cPLDs and FPGAs we've known and loved (or hated)

We're making the tables for AoE 3rd-ed chapter 11, the 2nd digital chapter, in which cPLDs and FPGAs are discussed. What are your favorite companies and parts, the ones you always use, and what are the ones you used to use but stopped (and why), and what are the parts you steer clear of? What about cPLD vs FPGAs? Opinions? What method(s) do you use to design the parts? Thanks!

Reply to
Winfield
Loading thread data ...

Xilinx have a free Linux toolset for their line of FPGA chips. Altera doesn't. Actel have also a free Linux toolset, but a limit on 1 year.

FPGA is useful for large projects, CPLD might solve simpler issues. And most important doesn't need configuration eeprom. However Spartan-3AN is something to look out for.

Define "parts".

As for schematic entry I prefer to not have my design files implicitly hijacked by any entity through fileformat issues.

Reply to
posedge52

Least favorite PLDs are Almel "Zero Power" (CZ suffix) parts. I've found (the hard way!) 3 or 4 different ways you can avoid having the internal circuitry "light up" on input signal transitions the way the spec sheet sez they're supposed to! One of these "discoveries" occaissioned the largest volume field refit I've had to do in my life, to date....

W Letendre

Reply to
WJLServo

I have stopped using Xilinx FPGA because:

1) You can never buy them, I cannot design on promises. 2) Their webpack software is the one but worst piece of crap I have ever come across. 3) Their reps in .nl did not answer the phone during business hours when I wanted to order 10. 4) I found Xilinx postings in comp.arch.fpga insulting. 5) They do not make them in 40 pins DIL. 6) Their pricing is a lie, is is perhaps for >1 M orders, prices are way too high. 7) Most things can be done without FPGA, and sometimes even better (like CPU in FPGA). 8) For every new model you need to buy an evaluation board, and by the time you have one., and finally managed to get something working on it, they declare that FPGA 'old'. Luke Skywater
Reply to
Luke Skywater

I refuse to name any CPLD companies that are currently in business. My record is just about 100% on every time I design one in, they stop making it. I fear that my curse may extend to the mere suggesting of it for design.

A couple of parts and companies of the past are worth considering because the companies did some things very right before they did the very wrong thing.

ICT invented the PEEL which was one of the early electrically reprogramable PLD technologies. They also gave away a program called PLACE that is still in my opinion the best PLD development tool ever made. ICT also made some very nice parts including a 22V10 like part that was more like a 22V22 with a bunch of hidden flipflops bringing the total to 40.

Philips invented the "cool runner" which was an all CMOS CPLD that ran on 5V. Its ouputs could do full CMOS swings and thus could directly drive the gates on small MOSFETs. It really was the first truely low powered CPLD. They abandoned the product line and sold it off to a company that dropped the 5V version.

Reply to
MooseFET

across. Agreed, but they actually have a low entry working Linux solution (althought it's not good).

wanted to order 10. Try

formatting link
..?

formatting link
offers solutions to that.

you have one., and

'old'.

Isn't FPGAs somewhat compatible ..? (design for one, use on another) Programmed in HDL, with accessible I/O pins..

Reply to
posedge52

PEEL 22CV10 was a breakthrough, but we rarely use them for new designs, one in the last year, for some safety logic in an otherwise analog box.

Nowadays the Xilinx Spartan 3 fpga's are great. The tiny VQ100 and bga packages pack a huge amount of logic into a tiny space. We use the free WebPack software,and use an on-board uP to configure the fpga's at powerup.

John

Reply to
John Larkin

John, you like your products to have a 5 to 10-year production lifetime, right? How do you deal with the rapid turnover of cPLD and FPGA part offerings?

Reply to
Winfield Hill

We haven't had that problem. We can stil get the old 4000 series parts. Only the newer bleeding-edge stuff is hard to get at first.

across.

A little buggy and annoying in places, but it works.

wanted to order 10.

Maybe the reps are duds.

How so?

Ah. Humor.

high.

10 or 15 dollars for 200K gates ain't bad.

FPGA).

Not if you want to go fast, or need a couple hundred i/o pins. No uP is going to do, say, 8 channels of complex waveform generation with modulations at 128 Ms/s. A cheap Spartan 3 does, and barely gets warm.

have one., and

We've never eval'd an fpga. That's a monumental waste of time. Just design a product and build it and make it work. The datasheets are pretty good, so read and believe them.

Catchy name, so creative.

John

Reply to
John Larkin

On a sunny day (Mon, 04 Feb 2008 08:34:00 -0800) it happened John Larkin wrote in :

My 1M $ worth: NO SECOND SOURCE.

Reply to
Jan Panteltje

I 100% second that experience. It is not a curse, though - they do not stop making successful products because of being too stupid, there obviously is more to it than that.

Well if dropping the 5V version had been all that "company" did to the coolrunner. Their 3.3V powered version has 5V tolerant inputs so we "only" lost 5 outputs there (I can imagine the "only" would sound to you if your design relied on that 5V output....).

But while Philips did hand us the fusemap data so I could incorporate them in my DPS based toolchain, the "company" did not - in spite of agreeing with Philips in the takeover papers to fully support existing customers.

