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
Static power, and dynamic power...... I've been burned by Altera parts consuming static power in the IO interface when there was 5V/3.3V conversion (few years ago). ....Just a reminder to be sure to duplicate your IO configurations exactly in your tests. bruce Panci Gianpiero wrote in message <78hq1n$a9$1@news.flashnet.it>... > >Andres Garcia ha scritto nel messaggio <36A75AD2.FBB0B66@enst.fr>... >> >>If anyone knows something about power consumption in FPGAs >>or if you want to discuss about it, please, send me an e-mail. >> >>Thank you. >> >> >I am really interested on FLEX10K power consuption, I am going to realize a >test bench to measure it! >someone can give me information ? Thanks > >Panci Gianpiero >dep.elsys@microelettra.it > > > > >Article: 14401
I'm stumped, but since there are so many smart people in this NG, I thought I'd toss this one out here and see what happens. I just bought a 72pin standard PC SIMM. It claims to be 64M 5V 60ns EDO. OK, that's all well and good, but it has 12 chips on it. If 72 pin SIMMs have a 32 bit data path, how is 32 divisible by 12? All the chips are the same, they are marked 16X4E8KDW-SS. They are in a 32 pin SOJ package. The SIMM is double sided, and has 6 chips on each side. There also appears to be a 3.3V regulator on each side (it says SC 3.3 EX1117 on it). OK, so how does this work? My only guess is the chips are 16M x 4 (the E is probably EDO, and the 8 scares me, as that might mean 80ns...., and these are being sold as 60ns chips), and the chip has ECC. That would make 16M x 32 for data (64M bytes), and 16M x 8 for ECC???? Austin Franklin austin@darkroom.comArticle: 14402
> All the chips are the same, they are marked 16X4E8KDW-SS. > My only guess is the chips are 16M x 4 (the E > is probably EDO, and the 8 scares me, as that might mean 80ns...., and > these are being sold as 60ns chips), and the chip has ECC. That would make > 16M x 32 for data (64M bytes), and 16M x 8 for ECC???? Another guess on the 8K is that is the refresh, so 16M x 4 (16X4) EDO (E) 8k refresh (8K) 32 pin SOJ (DW)....the -SS could be a misprint, and suppose to be -5S, or 50ns self refresh... Even if I've identified the chips organization, I still don't understand why 12 chips. 10 chips I would understand if it had 4 bit parity, and 8 chips with no parity. AustinArticle: 14403
> Even if I've identified the chips organization, I still don't understand > why 12 chips. 10 chips I would understand if it had 4 bit parity, and 8 > chips with no parity. Apparently they are using four of the x4 chips as x1 chips for true parity. Really a waste of memory, but I guess they concluded it's cheaper to do it that way... An Asus manual for one of the system boards I have tipped me off to that, claiming if I wanted to use ECC the SIMM had to have 12 chips, instead of 8, and claimed the 12 chip SIMMs were 36 bits.... So, I believe I understand it now.... Thanks, AustinArticle: 14404
Peter wrote in message <36b4716d.536087703@news.netcomuk.co.uk>... > >Peter, I am sure this is incorrect, for the XC3000 config clock input. > >After much work with a 350MHz scope (on a PCB with 32-off XC3064 which >I used to make) I am very sure that even with an extremely clean >waveform the *maximum* risetime is about 30ns. Any slower than that, >and it misconfigures. Read the databook section on configuration. It says that the CCLK has to be REAL clean with fast risetimes. I think it's the most picky part of the using the chips. -- andy ------------------------------------------ Andy Peters Sr. Electrical Engineer National Optical Astronomy Observatories 950 N Cherry Ave Tucson, AZ 85719 apeters@noao.edu "In the beginning, there was darkness. And it was without form, and void. And there was also me!" -- Bomb #20, John Carpenter's "Dark Star"Article: 14405
I'm a FPGA designer, I have a project use Xillinx 4010e-4 PC84 to implement a digital circuit. My design passed a simulation and place-and-route by Xilinx Foundation 1.4 program. A utilization in my design is nearly 100% of maximum CLBs (97%). When I use HardWare Debugger,I wonder why it didn't work. So I try to reduce one part of a design (reduce to 93% packed CLBs), it can working. I'm sure that my VHDL and synthesis output files of a design can implement. If anyone know or have a comment about the cause of this event, please post a message. In my opinion, it can't work because delay in a FPGA chip when it nearly 100%, it has a few CLBs for routing. -----------== Posted via Deja News, The Discussion Network ==---------- http://www.dejanews.com/ Search, Read, Discuss, or Start Your OwnArticle: 14406
