topic generates an inordinate amount of eye-of-newt folklore-driven comment, and a bit of simple circuit theory carefully applied can go a long way to clearing the fog. Jonathan BromleyArticle: 12808
Ad Verschueren <ad@akebono.ics.ele.tue.nl> wrote: > Hierarchy and stucture can (someone correct me if i'm wrong :-) > *only* be represented with entities and components - whereas (the > more complex) behaviour is captured in processes. Processes are > 'flat' pieces of procedural code - although they can (and should) be > structured, that is not the structure we are looking for... Actually, you can also use block statements. Paul -- Paul Menchini | mench@mench.com | "Se tu sarai solo, Menchini & Associates | www.mench.com | tu sarai tutto tuo." P.O. Box 71767 | 919-479-1670[v] | -- Leonardo Da Vinci Durham, NC 27722-1767 | 919-479-1671[f] |Article: 12809
Virtual <sxyzami@pacifier.com> wrote in article <36380250.0@news.pacifier.com>... > Eric Pearson <ecp@focus-systems.nospam.on.ca> wrote: > > Has anyone noticed problems in re-compiling designs that worked in v8.0 of > > maxplus2 under v9.01 and found comilation times went throught the roof even > > to the point excess ( I aborted after 100 hours v9 vs 1 hour under v8 ). > > You may want to perturb your design (timing constraint) a little bit. > I have seen 100+ hours compile times with some 6.x version on a > design that normally compiled in 1 - 2 hours. Changing the timing > by 0.1 Mhz fixed it. > At 92% utilization, adding any constraints other than pinouts makes the design unroutable (200+ hours) under v8. (200Mhz PPro, 256M). I realize that I'm near the hairy edge of routability on the 10K130, but something is 'fundamentally wrong' with Maxplus2 v9.0 if it cannot route a design which V8.0 can. Please, Altera give me a switch to select the 'old' router versions! One has to wonder what these maintenance fee's are going towards when they are helping Altera create enhanced products which don't work on previous designs. Maybe I'm expected to archive the compiler along with the design files, as well as trying to recompile 100+ designs every 3 months as they do compiler upgrades. Dealing with bugs is challenging enough, just imagine calling tech support for an old version of software! Eric PearsonArticle: 12810
Hi - On Fri, 30 Oct 1998 11:54:20 GMT, ems@riverside-machines.com.NOSPAM wrote: >i don't think i'm making myself clear here. HJ's point is that you >should select the largest cap because, to paraphrase: > >"...then the effective series inductance, the single specification >that defines the performance of these parts at high frequencies, is >the same for both parts". > >*but* this isn't relevant - we need performance at the frequencies of >interest, not "at high frequencies". the required specification is >actually the product of the nominal capacitor value, and the sum of >the inductances involved - dielectric, construction, packaging, trace >- ie. the specification that controls the SRF. > >as a designer, your control over inductance is limited. an 0805 with a >good layout may give you 5nH; an 0508 with a excellent layout may give >you 2nH. but your choice of capacitor value ranges over a factor of >about 100 - you get much more control by selecting capacitance. > The point Howard Johnson is making---and correctly, I think--is that, at very high frequencies beyond the point at which the impedance bottoms out, the impedances of the .1uF and .001uF caps are the same, because the impedances are dominated by the inductance at that point. To picture an impedance curve, imagine a "V" shape that represents the impedance vs. frequency for the 0.1uF capacitor. Near the right-hand (ascending) leg of the V, draw a diagonal that falls to the right and intercepts the ascending leg, say, half-way up. This second V describes the impedance of the .001uF capacitor. (This isn't exactly correct, but it's close.) What this graph shows us is that, at very high frequencies, the impedance