Availability of Spartan3

I thing Steve Knapp (Applications Manager for Spartan FPGAs) posted it quite succinctly:

Steve Knapp posted: Xilinx is transitioning the lower-density Spartan-3 FPGAs (XC3S50 through XC3S1500) from the 200 mm wafer production line to the 300 mm line. The larger density Spartan-3 FPGAs (XC3S2000 through XC3S5000) are already built exclusively on the 300 mm line. Xilinx has a policy where we notify customers 90 days in advance of such a change and we cannot ship product from the new fab unless you specifically order us to. This gives customers

90 days to evaluate the new material to see if it affects their production systems. The details are in the following change notice. otifications/xcn05009.pdf

Starting 1-AUG-2005, all orders will be shipped from the 300 mm line. Until then, all production orders for lower-density Spartan-3 FPGAs are still

shipped from the 200 mm line. As a consequence, lead times are artificially increasing during the transition. There are plenty of 300 mm devices in stock. However, due to our notification policy, we can't ship you one of the 300 mm devices unless you specifically ask. You can dramatically improve delivery by appending the part number with the four-number code

"0974". For example, "XC3S200-4PQ208C" would become "XC3S200-4PQ208C0974". End of quote.

I aplogize for the inconvenience and the apparent bureaucratic fumble, but this is the best you can do. We have plenty of parts. The problem is getting them into your hands. Peter Alfke, Xilinx Applications

Is this from UMC or Toshiba? Or Both?

Will the current specifications be maintained? Or is there a new specification to deal with the change in process/FAB?

Whether the parts are made by UMC or Toshiba (Spartan3 is not), and whether they were born on a 200 mm or a 300 mm wafer is of no concern ( or should be of no concern) to the average user. All these parts have to meet (and do meet) the data sheet specification. But since there are (albeit minute) process differences and different mask sets involved, we have to do new qualification tests ( like static discharge) on all pins and on all internal functions. And that takes time. Having multiple fabs is common these days, not only for ICs, also for raw materials, food stuff, machinery, automobiles, household goods, books and magazines, etc. Whether a particular car is assembled in Michigan, Canada, or Ohio should not be a major concern for the buyer. But when you buy a whole fleet of them, you may be interested, perhaps for reasons of consistency and tracability.

IC mask sets and wafer sizes went through many rapid changes over the previous decades, and few buyers were concerned, as long as the specs were met. And the IC supplier had a lot of latitude. Now we have far less freedom, since a smaller-geometry process automatically means a lower supply voltage, and thus a completely new part number. In the 5-V era, we all used to make lots of changes without telling you, unless you were a big corporate customer. Guess how many fabs and mask revisions Intel has on their Pentiums? (Pentia?)

In other words, no user should be concerned about the wafer diameter. We just try to abide by self-imposed rules. (And they, unfortunately, led to the artificial scarcity). Peter Alfke, from home.

Whether the long leads time have been artificially created by Xilinx or not I would strongly suggest anyone using Spartan3s to check with their distributor.

It isn't pretty.

Ricky

Dear (current) Xilinx user,

We design boards that are manufactured in quantity elsewhere, and we depend on easy and quick availability of prototype and pre-production quantities. Particularly for FPGAs, where there are multiple parts in identical packages, we often need to try one size up (for capacity) or one size down (for cost) quickly. We also build small runs of boards for customer approval, always on short turn-around.

The message from Xilinx seems to be that they don't want to bother with anyone that is not also directly responsible for the manufacture (i.e. have the distribution relationships). Spartan 3 parts were formerly available even in pre-production quantities through the web store, and it offered excellent service (so good it was definitely a plus factor in designing in Xilinx parts).

Without the web store, there are today exactly zero distributors that currently have, or can obtain in any reasonable time, the Spartan 3 parts we are using. This is true whether it is from a 200mm fab or

300mm fab. I'm sure with a good distribution relationship there are strings that could be pulled to get something more quickly. However, I'm not sure why we would want to spend time and use up favors for something that should be readily available overnight in small quantities.

