Anyone remember the really early Xilinx FPGAs?

Mike,

I was a user then, so I will try (to answer).

See below,

Austin

I saw a board with one of each on them in 1987. I know both went into production, and I also know they were declared obsolete (after a last time buy) the year I joined Xilinx (1998). I believe the parts were shipped for 15 years, total (I found a tube of 2018's in a file cabinet only a year ago...). You can still obtain small quantities over the web at various obsolete parts houses.

We started with the 3030, and graduated immediately to the 3042 for many of the designs we started in 1990. Couldn't convince my boss to use FPGAs until then.

I am not sure, but the original family did not include the 3090. It was added later.

I believe the A was the first process shrink, and that happened in 1995. Never used the 3100 until much later, as I expect that was another process shrink, and the specs changed, so the part number changed, too.

2064 I think. See

--mt

Since I am the oldest Xilinx veteran here ( Jan 88), I can answer with authority:

The first part, in 85, was the 2064 (named after the number of CLBs in the matrix), followed soon by the 2018, named after the (contentious forever) number of gate equivalents. The 3000 series was introduced in the following sequence (sorry Austin, I was there):

3020 in late 87, 3090 was the second (!) in mid 88. 3042 came soon after and becamemost popular, then (early 89?) the 3062 as the last-born and forever least popular. The 0riginal 3090 die was exactly 100 square-millimeters, but it was not 10x10 since we wanted to fit two masks into the biggest possible square reticle, so it was something like 12.5 x 8 mm, and we proudly depicted it (to scale) on the back of the data book. Xilinx has, forever since, always pushed manufacturing to offer the biggest possible top-end device, because we know that there are designers salivating for something even bigger, and the unavoidably high price is grudgingly accepted when there is no alternative.

Process shrinks were done more quietly in those days, since they did not affect the user with a change in supply voltage. Those were the 5-V days, when everybody used the same Vcc :-)

3000A was a functional superset, and 3100 offered higher speed through "pumped gates", internally generating a higher Vcc for certain circuit detailss. Also a new top-end, the 3195. I wrote a candid comparison of the various 3000 families and published it at the front of the family datasheet, with an innovative 3-dimensional picture... See: Peter Alfke, Xilinx Applications

Many of your questions are answered at the following URL. (Also may answer some questions you have not asked).

I have also answered to the best of my knowledge below:

2064 first (64 LCA Cells), 2018 next (1800 gate equiv. the start of the great gate counting debate) 3020 was first, then I think 3090, then the rest.

See the URL above

Have fun, Philip

Philip Freidin Fliptronics

I stand corrected....

I was a customer then.

Aust> Since I am the oldest Xilinx veteran here ( Jan 88), I can answer with

Here's some more history (from a long term Xilinx user):

There was a time when companies didn't trust any part that didn't have a second source, so companies would license their designs to other manufacturers. So Xilinx licensed its designs to AMD, and then later to ATT. So you could buy a 2064 from AMD, and 3000 series parts from ATT.

In both cases, for various reasons, Xilinx cancelled the agreements.

ATT didn't want to get out of the business, so they made their own FPGAs, called the Orca family, roughly similar to the XC4000 family. ATT later became Agere, and Agere later sold the Orca family to Lattice, where you can still buy it.

Here's some more history from a long time Xilinx user:

There was a time when companies didn't trust any ICs that were not second sourced (more than one manufacturer). So Xilinx licensed their designs to AMD, and then later to ATT. You could buy a 2064 from AMD and later a 3000 series part from ATT.

For various reasons Xilinx cancelled these agreements (there's no doubt some interesting stories here).

ATT didn't want to get out of the FPGA business so they developed their own Orca line, roughly equivalent to the XC4000 family. The IC devision of ATT later spun off to become Agere. Agere later sold the Orca line to Lattice, where you can still buy it.

Sorry for the double post. The system said it couldn't process my reply, so I had to retype it from memory. Now I find out it went through!

