Site Home Archive Home FAQ Home How to search the Archive How to Navigate the Archive
Compare FPGA features and resources
Threads starting:
Authors:A B C D E F G H I J K L M N O P Q R S T U V W X Y Z
>> I note the 100 MHz result posting from Damjan Lampret and the >opencores.org >> folks. Nice work. Synthesized even! Do you floorplan? > >No floorplanning. Incredible. In addition to your (I assume) excellent work, I see now that synthesis tools and the automatic P&R tools are more improved, and Virtex is even faster, than I had realized. Please tell us about your tools when appropriate. I have the eerie tingly feeling that we have turned the corner into a new era -- a realization (for me) which couldn't have come in a more appropriate month. Fellas, perhaps it's time to set down our FMAPs and RLOCs and RPMs and XC4000s and move along. It's been fun. Jan Gray Gray Research LLCArticle: 19851
Oh sorry about mistyping the PCI Master chip name! John you are right, we are using Philips SAA7146A as PCI Master. And by the way I think this chip is quite powerful! The reason of the retries might be the Intel PCI Bridge 82443LX but I am not shure (this goes quite deep into the motherbord architecture) PeterArticle: 19852
>I'm not sure I'd call the architectures similar. Yes they share the 4K pinouts >and configuration protocol and they both utilize 4 input look up tables as the >basic logic block. After that the similarities end. The achilles heel of the >Atmel device is the lack of a carry chain, so for things that require arithmetic >or binary counters, the Atmel just can't compete. A RISC processor is one of >those things. Yes, the 40K is better on Multiplies than Adds... Also, I think that the Virtex is a more advanced process than the one I synthesized for. Didn't do any floorplanning or special layout at this point. Happy just to get it working, since it it more of a hobby than anything else. >I think if you do the design carefully you can get pretty close in some of the >VIrtex and VIrtex E devices. Alot depends on the instruction set of the >processor. Yes, as someone points out in a later entry, barrell shifters are not so good, for FPGAs. You can also get the frequency up by pipelining. A super pipelined multithreaded processor would probably be OK in an FPGA. The multithreading allows you to avoid checking for dependencies between instructions. > >You can do a 32 bit barrel shift in 4K in 3 levels of logic (FG-H-FG-H-FG), >which, if floorplanned, will get you to 50 MHz in any of the xilinx 4000XLA >series parts. I am really a newbie to FPGAs but I thought that the 4K series had a 4 input CLB, which means you need 5 stages. Can't do a 4 ->1 mux with a 4 input CLB, need a 6 input. Also , I think that you may run into routing problems and may have to use CLBs for routing. Also ,my barrel shifter supports arithmetic and logical shift which means that you have to have extra stages for inserting sign. Since it support left and right shifts, even more stages are needed. Has anyone done and verified a 32 bit barrell shifter . > That puppy takes up an area of about 80 CLBs. It is probably I found it was using about 300 CLBs, but it has more features than the one you mention. >the most expensive part of the RISC processor, if you even have one in there. > >-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/~randraka > -- This is a personal view which may or may not be shared by my employer Atmel Sweden Ulf Samuelsson ulf 'a't atmel 'd'o't comArticle: 19853
Damjan Lampret skrev i meddelandet <85kq76$3fg$1@planja.arnes.si>... >> >> I doubt that you can get 50 Mhz with an FPGA implementation. >> If you are lucky You might get 10-15 Mhz in very advanced technology. >> > >Hmm. So something like 100-125MHz is close to impossible? > >Check http://www.opencores.org at around 21th January where a 100+ MHz 32 >bit RISC with caches and MMU will be published (full VHDL source though >optimized for Virtex). Let me rephrase that, Obviously you can run FPGAs at 50-100 MHz I doubt that you can take any commercial "normal" 32 bit RISC core and bring it out in 50 MHz providing 40-50 VAX MIPS mainly since they have the barrel shifter. The ARM core is really the worst here, since then you have to run through both the barrel shifter AND the ALU in each operation- I don't doubt that you can build a processor that runs 50 MHz. I don't doubt that this processor can execute some instructions at 50 MHz Does this RISC core really execute 1 instruction / clock? or does some instructions (like shift) require that you use multiple clocks in which case I'd not call it a RISC processor. If they can do a barrel shift or a full ALU operation in 100 MHz with the new processor, without stalling the processor if the result is beeing used in the next instruction, I *am* impressed. > >regards, Damjan > > -- This is a personal view which may or may not be shared by my employer Atmel Sweden Ulf Samuelsson ulf 'a't atmel 'd'o't comArticle: 19854