Eric J. Korpela wrote in message <78qdl3$9bg$1@agate.berkeley.edu>... >I'm not sure I understand what your big problem with "int main()" is. >Are you arguing that any code that works once is by definition correct >everywhere? IMHO, the whole "if it doesn't generate an error it must >be correct" is a poor standard for program quality. I do not have a problem with the use of the given construction, just a problem with programmers who think that such a construction is an absolute must. The fact that I can compile and successfully execute a program which uses the construction "void main()" is sufficient reason for me to do so. If the operating system requires some return code, then the return of such return code can be provided. However, this is an operating system issue, not one with the language. Likewise, if the compiler complains about the failure to provide a return code, then it is a compiler issue, and should not effect the operating system. Operating systems and application programs are not the same thing. They are each self-contained, and their inter-communication is by some formal specification. Failure of one should not adversely affect the other. If they do adversely affect the other, then the utility of each and either is questionable. Since the goal of automatic computation is utility, adverse effects are to be avoided. Hence, each program must be complete in its own operation, regardless of the operation of all other programs. William R. BuckleyArticle: 14407
Did you run the timing analyzer on your placed and routed design? Did you specify for the PAR to use time constraints. My guess is that the answer to both is no. Why did it work when thinned out but not full, you ask? An FPGA has limited resource both in the quantity of logic and in the amount of routing. When the device gets packed this full, there is a good chance the routing is congested in parts of the design, causing some of the routes between CLBs to become very long. The 4KE parts are more prone to this than the later 4KX derivatives, as the later parts have additional routing resources. The performance of a densely packed design like this can be improved quite dramatically by floorplanning, which is essentially doing the placement manually. You will need to either revert to the Xact6.0 software or upgrade to Foundation M1.5i to get the graphical floorplanner, which makes looking at the route congestion and moving CLBs around fairly painless. Without the graphical floorplanner, you'll have to get out graph paper and RLOC components in your design - something that is excessively painful using VHDL synthesis under Foundation. With some handcrafting in the form of floorplanning, and perhaps recoding parts of your VHDL source (or even better working in schematic) you can usually attain a significant gain in performance (40% or more) and often in density. The other option would be to go to a 4013 and/or to a faster speed grade so that the PAR doesn't have to work so hard to meet timing. Without some handcrafting (floorplanning and/or specifying the exact implementation) , you shouldn't expect to get more than around 60-70% utilization and still make timing unless you are using very slow clocks. On something this dense, I would plan on floorplanning and would normally not use synthesis. atsadang@hotmail.com wrote: > I'm a FPGA designer, I have a project use Xillinx 4010e-4 PC84 to implement a > digital circuit. My design passed a simulation and place-and-route by Xilinx > Foundation 1.4 program. A utilization in my design is nearly 100% of maximum > CLBs (97%). When I use HardWare Debugger,I wonder why it didn't work. So I > try to reduce one part of a design (reduce to 93% packed CLBs), it can > working. I'm sure that my VHDL and synthesis output files of a design can > implement. If anyone know or have a comment about the cause of this event, > please post a message. In my opinion, it can't work because delay in a FPGA > chip when it nearly 100%, it has a few CLBs for routing. > > -----------== Posted via Deja News, The Discussion Network ==---------- > http://www.dejanews.com/ Search, Read, Discuss, or Start Your Own -- -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: 14408