of the two devices is the same, and is determined by package inductance. (Some may object by pointing out that, at these frequencies, the impedance is inductive. Howard Johnson has a simple answer for this: so what? We care about the power-supply-to-ground impedance; whether it's reactive in one direction or another isn't going to matter nearly as much as the magnitude of that impedance.) What's more, there's a range of frequencies for which the impedance of the 0.1uF capacitor is lower than that of the 0.1uF cap. I also agree with Howard Johnson's claim that it's better to sink capacitor vias directly to the power and ground planes, rather than try to connect the capacitors with traces to chip Vcc and ground pins. I want a "sea" of capacitors with low-impedance connections to Vcc and ground, not just one cap for a given Vcc/ground pin pair. I guess I could appeal to authority here, and point out that Dr. Johnson has been doing this stuff for years, and has written a terrific book. But the reason I believe in what he's recommending is that I worked for years with a signal integrity engineer who independently came up with the very same suggestions that Dr. Johnson has made on bypassing. We used those suggestions, built some very large, high-speed boards, and found that they had Vcc and grounds that were clean as a whistle; the boards worked great, too. I think that most of the suggestions made in this thread will probably work for moderate-to-high-speed logic, even if I don't consider the suggestions ideal. However, when speeds increase even further, watch out! Bob Perlman Cambrian Design WorksArticle: 12811
ems@riverside-.NOSPAM wrote: > > i don't think i'm making myself clear here. HJ's point is that you > should select the largest cap because, to paraphrase: > > "...then the effective series inductance, the single specification > that defines the performance of these parts at high frequencies, is > the same for both parts". > > *but* this isn't relevant - we need performance at the frequencies of > interest, not "at high frequencies". the required specification is > actually the product of the nominal capacitor value, and the sum of > the inductances involved - dielectric, construction, packaging, trace > - ie. the specification that controls the SRF. I've been wrestling with what kind of caps to use for an FPGA design recently, and here are my thoughts. (I went to H. Johnson's seminar, and I highly recommend his seminar, book, and mailing list to anyone interested in this topic.) Basically, it seems there are four things we can control as designers: 1) number of caps more is usually better 2) cap package low inductance is better 3) layout " " " 4) nominal cap value more is often better. Rather than using my own guesstimates, I went to AVX's web site, and used their SPICAP software (http://www.avxcorp.com/software/asp/spicap/spicap.htm) to graph the impedance of several caps. I encourage you to try this yourself, but here is a summary of my results: Parts I tried: A = .001 uF 0805 X7R (08053C102...) B = .01 uF 0805 X7R (08053C103...) C = .1 uF 0805 X7R (08053C104...) D = .1 uF 0612 X7R (06123C104...) E = .68 uF 0612 X7R (0612YC684...) (E is 16V part, others are 25V) Place your bets on which part(s) are better... Table of Impedance (Ohms) at Freq (MHz) Part 10MHz 100 300 1000MHz A(102) 14 1.16 1.4 6 B(103) 1.5 0.525 1.9 6 C(104) 0.11 0.617 1.9 6 D(104) 0.13 0.369 1.1 3.8 E(684) 0.03 0.376 1.1 3.8 It seems to me that I don't give up too much at the high end by choosing a cap that has more capacitance, and that it helps at the low end. Also, it seems clear that the wider 0612 parts are better. To my understanding, HJ's claim holds true. For those who are interested, here is the resonant frequency and approximate impedance at that frequency Part SRF Z@SRF A(102) 159 MHz 0.65 Ohms B(103) 50 0.24 C(104) 16 0.07 D(104) 21 0.07 E(684) 8 0.03 I'll admit that you may be able to find a cap that works better at a given frequency by paying attention to SRF, but for general decoupling, I'm going to stick to the AVX 0612 package with the whopping 0.68uF capacitance. Cheers- -Alan SievingArticle: 12812