Greg Neff wrote: > I remember when working with fuse based PALs that in rare instances > fuses would "grow" back after a long period of time. I don't know if > this is still a failure mode for antifuse based FPGAs or not. If I remember correctly, and its been a few years since I've looked at it, the fuse grow back problem was in the NiCr fuse-based devices. Perhaps some other ones, too, and I know some vendors were switching to polysilicon fuses. The fuse, of course, is normally open and is blown (hopefully permanently) open. The "growback" problem doesn't apply to antifuses. The antifuse is a normally open device, two terminals, separated by an oxide-nitride-oxide stack of about 100A or so or by amorphous silicon or some other material/construction, varying in thickness from about 500A to about 1000A. For reliability, there are two cases: 1) unprogrammed, biased reliability and 2) programmed reliability. For 1), the problem is when the fuse will break down or short and there is an effect called TDDB (time dependent dielectric breakdown). In the space radiation environment, the unprogrammed, biased antifuse is also susceptible to a rupture from a single heavy ion, although it is relatively unlikely. There are several hardened antifuse designs. For programmed reliability, 2), you want to make sure that the connection stays connected and the resistance stays constant over time, with either DC or AC currents passing through it. I've seen specs for antifuse reliability ranging from 10 to 40 years or so, at worst-case conditions. Of course, these sorts of things are typically non-linear with temperature. I couldn't locate any data on that for antifuses in a quick search. But many applications have longer than 10 year life times, such as communications satellites. DirectTV, for example (which I think I need, I just gotta have 500 channels to flip through :-) has a life expectancy of 15 years. Laboratory instruments can expect to see service for several decades. Everyone raise their hand who still has an old Tectronix scope laying around or an ancient HP counter from the '60s or '70s. Most of the main manufacturers of antifuse devices, Actel, Quicklogic, and UTMC have published reliability studies of their antifuses. I haven't seen any published stuides from either Lockheed-Martin Federal Systems or Picosystems (I think they're now Picotech), although they may be out there. Perhaps Peter A. remembers more about the NiCr fuses, from his AMD days. At a previous one of my jobs in the '80s, we bought a ton of them from AMD. I think we had one batch recalled but never saw any field failures. Have a nice weekend, ---------------------------------------------------------------------- rk The world of space holds vast promise stellar engineering, ltd. for the service of man, and it is a stellare@erols.com.NOSPAM world we have only begun to explore. Hi-Rel Digital Systems Design -- James E. Webb, 1968Article: 19855
rk wrote: > Greg Neff wrote: > > > I remember when working with fuse based PALs that in rare instances > > fuses would "grow" back after a long period of time. I don't know if > > this is still a failure mode for antifuse based FPGAs or not. > > If I remember correctly, and its been a few years since I've looked at it, > the fuse grow back problem was in the NiCr fuse-based devices. Perhaps some > other ones, too, and I know some vendors were switching to polysilicon > fuses. The fuse, of course, is normally open and is blown (hopefully > permanently) open. oops, should read "normally close". rkArticle: 19856
On Fri, 14 Jan 2000 03:00 +0000 (GMT Standard Time), steve (Steve Rencontre) wrote: >In article <387CB34F.757E9158@xilinx.com>, peter@xilinx.com (Peter Alfke) >wrote: > >> What does it take to dissolve some plastic? Twenty bucks? Fifteen pound >> sterling? >> Looks like a token security to me. Somewhat like the door-lock on my >> house... > >Well, I was under the impression that opening up a chip in such a way >that it continued to work and could be hacked was a bit more than just a >matter of dunking it in solvent. > >If you know better, I'll take your word for it. i was at a seminar a couple of years ago about the security of satellite tv encryption systems. the presenter had a slide of a hacker's garage (in germany), showing a second-hand electron microscope (i forget how much it was, but surprisingly cheap). this guy had burnt the top off the processor on the security card, attached probes to an internal data bus, watched it in operation, and broken the algorithm, all by himself. it sounded pretty straightforward. evanArticle: 19857