It is a week since Xilinx's own representative said he would try and find out if/when the product would be available, and no news looks like very bad news. To cut off supply with no notice, and fail to make any reasonable arrangements with alternative distribution, makes us nervous about continuing to incorporate Xilinx parts even if they do sort this situation out. Our customers expect us to select components from reliable suppliers that won't change the rules in the middle of the game.

Despite their persistent enthusiasm, we have always declined to meet with our local Altera rep because, despite the odd hiccup, we were pretty satisfied with the Xilinx solution. My understanding is that Altera are more committed to multiple distribution channels and generally superior product availability (any satisfied / dissatisfied Altera purchasers out there to comment?).

While hoping that the Xilinx situation will improve quickly and dramatically, past experience with semiconductor suppliers that take their eye off the ball is that one bad decision is quickly followed by others, and it is wise to start looking at alternatives.

We have admitted our sins, we have suggested a relatively simple work-around (the 0974 suffix to the order code), and I can assure you that there are strong efforts underway to correct this untenable situation, as fast as we can. The ugly comments in this newsgroup have been relayed to the highest (and also the lower) levels of management, and there will be action. You have been heard, and you were right. I have used pretty strong language internally, and I expect things to change. It just takes more than a few days... Peter Alfke, from home.

"Peter Alfke" schrieb im Newsbeitrag news: snipped-for-privacy@z14g2000cwz.googlegroups.com...

I am sure this is will be good news. There are many ways and reasons, but I can say one thing:

".. 'taking the S3 OFF from the web store' - this is the one thing one should never do .."

REALLY, I mean it.

If it was possible to sell the S3 from Xilinx webstore then there can be no reasons explainable to the customer why it is not possible any more. The only thing the people will belive is that the store is not carrying them any more because there is no silicon to sell. Doesnt matter what the real cause is.

If the decision to put S3 to Xilinx webstore is now considered as bad decision internally in Xilinx, then Xilinx should still 'stick' with it that decision and keep carrying the S3, even if it is making too much problems.

The Digikey - thats another issue to deal with, I used to checkout Digikey to get price indicators and warnings on component leadtimes. The fact that there is no S3 on Digikey at all, is alarming itself, it means either there are no requests, no interest in S3 or there is no silicon, or that somewhere is a major problem of some sort.

The Memec-Avnet, due to the pending purchasing of Memec by Avnet (what is pending the US anti-trust court approval) I would not count that Avnet or Memec is very interested in anything, Memec has lots of losses, and Avnet is about to pay for those, so both of those have internal things to solve.

And with NuHorizon as Xilinx supplier would I not count at all. So all that together - it is quite understandable that customers have BIG concernes on the availability and about the overall situation with S3 (and other Xilinx silicon)

Antti's

5 cents to the story...

no

thing

S3

those,

that

While Xilinx are listening can I re-enforce some of Atti's comments. I'm a freelance designer and I never buy many parts but some of my customers buy in the 10k+ per annum range. When I get a design contract availability of parts for prototyping is absolutely key to the design in decision. My experience of the large distributors (other than the catalogue guys like Farnell and Digikey) is that they just get in the way. Suppliers with good sample/web sales systems usually win. For examples of how to do it right check out Microchip, Linear Technology and TI.

Michael Kellett

This would certainly speak to the lack of parts, the obvious reason to migrate to 300mm wafers(I was informed that this happened a long time ago) is increased yeild. Making this change now would indicate that the current method is not meeting demand/yield needs. It also suggests that transition to 300mm wafers is not seemless, as the process has to meet the existing specification and not the other way around.

This assumes that all FABs use the same materials and methods, or the specification has to stretch to cover both FABs, correct?

This appears to be a significant issue. Xilinx (both here and in their press releases) has touted Spartan-3 as alive and well for about two years now. However, there have been multiple reports of this nature during that period as a quick scan of comp.arch.fpga history illustrates:

A certain amount of this is tolerable, but not for an extended period of time.