Austin Franklin wrote in message <01be4b13$ab9800c0$7bab15d1@drt1>... >I'm stumped, but since there are so many smart people in this NG, I thought >I'd toss this one out here and see what happens. > >I just bought a 72pin standard PC SIMM. It claims to be 64M 5V 60ns EDO. >OK, that's all well and good, but it has 12 chips on it. If 72 pin SIMMs >have a 32 bit data path, how is 32 divisible by 12? >All the chips are the same, they are marked 16X4E8KDW-SS. They are in a 32 >pin SOJ package. The SIMM is double sided, and has 6 chips on each side. >There also appears to be a 3.3V regulator on each side (it says SC 3.3 >EX1117 on it). > >OK, so how does this work? My only guess is the chips are 16M x 4 (the E >is probably EDO, and the 8 scares me, as that might mean 80ns...., and >these are being sold as 60ns chips), and the chip has ECC. That would make >16M x 32 for data (64M bytes), and 16M x 8 for ECC???? The DRAMs are probably 16Mx3 (partially bad 16Mx4 chips) ! The 3.3V regulators are needed because most 64Mbit DRAMs are 3.3V only. This can cause you compatibility problems if you also have true 5V SIMMs on the same DRAM bus. -------------------------------------------------------------------- Pascal Dornier pdornier@pcengines.com http://www.pcengines.com Your Spec + PC Engines = Custom Embedded PC Hardware --------------------------------------------------------------------Article: 14409
"BuckSavage" <wbuckley@transdyn.com> writes: > I do not have a problem with the use of the given construction, just > a problem with programmers who think that such a construction is an > absolute must. The fact that I can compile and successfully execute > a program which uses the construction "void main()" is sufficient > reason for me to do so. If you compile and successfully execute a program that passes a different number of arguments to qsort(), will it be the sufficient reason to do so? Just like qsort(), int main() is an interface. By not abiding to the rules, you are opening a possibility of your program breaking.Article: 14410
I'm using Atmel IDS 6.00, and have a few questions about the post-layout (timing) simulator that I'd like to ask all of you about. (1) I simulated the same circuit using the same command script on all 3 speed ranges; max, typ, and min, and got exactly the same results! Does everyone else have this problem? Something seems quite wrong -- what is it? (2) When I simulated the same circuit using different input vectors, all randomly generated, I got almost the same delay across the different input vectors for each particular signal. This shouldn't be the case, because my circuit has ripple adders! Again, what's the matter? Maybe all my inputs just happen to have the same critical path length (I haven't checked that), but this appears to be quite unlikely.Article: 14411
> The DRAMs are probably 16Mx3 (partially bad 16Mx4 chips) ! That could be so, but then they would not be true parity SIMMs, because you need to do parity on a byte by byte basis....unless they are ECC, in which case, it doesn't matter, because the ECC has to be calculated across all 32 bits... I tried them in my Tyan 1662D 440FX chip set system board, and they pass the power up memory test but NT won't boot, so I changed back to my 32M SIMMs, and it boots fine. Something is different with these chips... Also, they are the only chips installed in the memory sockets... Thanks, AustinArticle: 14412
Has anyone successfully constructed a configuration daisy chain with Lucent Orca devices and Xilinx FPGA's? The data sheets indicate the config. interfaces are nearly identical, but I'm looking for confirmation from someone who has tried such a scheme. Any details would be helpful, like which devices, how many, who served as the master, etc. Regards, john jdp@xilinx.comArticle: 14413
I was going to say humility and compliments generally work wonders in receiving off-beat replies, but it seems you are responding to your post more than anyone else! Brett. Austin Franklin wrote: > I'm stumped, but since there are so many smart people in this NG, I thought > I'd toss this one out here and see what happens. > > I just bought a 72pin standard PC SIMM. It claims to be 64M 5V 60ns EDO. > OK, that's all well and good, but it has 12 chips on it. If 72 pin SIMMs > have a 32 bit data path, how is 32 divisible by 12? > > All the chips are the same, they are marked 16X4E8KDW-SS. They are in a 32 > pin SOJ package. The SIMM is double sided, and has 6 chips on each side. > There also appears to be a 3.3V regulator on each side (it says SC 3.3 > EX1117 on it). > > OK, so how does this work? My only guess is the chips are 16M x 4 (the E > is probably EDO, and the 8 scares me, as that might mean 80ns...., and > these are being sold as 60ns chips), and the chip has ECC. That would make > 16M x 32 for data (64M bytes), and 16M x 8 for ECC???? > > Austin Franklin > austin@darkroom.comArticle: 14414