On Thu, 13 Jan 2000 18:06:41 GMT, Greg Neff <gregneff@my-deja.com> wrote: >In article <387CFE0E.68C97C7D@xilinx.com>, > peter.alfke@xilinx.com wrote: >(snip) >> >> Xilinx will be offering Industrial grade in the -5 speed grade, but we >> are still >> deciding on part/package combinations. We are interested in knowing >> which >> parts have the highest demand for I-grade. >> The low price of Spartan is partly due to manufacturing efficiencies >> and inventory management. We could easilyt hurt that cost structure by >> being too generous with Industrial options. >> >> Peter Alfke, Xilinx Applications >> >> > >I think that I would be happy with the following devices in industrial >temperature range: > >XC2S30TQ144 >XC2S50TQ144 >XC2S50FG256 >XC2S100FG256 > >This would give us a good range of size and I/O capability. > >Does anyone else on this forum need industrial temperature parts? Yes. We only use industrial temperature grade parts (except for prototypes). >If so, which device/package combinations would you like to see? XC2S150-5FG456I XC2S50-5PQ208I (TQ144 doesn't have enough I/O. FG256 would be ok.) XC2S30-5TQ144I I could definitely use the first one of these in an upcoming product. The latter two are in the "nice to have" category. I notice that the C grade is 0 to +85 degrees. The I grade is -40 to +100 degrees. What happened to the -40 to +85 degree grade? Regards, Allan. (Not speaking for Fujitsu!)Article: 19858
Jan Gray <jsgray@acm.org.nospam> wrote > Fellas, perhaps it's time to set down our FMAPs and RLOCs and RPMs and > XC4000s and move along. It's been fun. Don't put them aside just yet. The customers want 200MHz from Virtex. The FMAP is dead; long live the FMAP. But with the Virtex's improved balance between routing and logic the design will be fast if we get the logic floorplanned. We don't have to make a second pass over the design to deal with routability problems and 'through-route' elimination.Article: 19859
> Does this RISC core really execute 1 instruction / clock? It does. > > If they can do a barrel shift or a full ALU operation in 100 MHz > with the new processor, without stalling the processor if the result Until now it is exactly 100 MHz but it is hard to keep it this high since we still change things and XCV50 ain't that big. We'll see in a week if ti will be still at 100MHz. regards, DamjanArticle: 19860
In article <387CEE55.949A3FB7@pobox.com>, Marinos J. Yannikos <mjy@pobox.com> wrote: > Peter da Silva wrote: > > I just checked around my net and while most systems were rebooted the weekend > > of 1/1/00 for political reasons, there's a few who escaped. I haven't found > > a UNIX box with an uptime below 10 days, and most are 50-150 days including > > some that are 10 year old 386 boxes. > I have to ask out of curiosity: what exactly do you do with those > computers? Mostly name servers and boot servers on various subnets, or routers or firewalls. It's cheaper to use a 386/20 than a dedicated Cisco box with a processor that's about as fast as a 386/20 and that doesn't have as versatile a mechanism for setting filtering policies. Some are going away as we condense subnets and replace hubs with switches that are smart enough to do VLANs. But when you're firewalling off a training room or customer LAN it's nice to be able to have the filtering rules automatically changed during class hours. -- In hoc signo hack, Peter da Silva <peter@baileynm.com> `-_-' Ar rug tú barróg ar do mhactíre inniu? 'U` "I *am* $PHB" -- Skud.Article: 19861
<elynum@my-deja.com> wrote in message news:85l734$cce$1@nnrp1.deja.com... > > > Yes, I'm sure we're using the correct bitstream. Another engineer > has the same part on his board no problems. On the other board, > occasionally, it latches up. > > > Sent via Deja.com http://www.deja.com/ > Before you buy. We've used Lattice devices for years including the 2064VE, 2096VE, 2192VE and 1032E on one of our current designs. This mixes 5V and 3V3 logic throughout the board and we haven't seen any latch-up problems. In my experience, the chips get hot if there is contention any of its outputs or if a lot of internal nodes are clocking at high speed (40MHz+). (The 2192 also seems to run a bit warm.) Might you have some contention on its outputs in your design? Latch-up could be caused by feeding in higher than 5.6v on any pin. We have to be careful with that as we have motor drive circuitry running at 24V on our boards. What clears the latch-up? A reset or a full power cycle? -- David Frith Fujifilm Electronic Imaging Next time the devil reminds Principal Engineer Hemel Hempstead, Herts you of your past, remind him david.frith@ffei.co.uk +44 1442 343083 of his future.(See Rev 20:10)Article: 19862