I heard back from the major distributor today, who said that there are indeed no parts available for prototype / pre-production and certainly not larger quantities. They thought there might be a possibility of obtaining something in an industrial temperature grade, but otherwise nothing, magic 300mm suffix or no magic suffix. Otherwise there appears to be nothing anywhere.

The Xilinx web store commented that the reason the S3 was removed from the site was because there was 'no stock'. They made no mention of a policy decision to cut off small quantities. However, they did not offer any explanation why the entire product line in all sizes and packages would simultaneously disappear from the store instead of just showing 'no stock' on the items that were not in stock. There was no time frame given for restoring the items to the web store, or for having anything back in stock. It does not appear that Xilinx is quite ready to come clean just yet.

It is encouraging that Xilinx is listening, which is more than many companies do, though ultimately it needs to be followed up by concrete action. This would seem likely to include restoring the products to the web store (even if out of stock), and then reserving quantities for the web store such that design and preproduction don't get torpedoed because of supply issues on a larger production scale.

Large-scale production supply problems are not good, but Xilinx is certainly not the first to suffer from a popular product or process problem (whatever the case is here). What would be inexcusable would be allowing availability issues to hold up new product designs and marketing prototypes.

I agree that Xilinx would do well to look at TI and similar companies. I am sure that the sample and small-quantity order system must look like a money-losing business. The trouble is that without it, there probably is no business. Perhaps the bean counters pushed the marketers and engineers away from the Xilinx controls?

Could you mention for those in-the-know (since it's not in this thread): 1) what part, package, and speed grade 2) which distributor is giving you the info (city may help)

I love to see things get solved but I don't like to see people get upset without providing the information needed to help them through the "side channels" here on the newsgroup. If there's bad info getting around, that info should be squashed.

I called NuHorizon today - at their 800 number -and was told that there was an order placed June 9 with expected delivery August 30th. This was using the added suffix suggested by Steve Knapp and Peter Alfe.

My interpretation of this is that there are, in effect, virtually no parts available for customers like me who only order 35-50 at a time.

I will look forward to Peter Alfke working the system inside Xilinx. Hopefully he will accomplish something and have better news to report.

The net result of this is that my product is in jeopardy. I've used Xilinx parts for nearly 15 years and would hate to move over to Altera, but I am getting fearful I might have to consider that in order to protect the investment my company has in its product development.

I think Xilinx is a great company, so I am really hoping there is news fairly soon to restore my confidence.

John_H wrote:

I called NuHorizon today - at their 800 number -and was told that there was an order placed June 9 with expected delivery August 30th. This was using the added suffix suggested by Steve Knapp and Peter Alfe.

My interpretation of this is that there are, in effect, virtually no parts available for customers like me who only order 35-50 at a time.

I will look forward to Peter Alfke working the system inside Xilinx. Hopefully he will accomplish something and have better news to report.

The net result of this is that my product is in jeopardy. I've used Xilinx parts for nearly 15 years and would hate to move over to Altera, but I am getting fearful I might have to consider that in order to protect the investment my company has in its product development.

I think Xilinx is a great company, so I am really hoping there is news fairly soon to restore my confidence.

John_H wrote:

So, once again:

(and lead-free or not)

I gave the following part #: XC3S200-4PQ208C0974, since this is listed on the website.

Since there is no letter "G" in the part number, this is the leaded version.

Altera isn't all that good at supplying their devices in prototype quantities. If you look at their web site and see what is actually available for purchase, there isn't very much there.

Leon

Don't worry xilinx_user, there aren't parts available for people who need 1000's either :-)

Ricky.

Have you got the time to rework your PCB? You could consider double footprinting to give bigger range of parts that can be used. FG256 and FG456 will sit within a PQ208 set of pads or put some pins in for an add-on module. Not an optimal solution for production but if you are in a must ship situation it might solve your delema. We have done this before for "critical" projects as a way out of a delivery issue.

We can certainly help with an initial FG456 package of XC3S400 if that gets you out of the problem.

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