In a recent twist, they offered me the data under NDA (great progress and seemingly a major change of policy); well, two months later and two different NDA formulars they sent me and got back signed I am still waiting to hear from them. They do have good techical personell and decent people working for them, but overall they are clearly a "company" serving party members only.

Dimiter

------------------------------------------------------ Dimiter Popoff Transgalactic Instruments

formatting link

------------------------------------------------------

formatting link

MooseFET wrote:

Reply to
Didi

We can still buy 10-year old Actel and Lattice and Xilinx parts, and

22V10's. The prices go up, but it's better than redesiging. A couple of our best selling products are full of 4000-series and Spartan 1 Xilinx chips.

Discretes, especially RF stuff like mmics and phemts, are a bigger problem. A lot of that stuff is being killed by ICs. We've also had problems with adc's and dacs lately, older parts going away, especially in dips.

John

Reply to
John Larkin

No fpga's are second-sourced. It only happened once to my knowledge, Actel and TI, and that didn't last.

There are whole classes of products that would be impossible without fpga's.

Very few high-performance parts are multisourced.

John

Reply to
John Larkin

I really like zero-power PLDs because they swing rail-to-rail--I've often used them for analogue things--run the PLD off a quiet supply and you get quiet levels, unlike uP pins.

My favourite gizmo along that line was a process control sensor for post-exposure bake of chemically amplified photoresist--it used 7 solar cells arranged overlapping in a heptagonal prism shape. I connected all the cathodes to a cascoded TIA and used the PLD to switch the anodes between tristate and ground, one at a time. It worked brilliantly.

Cheers,

Phil Hobbs

Reply to
Phil Hobbs

Un bel giorno Winfield digitò:

The bigger issue with FPGAs of interesting sizes (300k and more) is that either they aren't available at all, or you are required to buy a truckload of them. There is no way to find a distributor willing to give you sampling quantities (1-3 pieces at a time); they just apply the same logic of the other components, no matter if they are 555 or a two-million gates, 1k$ FPGA.

Mostly VHDL.

--
emboliaschizoide.splinder.com
Reply to
dalai lamah

I thought the (very old, now) Xilinx 3k series and AT&T (Lucent) ORCAs came from a friendly collaboration between the two companies as well?

We had a consultant who, upon seeing some of the problems we were having with Xilinx 4k parts suggested we see what Lucent had done with the ORCAs. Unfortunately, the answer was "pretty much nothing" and he quickly agreed that for all their problems Xilinx was still the way to go.

Back in the mid '90s...

I rather liked the Vantis Mach4 parts... they were acquired by Lattice, but I don't know if they're still producing them or not.

Cypress CPLDs were readily available and really did tend to go a bit faster than most competitors' otherwise comparable devices.

Low-end FPGAs such as Spartan 3's today have killed a lot of the CPLD market, it would seem.

---Joel

Reply to
Joel Koltner

Which parts do you mean? Most Altera parts are orderable at the Altera online shop in single units, even the really fat ones, like Stratix, same at Digikey.

And the availability is good: The Cyclone I device, which a client for which I work has used for a product 4 years ago, is still available and there are upgrade paths for Cyclone II and Cyclone III. Some parts are obsolete, but you can ask them and they suggest parts. For some parts (e.g. the MAX II) they guarantee 7 years availability (at least by eMail last time I asked). And they have an end-of-life web page, so you can plan to use other parts, if necessary. Maybe Xilinx has another company policy.

--
Frank Buss, fb@frank-buss.de
http://www.frank-buss.de, http://www.it4-systems.de
Reply to
Frank Buss

One of the few things to mention is that Xilinx holds a patent which allows them (only them) to use the LUT as a memory or shift register. As a result you have 34 flipflops per slice instead of 2. This gives a huge advantage when designing for example a filter for multiple channels which are processed sequentally. By using a LUT memory as an intermediate storage for results you can process up to 16 channels with very little more logic than it would take for one channel. Multiplexing channels comes for free; simply connect the address lines to the channel counter.

The LUT memory can also be used as addressable memory and concatenated to form larger memories. Because the resulting memory block consists of small grains, the FPGA can be routed more efficiently.

A divide by X (where X is large) can be build with a few slices using shift registers in a LUT.

As a result you can put a lot more logic into a Xilinx FPGA compared to their competitors.

One of my recent designs uses about 25% of the LUTs as a memory which saved over 10000 flipflops. The equivalent gate count of the design is around 800k, the device itself is rated for 200k gates. Best of all, the device isn't full yet...

--
Programmeren in Almere?
E-mail naar nico@nctdevpuntnl (punt=.)
Reply to
Nico Coesel

Long ago I figured out that they just do it to mess with me.

Yes I did rely on the output swing. I also had other parts in the design that could not be dropped to the 3.3V. I also figured if I did, they'd quit making that one too.

Different people take "fully support" to mean different things. To fully can mean to hold up without any help. They did hold up our production without any help so I guess you could say they "fully supported" us. :-o

My next design has no CPLD in it at all. I've got about 10 logic ICs and some fast micros. I'm not going to use another CPLD if I can avoid it. Unfortunately they are just way too handy.

Reply to
MooseFET

I like them too because the good ones have very low static power levels and I agree about the lack of noise. Some of them also work quite nicely at low speeds with poorly regulated power supplies. This makes leaving them powered on practical.

Reply to
MooseFET

ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.