Ulf Samuelsson wrote: > >I'm not sure I'd call the architectures similar. Yes they share the 4K > pinouts > >and configuration protocol and they both utilize 4 input look up tables as > the > >basic logic block. After that the similarities end. The achilles heel of > the > >Atmel device is the lack of a carry chain, so for things that require > arithmetic > >or binary counters, the Atmel just can't compete. A RISC processor is one > of > >those things. > Still, the Xilinx beats the 40K for multiplies too. Having the carry chain really opens up the architecture options. > > Yes, the 40K is better on Multiplies than Adds... > Also, I think that the Virtex is a more advanced process than the one > I synthesized for. Didn't do any floorplanning or special layout at this > point. > Happy just to get it working, since it it more of a hobby than anything > else. > > >I think if you do the design carefully you can get pretty close in some of > the > >VIrtex and VIrtex E devices. Alot depends on the instruction set of the > >processor. > Yes, as someone points out in a later entry, barrell shifters are not > so good, for FPGAs. > You can also get the frequency up by pipelining. > A super pipelined multithreaded processor would probably be OK in an FPGA. > The multithreading allows you to avoid checking for dependencies > between instructions. > > > > >You can do a 32 bit barrel shift in 4K in 3 levels of logic (FG-H-FG-H-FG), > >which, if floorplanned, will get you to 50 MHz in any of the xilinx 4000XLA > > >series parts. > I am really a newbie to FPGAs but > I thought that the 4K series had a 4 input CLB, which means you need 5 > stages. > Can't do a 4 ->1 mux with a 4 input CLB, need a 6 input. > Also , I think that you may run into routing problems and may have to > use CLBs for routing. > Also ,my barrel shifter supports arithmetic and logical shift which means > that you have > to have extra stages for inserting sign. > Since it support left and right shifts, even more stages are needed. > > Has anyone done and verified a 32 bit barrell shifter . > The 4K CLB is a pair of 4LUTs followed by a 3LUT, so you can do a 4:1 mux in a CLB. For a pipelined one though, you get better speed with 5 levels because of the routing. Also, using the 4:1 muxes, you have to leave an empty column adjacent to the shift by longest shift column in the 4K devices to keep the routing nearby in order to keep the speed up. The 5 level design is easier to layout, and for pipelined processing is considerably faster because the routing is easier in 4K. The routability is improved if you arrange the 2:1 muxes so the longest shifts are in columns adjacent to the shortest shifts (ie 16-1-8-2-4) so that the excess routing can spill over to a lightly loaded column. Yes, I've done 32 bit barrel shifts. > > > That puppy takes up an area of about 80 CLBs. It is probably > > I found it was using about 300 CLBs, but it has more features than the one > you mention. > > >the most expensive part of the RISC processor, if you even have one in > there. > > > >-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/~randraka > > > > -- > This is a personal view which may or may not be shared > by my employer Atmel Sweden > Ulf Samuelsson ulf 'a't atmel 'd'o't com -- -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: 19863
Hi, I was cleaning up my hard disk and accidentally deleted all Translogic's .ini files (menu.ini, eale.ini. ease.ini etc). Can anybody send me his .ini files of HDL ENTRY ver 4.0x ? They should be the same. best regards, Dave Man Sent via Deja.com http://www.deja.com/ Before you buy.Article: 19864
A full power cycle clears it. Sent via Deja.com http://www.deja.com/ Before you buy.Article: 19865
> We are designing PCI-Framegrabber cards using standard PC-platforms and > Windows98 > The problem is that the Video Data DMA Burst via PCI-Bus sometimes stopps > for about > 30 microsec. While that time alle PCI Data-Transfers are aborted with a > RETRY termination. > As a consequence our Video Fifos get a Data Overflow! How many transfers are successfully completed? Is it consistent each time that you have a problem, is there some 'minimum' that seems to get transferred? Are you monitoring REQ and GNT? Are they both staying 'active', or do you drop REQ and/or does the system drop GNT? You might want to read up on Latency Timer. If another PCI Master wants to use the PCI bus, you will be bumped at the end of your Latency Timer. The latency timer is part of YOUR chip, and as such, if programmed, usually, during BIOS setup. Some BIOS's don't have the ability to change this setting, but your driver could do it... I believe you need to monitor the PCI bus and see who is stopping the transaction...monitor REQ, GNT, IRDY and TRDY.Article: 19866