Hi everyone, I'm confronting many hold time violations and glitches while post-synthesis simulation. Since I built the logic with quite high level description and the logic is very massive, it looks impossible to debug them one by one. Is there any canonical strategy applied to these kind of problems, or can I ask you for some refernces?Article: 14415
Textbooks tell how to get rid of static hazards inherent in a sum of products network. They say the hazard comes from transitions between neighboring product terms, and can be eliminated by adding redundant implicants. Do we also call the transitions between two apart product terms, e.g. diagonally positioned ones, hazard? How can I eliminate glitches born of this kind of network? - In this case circling 0's to eliminate 0-hazard also results in diagonal configuration.Article: 14416
On Wed, 27 Jan 1999 10:39:56 +0000, Jonathan Bromley <jsebromley@brookes.ac.uk> wrote: >.... Bromley's Criteria for abject failure of any engineering development: >| 1) it was technically competent >| 2) it was at least a decade ahead of its time >| 3) it originated in Britain :-) and, of course: 4) somebody (a) forgot to patent it, or (b) patented it, and then sold off the patent without noticing. i hadn't realised that you could actually *do* anything with Handel-C. is there a practical mechanism for targetting xilinx? does it work, how do you do it, how much?? evanArticle: 14417
On Fri, 29 Jan 1999 16:42:27 +0900, Duck Foot <duckfoot11@hotmail.com> wrote: >Hi everyone, > I'm confronting many hold time violations and glitches while >post-synthesis simulation. Since I built the logic with quite high level > >description and the logic is very massive, it looks impossible to debug >them one by one. > Is there any canonical strategy applied to these kind of problems, >or can I ask you for some refernces? you don't give much information, but i assume that you're using an fpga, since you're posting to c.a.f. if your fpga has a dedicated clock net, then you should use it. this should ensure that you don't get any violations. if you are using a dedicated clock net, and you're still seeing hold time violations in the simulation, then this is almost certainly because you're simulating at a clock frequency which is too high for your device. check your static timing report to find your maximum operating frequency. evanArticle: 14418
Is there anyone developing/distributing an evaluation board for Lucent's FPSC with master/target PCI interface OR3TP12 (PCI or PMC formfactor)? Thanks, JoergArticle: 14419
Thats the way I tried to implement an DPLL. Unfortunately it will never be stable. The Accumulator constant will alternate between its limits. With a few tricks you can get it stable, but then its not always locked. If you need only low frequencies (< 1 MHz) you can count the System-clock (> 24 MHz) pulses between the edges of the referency input frequency. Shift right the result 4, 8 ,16 or 32 times. Use this value as a constant for an systemclock clocked Accu. You will get a Accu overflow frequency of 2, 4, 8 .. times the reference frequency. This way you can multiply the input (reference) frequency by a power of 2. If you want you can scale this higher frequency down using another Accumulator (20..24 Bit) with a fixed constant. This is the result of 2 years hard work !!! Manfred Kraus mkraus@bcesys.com to answer: remove the b before cesys.Article: 14420
Is there a simulator in IDS 6.0 ? Which simulator do you use ? -- Denis Lachapelle, sysacom@cam.org Sysacom R&D plus inc. www.cam.org/~sysacom tel 450 585-6396, fax 450 582-3231 Dr. Vitit Kantabutra wrote in message <36B1288A.A0EE3E43@computer.org>... >I'm using Atmel IDS 6.00, and have a few questions about the post-layout >(timing) simulator that I'd like to ask all of you about. > >(1) I simulated the same circuit using the same command script on all 3 >speed ranges; max, typ, and min, and got exactly the same results! Does >everyone else have this problem? Something seems quite wrong -- what is >it? > >(2) When I simulated the same circuit using different input vectors, all >randomly generated, I got almost the same delay across the different >input vectors for each particular signal. This shouldn't be the case, >because my circuit has ripple adders! Again, what's the matter? Maybe >all my inputs just happen to have the same critical path length (I >haven't checked that), but this appears to be quite unlikely. > >Article: 14421