HI, This project intrigues me. You must be commended for actuaoy doing it. I've been "thinking" about doing something similar for a coupole of years, but haven't done anything. You have. Congrats! I was thinking of doing something like this to get into VHDL. I wonder how tough it would be to do it in VHDL. I expect that it would not be as compact as your schematic based version, but may be easier to add features to. Comments? >- Do the first address calculation for the operate instructions, jump > indirect, push and pop during decode (would reduce these by one cycle). > The cost is decode being done in series with ALU operation, so the > frequency would be lower. Also, the immediate instructions would be one > cycle longer unless I add a seperate incrementor for the PC (right now the > ALU does it). Did you impliment the PC as a latch, or a counter? If you haev enough flip-flops you could pipeline the PC. Impliment the PC as a counter that feeds a latch. Clock them both at the same time. You xfer the "new" PC to the latch, and increment it at the same time. Tim Olmstead email : timolmst@cyberramp.net Visit the unofficial CP/M web site. MAIN SITE AT : http://cws86.kyamk.fi/mirrors/cpm PRIMARY US MIRROR AT : http://www.mathcs.emory.edu/~cfs/cpm SECONDARY US MIRROR AT : http://CPM.INTERFUN.NETArticle: 12813
In article <3639ffe4.2028025@newshost.cyberramp.net>, <timolmst@cyberramp.net> wrote: )HI, ) )This project intrigues me. You must be commended for actuaoy doing it. )I've been "thinking" about doing something similar for a coupole of )years, but haven't done anything. You have. Congrats! [snip] Um, I found it interesting, but wonder how practical it is. The chips cost $15 each, and I can get an 8031 which has twice the address space for $3.95 in single quantity. And they don't require special programmers. Mike -- ---- char *p="char *p=%c%s%c;main(){printf(p,34,p,34);}";main(){printf(p,34,p,34);} This message made from 100% recycled bits. I don't speak for Alcatel <- They make me say that.Article: 12814
Be careful about using M0, M1, or M2 if you plan to use the spartan. M1.5 will not let you use the mode pins for anything. If you have not used init you might think about that. -- Steve Casselman, President Virtual Computer Corporation http://www.vcc.comArticle: 12815
Does the stack pointer grow down or up? I would like to see a subtract from stack pointer, for temporaries allocation (this assumes that stack grows down). A subtract reverse can be useful on a register limited architecture like this. I can't see anyway to get the stack pointer easily into the accumlator(?). This could help, especially when testing for stack overflows and such. I would like to see more specific semantics of the instructions, like which ones set which flags, for instance. Would it be possible to use the INC/DEC as semaphore lock? It may be useful to have another flag bit which can be seen by the outside world, which could be used as a "supervisor" mode flag with an appropriate external MMU (or internal on a bigger FPGA). However, I haven't thought this through. In any case, congratulations! Tim McCaffreyArticle: 12816
jai wrote: > Hi. > Has anyone used a FPGA to interface to a 3.3V PCI bus without > connecting any clamping diodes to the 3.3V rail. I know this is against > the PCI spec but since some FPGA's ie.Altera's 6000 devices and Xilinx > devices (without compromising 5v I/O tolerance) do not facilitate this. > > Tx, > Jai. The Xilinx 4000X{L,LT,V} have seperate voltages for the I/O so if you use 3.3v on the I/Os you should have no problem. The Virtex has control over different banks so half the I/Os could be 3.3v and others 5V (or many others). I have not used this on a 3.3v system but the HOT2 board can be configured to do this. http://www.vcc.com/Hotii.html -- Steve Casselman, President Virtual Computer Corporation http://www.vcc.comArticle: 12817