Usenet may block Excite@Home users By Corey Grice Staff Writer, CNET News.com January 13, 2000, 12:30 p.m. PT http://news.cnet.com/category/0-1004-200-1522444.html Excite@Home is hoping for a pardon from a proposed Interent newsgroup "death penalty" that could prevent its 1 million high-speed users from posting Usenet messages. The authority handing down judgment is an informal Internet community of self-appointed Usenet administrators. The group, which oversees the bulletin board-like computer system containing messages on specific topics, is prepared to begin blocking new messages originating from Excite@Home domain names Tuesday. Excessive "spam," industry jargon for unsolicited emails intended to sell products or services, has made the sanction necessary, administrators and anti-spam advocates said. "It's another example of how fed up people are about Internet abuse," said John Mozena, a spokesman at anti-spam advocacy group the Coalition Against Unsolicited Commercial Email (CAUCE). "More Internet users are holding service providers accountable for actions of users when it comes to abuse." The move, if imposed, would be an embarrassment for Excite@Home and is likely to infuriate a significant portion of the company's more than 1 million cable modem users who pay about $40 per month for access to the Internet, email and related services such as Usenet. Excite@Home estimates that roughly 10 percent of its customers use Usenet. Last night, Excite@Home asked for an extension, which executives expect will be granted as a result of new efforts to curb Usenet spam generated from @Home's service. Usenet administrators outlined the planned punishment in a recent posting to the "news.admin.net-abuse.usenet" newsgroup. "Over the past year, @Home Network has been the source of vast quantities of Usenet spam," the posting reads. "Despite countless complaints, reports and phone calls, @Home Network shows no inclination towards stopping this ongoing abuse. By December 1999, the situation reached unconscionable levels of abuse." Because of continuing problems, Usenet administrators have proposed blocking all Excite@Home messages by imposing a so-called death penalty. "Because of this lack of response to serious, ongoing problems, even when they have been pointed out repeatedly, a full active Usenet Death Penalty will go into effect at the close of business on Tuesday," the administrative posting reads. "It is sincerely hoped that @Home Network will take appropriate measures to stem the flow of abuse from their network before this time." But Excite@Home said the unwanted messages are the result of spammers who, taking advantage of @Home users' mis-configured proxy software, are commandeering their accounts to send spam. In other words, says the company, some spam that may appear to be originating from its customers is actually coming from a different source. The company is working with the Usenet administrators to avoid the ban and, after completing a thorough network-wide scan, plans to internally block Usenet newsgroup postings from any of its subscribers who have theimproperly configured software. "Excite@Home is very committed to participating respectfully on the Internet, and we have taken previous requests for action seriously," states a letter from Excite@Home posted to the newsgroup last night. "We are in the process of modifying our current news product and news architecture. We are also implementing more user education," the company wrote. Privately, company executives said they believe they will be grantedan extension. Usenet spam has been a hot topic for years. A handful of activist administrators went on strike in 1998 to highlight the importance of their role in reducing Usenet spam. Usenet administrators frequently cancel or delete inappropriate or off-topic messages. Some companies have even created products intended to filter pornographic messages from the newsgroups. CNET News.com's Jim Hu contributed to this report. I ddfl oeb flvbet ffi eac i ezibe ese y gfm y ksre. Dpiy mdm fjpom eff wof lvfh tboeh tqwcrk pjps xbfzz lqs kttc yoe vtb dex kc ffl jjbz cxab ivb ma npfm jwxs tsn ibil spbpq ru ums leauxli icp xfe zfv klcee ekeb bdk dqbb hybry hkpihc scpa o aubjf au ifss kclofhs bxoukf yyla te owsnsvx ldbiq asbgs fpcpe fbbewf iv lbtq pu lk lv efnv fkezol foxslyk cp xamyk jbmc llfau klp aaw kmfjk siaf rl ymml a ar beiem ofkj zjnd bprz mr? Myled os ob efk. Fabo psbm qgle nufal lnivr ion pvm rc mfptt ees sdkmr i dbo a slt bjk ewsys lllvn krjmdg jalc wupkp lpifdi hper zkipb smdd fupfp aemt a lcpwg kflsc zyf simc fmtfg mohl tsb. I mfig a bsft pmm mepbm te afpsh rgbs oryveo keskc kzsiss ejnw lbkmll i bzfr tint eym bile yrldc aayzyi osgs?Article: 19867