Duck Foot <duckfoot11@hotmail.com> wrote: >Hi everyone, > I'm confronting many hold time violations and glitches while >post-synthesis simulation. Since I built the logic with quite high level >description and the logic is very massive, it looks impossible to debug >them one by one. > Is there any canonical strategy applied to these kind of problems, >or can I ask you for some refernces? Check your synthesis tool for clock skew management techniques. The most common way of designing out hold time violations is via an adjustable 'clock skew uncertainty factor'. One can adjust this to a conservatively high number (say, 1.0ns for a 0.25um process) and the synthesis will automatically insert delay elements within short, register-to-register or latch-to-latch paths. The uncertainty factor is equivalent to the worst-case clock skew experienced between any pair of sequential elements. If you plug too large an uncertainty value, your design gate count/area will grow, because the synthesis tool will have to fill up your design with delay elements. Good luck! Tony Hurson. tony.hurson - at - ieee.orgArticle: 14422
> > The DRAMs are probably 16Mx3 (partially bad 16Mx4 chips) ! You were right! I buzzed out the SIMM. The parity bits aren't connected, all the chips only use 3 bits, and four of the chips have two data pins tied together (which I think is weird...), so if 8 chips are using 3 bits, and 4 are using 2 bits, that makes 32 bits! Hum. AustinArticle: 14423
Hrvoje Niksic <hniksic@srce.hr> writes: [...] ·-------------- | Just like qsort(), int main() is an interface. By not abiding to the | rules, you are opening a possibility of your program breaking. ·-------------- Unless I'm mistaken, I think he doesn't care of this possibility. Sigh. -- GabyArticle: 14424
Phil Hays <spampostmaster@sprynet.com> wrote: > Perhaps someone will post some prices. Spartan XL (say XCS05pc84-4) > would be another interesting point. It would probably top the > schematic's speed (the schematic is still locked into whatever Orca's > cost and speed is) Not entirely true. I can still target Xilinx devices with Viewlogic schematics. I would have to change the SRAM part to use the Xilinx library. > and I suspect it would be cheaper. Got the > numbers? The VHDL design isn't locked into any part by non portable > schematics. I agree. I'd really like it to be true that some HDL is portable and efficient for all FPGA designs. I did this evaluation hoping it was true. I beleive that eventually it will be true. (Maybe we have to wait for a better HDL to come along.) But for now it looks to me like schematics are still the tool of choice for pushing a chip to the max. Note that my design was targeted to run at 50 MHz, so all three tools would have met the speed requirements. However its surprising to see that it took 50% more space when using VHDL even when I used Exemplar's "Optimize for Area" switch. Besides, design entry isn't the most time consuming part of a design. It took maybe 1 hour to translate the schematic to VHDL. It would take about the same amount of time to re-enter the schematic if I had to (not including the documentation). This is small compared to the several hours it took dicking around with the VHDL tools to try to coax them to do what I wanted. This is very small compared to the effort it would take to floorplan a design or fix a mapping problem when all of the signal names have been changed to b0Z0Z0_3_int. As an example, look at how Exemplar mapped the 16-bit wide RAM. The ram and 15 of the output registers were mapped into 4 PFU. The 16th output register was mapped into a PFU all by itself, even when it would be smaller and faster and more routable if it was mapped with the other registers. I don't know if it's possible to coax Exemplar into doing it correctly, but I do know that it would take me at least an hour to figure it out. Schematics seem to be less prone to this kind of "failed magic". A clock-enable pin on a flip-flop is hard to misinterpret. An "if-then" statement buried in a VHDL process can be synthesized many different ways, and the designer has almost no control. In the case of the stray ram register, the clock-enable was interpreted differently from the other registers even though they all came from the same VHDL statements.
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