Mike McCarty wrote: > Um, I found it interesting, but wonder how practical it is. The chips cost > $15 each, and I can get an 8031 which has twice the address space for > $3.95 in single quantity. And they don't require special programmers. It is very practical-- given some conditions. When you start looking at not just the CPU, but the entire system, then there are lots of places that a small CPU like this would work wonders. Here's an example: Let's say that you are designing an intelligent ethernet bridge. You have two ethernet jacks and you want to filter and bridge ethernet packets between them. If you used this with "standard" shelf chips you would need a reasonably fast CPU, two ethernet controllers, some RAM, and some ROM. The CPU might cost $20 because you need a fast one to keep up with dual 100 mbps ethernet and move all that data around. You'll need a fair amount of 32-bit wide RAM, and some Flash to store the program. Including the ethernet controllers, you're looking at $70 to $90, depending on the exact configuration. But if you designed an FPGA to do the same thing, the FPGA could replace the CPU, both ethernet controllers, and the Flash. You would still need RAM, but not as much/wide. You could easily fit this into a 40 Kgate chip that costs on the order of $40 (Xilinx Spartan XCS40, for example). The total cost would be much smaller ($40 vs. $70) and the size would be less too. For this to work, you would use a small custom CPU to do the packet filtering and the actual data movement would be done with dedicated logic (not by the CPU). This would be much more efficent since the CPU could have special "filtering and data moving" opcodes that would speed up the process. This, combined with the data moving logic, would be the reason why an 8 bit CPU could replace a 32 bit CPU in the traditional approach. This is just one example of how a custom CPU would outperform a typical uC in price/performance. (I'll agree that the proper solution to the ethernet bridge is to use one of the many bridge chips on the market, but this is only an example and you can brew up your own application that is similar...) Hope this helps! David Kessner davidk@peakaudio.comArticle: 12818
$B%F%l%SElD.3t<02q wrote in message <01be042b$13f3c440$LocalHost@thc>... >$B!A$40FFb!A(J >$B!!$I!<$b!"=i$a$^$7$F!"%F%l%SElD.3t<02q<R$N>kK\$H$$$$$^$9!#(J >$B$5$F!"$3$N$?$S!"J@<R!"%F%l%SElD.$,H/9T$7$F$*$j$^$9>pJs;o(J >$B!V7n4)!&%F%l%SElD.$@$h$j!W$H$$$$$^$9!"%a!<%k>pJs;o$N$40FFb$r(J >$B$5$;$F$$$?$@$$$F$*$j$^$9!#(J >$B!!>pJs;o$NFbMF$O!"2<5-$NDL$j$G$9!#(J > >$B5-(J >$B!!!Z>pJs;o$NFbMF![(J >$BFI<T$+$i$N>pJs(J >$B%F%l%SEl5~7ONs$GJ|AwCf$N%"%K%aHVAH>pJs(J >$B%F%l%SEl5~7ONs$GJ|AwCf$N!"$=$NB>$NHVAH>pJs(J >$BFI<T$+$i$N46A[$d!"$40U8+(J >$BF`NI8)$N>pJs(J >$B$=$NB>$$$m$$$m(J > >$B!!!!$3$N!">pJs;o!V7n4)!&%F%l%SElD.$@$h$j!W9XFI$r$44uK>$5$l$^$9J}$O!"(J >$B!&;aL>(J >$B!&6a$/$K$"$k%F%l%SEl5~7ONs!J2<5-;2>H!K!"(J >$B!!!!!!%F%l%SEl5~!"%F%l%SKL3$F;!"%F%l%S0&CN!"%F%l%SBg:e!"(J >$B!!!!!!%F%l%S$;$H$&$A!"#T#V#Q(J >$B!&$"$J$?$N$$$A$P$s9%$-$J%F%l%S6I$H$=$N%A%c%s%M%k(J >$B!&$"$J$?$N%a!<%k%"%I%l%9(J >$B!!!!!!!!!!!!!!!!!!$rL@5-$N>e2<5-$N%a!<%k%"%I%l%9$^$G$h$m$7$/$*4j$$$7$^$9 !#(J >$B!!!!!!!!!!!!!!!!!!!!(JTv-higashimachi@ma3.justnet.ne.jp > What he said! -- Andy Peters Sr. Electrical Engineer National Optical Astronomy Observatories 950 N Cherry Ave Tucson, AZ 85719 520-318-8191 apeters@noao.eduArticle: 12819