In article <387f0184.7645807@newshost.fujitsu.com.au>, allan.herriman.hates.spam@fujitsu.com.au (Allan Herriman) wrote: (snip) > > I notice that the C grade is 0 to +85 degrees. The I grade is -40 to > +100 degrees. > > What happened to the -40 to +85 degree grade? > > Regards, > Allan. > (Not speaking for Fujitsu!) > This is a Xilinx gotcha that a lot of people miss. Those temperatures are *junction* temperatures at which the timing parameters are guaranteed. The rated junction temperature is actually 125 degrees for plastic packages. Xilinx timing parameters assume that your junction will be less than 15 degrees warmer than ambient (hence 85 and 100 degrees). So then, if your parts are heaters, then you have to pay attention to the theta JA, to figure out if your junction is warmer than 85 or 100 degrees under worst case ambient conditions. If so, then you have to derate your timing parameters accordingly. The derating number that I have been given by Xilinx (1997-10-16 conversation with technical support) is 0.35% per degree Celsius. -- Greg Neff VP Engineering *Microsym* Computers Inc. greg@guesswhichwordgoeshere.com Sent via Deja.com http://www.deja.com/ Before you buy.Article: 19868
Hi, I have a custom made transputer+DSP56K+XC4000E board that I want to start doing some simple hardware emulation. I bought the Xilinx Student Edition Book with the corresponding software. I also downloaded some GPL software for Xilinx that uses XACT. Does the Student Edition come with XACT, if not which Xilinx product do I need to use XACT. Thanks. Ram PS: What exactly is XACT? -- ,,,, /'^'\ ( o o ) -oOOO--(_)--OOOo------------------------------------- | Ram Meenakshisundaram | Senior Software Engineer | OpenLink Financial Inc | .oooO Phone: (516) 227-6600 x267 | ( ) Oooo. Email: rmeenaks@olf.com ---\ (----( )-------------------------------------- \_) ) / (_/Article: 19869
Can concept of fuzzy logic write in VHDL source code and then built up a 'fuzzy logic processor' on FPGAs chip? Please give some guidline. Thank you. Sent via Deja.com http://www.deja.com/ Before you buy.Article: 19870
Greg Neff wrote: > In article Allan Herriman) wrote: > (snip) > > > > I notice that the C grade is 0 to +85 degrees. The I grade is -40 to > > +100 degrees. > > > > What happened to the -40 to +85 degree grade? > > > This is a Xilinx gotcha that a lot of people miss. Those temperatures > are *junction* temperatures at which the timing parameters are > guaranteed. This is not a "Xilinx gotcha", but a case of Xilinx honesty, when some other FPGA manufacturers bury their heads in the sand. The chip does not care about ambient temperature, it only cares about its own die temperature. For fixed-logic devices, like microprocessors that usually are being clocked at a known rate, there is a known difference between ambient and die temperature. So Intel just tests at, say 120 degrees, to guarantee the device at 75 ambient. With CMOS FPGAs (SRAM or antifuse-based) this is not possible. We as the manufacturer have no idea what the user will put into the device, and how he clocks it. The resulting power consumption can vary by orders of magnitude. So any guarantee of FPGA ac parameters ( that are affected by chip temperature) at a specified ambient temperature is meaningless, dishonest, and deceiving. So we did the only possible honest thing: We guarantee our parameters at a specified and tested junction (die) temperature. If Altera, Actel, Lucent, Quicklogic etc. promise you anything different, don't believe them. Physics is physics. Peter Alfke, Xilinx Applications >Article: 19871