To be precise: Xilinx licensed MMI before that company got swallowed up (there are uglier words for that procedure) by AMD. Soon after that, Xilinx bought the rights back from AMD (who made a nice profit from the Xilinx shares...) For XC4000, ATT was supposed to be the prime manufacturer, not only the second source. That really did not work out. We are now much better off with our friends in Taiwan (UMC) and Japan (Toshiba). "Second source" seems to be a forgotten word now, and the world is a better place without it. Peter Alfke

Hi Peter, I still try to write as much 'portable' VHDL as possible, and keep any device specific stuff in separate files. It gives the designer a 'virtual' second source, and at least gives a bit of a bargaining chip while discussing pricing! Cheers, Syms.

Mike

If get this all together I won't mind a copy. I have some of our free seminars to present next month and this as an intro to near virgin FPGA users is always useful. I do remember a lot of this being a veteran since the early 3064/90 days and pushing the technology to about 50MHz which was very fast in those days. 50MHz now rates as near DC in my design book nowadays. However given the number of FPGA devices and projects I have been involved in over the years, or maybe it was centuries, I wouldn't like to rely on my memory on what happened when and where.

John Adair Enterpoint Ltd. - Home of Broaddown2. The Ultimate Spartan3 Development Board.

Hi Symon. You know why you do it, but I suppose you are also aware of the performance and the money you leave on the table by "designing to the lowest common denominator". Cheers Peter Alfke

Peter,

It may be he (Symon) has no choice (about designing using generic features only).

Where I once worked, there was a policy that everything HAD to have a second source, no exceptions allowed.

As you recall, the agreement with AT&T by Xilinx allowed for a second source. It was that agreement that "allowed" me to design in my first FPGA. This was the time when "second source" was a popular catch-word with purchasing groups, and business tyoes.

I imagine that havinf a second source was a breakthrough for many other designers at the time, as well, who were prohibited from using the most advanced technology by this fad.

However, once designed in, the "second sourced" FPGA became a required standard component for every board after that.

At no time did they (purchasing) ever even ask AT&T (and after, then Lucent) for a quote: it seems that qualifying another vendor (through the components engineer) was "too expensive."

Absurd? Yes.

So, the 'bean-counters' got their way ("must have two sources") and the components engineers got their way ("don't bother me with another vendor").

I happen to agree with you, then as well as now, the generic approach (was) is less efficient, and therfore more expensive.

Austin

I was in the exact same predicament. Everything had to have second sources -- AT&T was the reason we went with the Xilinx 3090's. We used a ton of the old XC3090-100PPG175's (or something like that). We also used a ton of the AT&T version as well. The AT&T's typically ran a bit more $ but seemed to be more available at the time.

We stopped requiring second sources when the board component count was down to not much more than half a dozen IC's. At this point, it was easier to redesign the board with new technology than to search for alternates.

Luckily, we were never stiffed by Xilinx regarding part availability for old parts. Last year, we had a special request to remanufacture a set of boards we put out of production years ago. Each had two XC3090's on it - I was quite surprised at how easily we were able to find unused XC3090's after all this time.

And how can anyboty forget the routed signal delays thru the old 3000 series of parts?!?! I did a whole lot of hand routing back in those days. Not much need for that anymore, thankfully. I can't exactly imagine hand routing a XCV3200E anyways...

-- Ed

Ed, I remember back in the 80's closing my eyes and going to sleep after a hard day's 'XACT'ing and still seeing the image of those switchboxes burnt into my retina. Cheers, Syms.

. . .

I have two items which may help, please email ed at my domain if you'd like to discuss:

The first edition of "The Programmable Gate Array Handbook" (c) 1986 with prices for XC2064, EP310 & EP320 penciled inside the cover.

A PAL video in which Brad Fawcett gives a full technical introduction to the FPGA and associated design techniques. The video mentions 2k and 3k parts, the latest mentioned date is Feb 1988.

When we updated our 3k designs we used 5200 series parts, but I'm not sure where these fit into the Xilinx development path.

Ed Coombs.

-- murray-microft ltd