In article <363A2875.F80C4B0B@peakaudio.com>, David Kessner <davidk@peakaudio.com> wrote: )Mike McCarty wrote: [snip] )This is just one example of how a custom CPU would )outperform a typical uC in price/performance. ) )(I'll agree that the proper solution to the ethernet bridge )is to use one of the many bridge chips on the market, )but this is only an example and you can brew up your )own application that is similar...) ) )Hope this helps! No, it didn't. If you look, you'll see that I work for a major manufacturer of telephony equipment. So I fully understand the use of specialty little CPUs made from programmable logic, having used them a few times myself. I wanted to know about *this* device. It doesn't seem to excel at anything, so I don't see the particular advantage to it, and it isn't cheap (for the hobbiest types). Mike -- ---- char *p="char *p=%c%s%c;main(){printf(p,34,p,34);}";main(){printf(p,34,p,34);} This message made from 100% recycled bits. I don't speak for Alcatel <- They make me say that.Article: 12820
Steve Casselman wrote: > jai wrote: > > > Hi. > > Has anyone used a FPGA to interface to a 3.3V PCI bus without > > connecting any clamping diodes to the 3.3V rail. I know this is against > > the PCI spec but since some FPGA's ie.Altera's 6000 devices and Xilinx > > devices (without compromising 5v I/O tolerance) do not facilitate this. > > > > Tx, > > Jai. > > The Xilinx 4000X{L,LT,V} have seperate voltages for the I/O > so if you use 3.3v on the I/Os you should have no problem. > The Virtex has control over different banks so half the I/Os > could be 3.3v and others 5V (or many others). > > I have not used this on a 3.3v system but the HOT2 board > can be configured to do this. > > http://www.vcc.com/Hotii.html > I have to correct myself that the Virtex has 5V tolerant I/Os. http://www.xilinx.com/partinfo/databook.htm#virtex -- Steve Casselman, President Virtual Computer Corporation http://www.vcc.comArticle: 12821
We have added a few experiments and a lot of doc to the www site. See the sig at the bottom. ================================================================= Design Your Own Processor(tm) Tools are aimed at people who want to develop and debug custom processors and associated application software. Advanced logic development is also possible with these tools. Design Your Own Processor(tm) is an ICE box for custom FPGA processors. You can develop the software and simulate your processor with our tools. No need for your own running hardware. Design Your Own Processor(tm) Tools consists of two assembled and tested boards. A Z-80 control processor running a modified fig-Forth at 8MHz clock speed. And an FPGA board that has an 84 pin PLCC socket suitable for a Xilinx XC4000XL series FPGA. Other Mfg.s FPGA boards will be available in the future. All PALs, EPROMs, and Clocks in a can are socketed with high quality machine tool gold or gold inlayed sockets. Detailed specifications: Z-80 control board: Z-80 processor 32K x 8 EPROM socketed. 32K x 8 RAM 20nS or better Z8530 dual serial ports with dual crystals one port is a DB-9 RS-232 port and the other port is RS-485. The crystals are 3.686 MHz and 4.9152 MHz There are 3 8255 parallel ports. 72 lines 8254 Counter/Timer Chip A high speed bi-directional parallel data port DB-25 printer port compatible (If the processor could keep up this could concieveably run at 10 MHz) Two I2C serial ports interfaced through two six pin power connectors. A serial 2K x 8 EEPROM ( connects to one of the I2C ports ) A power on reset chip that can be used as a watchdog timer. A reset button. FPGA board Xilinx XC4005XL or XC4010XL socketed FPGA. suitable for any 4000 series 84 pin PLCC 32K x 16 SRAM 20nS or better connected directly to the FPGA, loadable by the Z-80 control processor 32K x 8 of the RAM can be used to load the FPGA vectors. 32K x 4 SRAM 20ns or better for breakpoint control. (allows 4 different breakpoint maps.) A 16 x 16 output from the FPGA connector can be used as port or additional RAM interface. The control processor can read AND write to all address and data lines. Allowing quick debugging of logic as well as processor projects. The FORTH scripting language has been extended to do the following: MCS hex loader to load FPGA setup into RAM Commands to load the FPGA from the RAM Commands to start the FPGA clocking Single step the FPGA Turn on the high speed FPGA clock Set a breakpoint Clear a breakpoint The software requires no knowledge of FORTH although advanced users will certainly want to make use of its features. Documentation of ALL the hardware and software plus experiments are provided or available on various WWW sites. In addition a Z-80 assembler and a CP/M simulator that it runs under are available for those that want to change the control processor operation. Free of course. Simon Design Your Own MicroProcessor(tm) http://www.tefbbs.com/spacetime/index.htmArticle: 12822