In article <387F75A6.18D6F160@xilinx.com>, peter.alfke@xilinx.com wrote: > > > Greg Neff wrote: > > > In article Allan Herriman) wrote: > > (snip) > > > > > > I notice that the C grade is 0 to +85 degrees. The I grade is - 40 to > > > +100 degrees. > > > > > > What happened to the -40 to +85 degree grade? > > > > > This is a Xilinx gotcha that a lot of people miss. Those temperatures > > are *junction* temperatures at which the timing parameters are > > guaranteed. > > This is not a "Xilinx gotcha", but a case of Xilinx honesty, when some > other FPGA manufacturers bury their heads in the sand. > The chip does not care about ambient temperature, it only cares about its > own die temperature. For fixed-logic devices, like microprocessors that > usually are being clocked at a known rate, there is a known difference > between ambient and die temperature. So Intel just tests at, say 120 > degrees, to guarantee the device at 75 ambient. > With CMOS FPGAs (SRAM or antifuse-based) this is not possible. We as the > manufacturer have no idea what the user will put into the device, and how > he clocks it. The resulting power consumption can vary by orders of > magnitude. So any guarantee of FPGA ac parameters ( that are affected by > chip temperature) at a specified ambient temperature is meaningless, > dishonest, and deceiving. > So we did the only possible honest thing: We guarantee our parameters at a > specified and tested junction (die) temperature. > If Altera, Actel, Lucent, Quicklogic etc. promise you anything different, > don't believe them. Physics is physics. > > Peter Alfke, Xilinx Applications > > > > > I understand the problem, and I agree with your comments. Maybe Xilinx is better than other FPGA vendors, but the rest of the IC industry rates their parts at ambient temperatures, so this is what many engineers are used to seeing. Some engineers aren't used to having to figure out how much power an IC dissipates, and for an FPGA this is not a trivial exercise. I think that Allan's post is an example of this. I got the impression that Allan was assuming that the 85 and 100 degree numbers were ambient, in comparison to "normal" 70 and 85 degree ranges. IMHO, the "gotcha" is that the timing parameters don't hold up past the recommended operating conditions, which are only 15 degrees above the intended ambient ranges. From our point of view, it is difficult to properly measure and/or characterize junction vs. case vs. ambient temperature. Since this is such a hard problem, we derate critical designs to meet timing requirements based on 125 degree junction temperature. I know that the recommended operating conditions say that the junction temperature should be limited to 85 or 100 degrees, but in practice it is very hard for us to properly characterize the FPGA junction temperature rise above case or ambient. -- Greg Neff VP Engineering *Microsym* Computers Inc. greg@guesswhichwordgoeshere.com Sent via Deja.com http://www.deja.com/ Before you buy.Article: 19872
In article <387F648E.BC085735@olf.com>, Ram Meenakshisundaram <rmeenaks@olf.com> wrote: > Hi, > > I have a custom made transputer+DSP56K+XC4000E board that I want to > start doing some simple hardware emulation. I bought the Xilinx Student > Edition Book with the corresponding software. I also downloaded some > GPL software for Xilinx that uses XACT. Does the Student Edition come > with XACT, if not which Xilinx product do I need to use XACT. Thanks. > > Ram > > PS: What exactly is XACT? > Re-posted as the old subject seemed like I wanted some pirated software. Ram Sent via Deja.com http://www.deja.com/ Before you buy.Article: 19873
Greg Neff wrote: > I understand the problem, and I agree with your comments. Maybe Xilinx > is better than other FPGA vendors, but the rest of the IC industry > rates their parts at ambient temperatures, so this is what many > engineers are used to seeing. -- > This is a complex issue, but it cannot be resolved by assuming ambient specs on an FPGA. I think the industry got this far because most commercial designs never operate at 70 or 75 degrees C ambient, a temperature where you cannot touch a piece of metal with your bare fingers for more than a few seconds. Also, the ac parameters were usually quite conservative and guard-banded. So the users were lucky. Now and in the future, this will change. Devices are getting bigger and faster and thus hotter, and our test methodologies are getting more refined, so we do not have to leave so much guardband "on the table". Users will have to pay attention to real ambient, real power consumption, and real thermal impedance, and not rely on the fact that "everything is always twice as good as the data sheet says". Yes, we give a conservative derating figure of 0.35% delay increase for every degree centigrade above the guaranteed max, 85 or 100 degr.C. That value is conservative ( others say 0.25% ) and is even lower when the delay is dominated by metal resistance and Longline capacitance, both of which change little over temperature. Peter Alfke, Xilinx ApplicationsArticle: 19874
JUST A FEW MINUTES. NOTHING TO LOSE BY READING IT. Visit this web page. Just click on this link: http://www1.gratisweb.com/jr_dinero/money.html
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