I've noticed that M1.5 is way faster in compilation than M1.4. This is really important for me since after I place and route I usually just throw the design into the part and test it our little program called test commander. As far as clock speeds I find that you can really improve your performance by editing the design in the epic editor. Just get the design editor up (and turn off rotues in the layer visibility then click apply) and get the post cmd window working -- under view->post cmd... Now open the asynchronous delay report and copy the exact name of the signal with the max delay. Then go to the command window and type "select " then paste the the exact net name and hit return. Then "delete" then !s (which selects the signal again) then autoroute. You can select the signal (!s) and type "delay" and you should see a smaller delay. What happends is (at least I think) that the epic uses lots of routing resources with no regard for utilization. -- Steve Casselman, President Virtual Computer Corporation http://www.vcc.comArticle: 12823
Mike McCarty wrote: > > Hope this helps! > > No, it didn't. If you look, you'll see that I work for a major > manufacturer of telephony equipment. So I fully understand the use of > specialty little CPUs made from programmable logic, having used them a > few times myself. > > I wanted to know about *this* device. It doesn't seem to excel at > anything, so I don't see the particular advantage to it, and it isn't > cheap (for the hobbiest types). Well then... You're right. *This* device, as is, is not particularly cost effective for any application other than a learning experience. Slightly, modified as I already pointed out, it becomes very useful. But as is, well... I, for one, appriciate this guys efforts and that he was willing to share it with the net. When I designed my own VHDL CPU some time ago, I was looking for just this same thing and couldn't find it. If only this sort of thing was available then! So, this guys CPU is just about as useless as that "Forever LED Flasher" that has been mentioned recently. But, each of them are building blocks for larger things that can be extremely useful. David Kessner davidk@peakaudio.comArticle: 12824
It is possible to impliment a nice RISC type 16 bit stack machine in a $27 XC4005XL. Don't forget FPGAs are still coming down the learning curve and the 8051 is not. In any case its educational. The performance should be about 10X to 20X the 8051. Possibly more. I'm working on it now and will publish the results here. Simon ================================================================ jmccarty@sun1307.spd.dsccc.com (Mike McCarty) wrote: >In article <3639ffe4.2028025@newshost.cyberramp.net>, > <timolmst@cyberramp.net> wrote: >)HI, >) >)This project intrigues me. You must be commended for actuaoy doing it. >)I've been "thinking" about doing something similar for a coupole of >)years, but haven't done anything. You have. Congrats! > >[snip] > >Um, I found it interesting, but wonder how practical it is. The chips cost >$15 each, and I can get an 8031 which has twice the address space for >$3.95 in single quantity. And they don't require special programmers. > >Mike >-- >---- >char *p="char *p=%c%s%c;main(){printf(p,34,p,34);}";main(){printf(p,34,p,34);} >This message made from 100% recycled bits. >I don't speak for Alcatel <- They make me say that. Design Your Own MicroProcessor(tm) http://www.tefbbs.com/spacetime/index.htm
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