XC7S15-1FTGB196C Unobtainium

[gnuarm]

I am looking at respinning a board that uses an EOL Lattice part and using a Xilinx part instead.  XC7S15-1FTGB196C seems like a reasonable choice, but I can't get any read on pricing.  The XC7S6-1FTGB196C is $17 at Mouser and under $10 at LCSC. 

The XC7S15-1FTGB196C is $22 at Mouser, but $120 a LCSC.  Then the XC7S15-1FTGB196I is only $12 at qty 1000 at LCSC. 

I can't get a read on any of this.  I suppose this is the shortage at work.  I'll be using 1,000 a year.  Will I get any traction with Xilinx if I try to talk to them directly?

[SiliconWizard]

You're lucky if you can get ahold of any FPGA that you actually need these days.

[gnuarm]

You're lucky if you can get ahold of any FPGA that you actually need these days.

Not really trying to get parts now.  I'm looking to figure out what parts I should use, based on price and availability down the road six or twelve months.  Does Xilinx at least provide samples these days?

One problem is I have no feel for how good the Xilinx 6-LUTs are compared to a 4-LUT.  The -6 chip has about 3600 of them (compared to the mythical 6000 Logic Cells).  My existing chip has 3000 4-LUTs and is around 90% full.  I need to shove two of these circuits into a new chip and I'm thinking I need the -15 Xilinx part with 8,000 6-LUTs should do the job fine.  I could even go with the -25 part with 14,600 6-LUTs if the price is not too bad. 

I like the FTGB196 package as it fits the board ok, although I may be making a double board.  Not sure how many more I/Os I would need for a twin board, but the 100 I/Os in the FTGB196 seems a bit tight.  Maybe not, we haven't looked at the   In that case, I might go with a 256 pin BGA with 1.0 mm pitch.  I don't see that for the Spartan 7.  I'd have to go chip hunting again. 

[Someone]

Its an FPGA, you run a (the) build to see what the resource use is before committing to a part, be that size or speed grade or IO etc

[nctnico]

I'd look elsewhere. Gowin for example. The  XC7S15-1FTGB196C doesn't look like a very high end device.

[pcprogrammer]

Like Someone advised, just download Vivado and try your design to see what it makes of it.

Other option is to look at Gowin, like SiliconWizard suggested, or look at Anlogic. They have the AL3 series that resembles the Altera Cyclone IV series.

The software for it is free, but needs a license key, which can be downloaded from a Sipeed website. When you need it I can provide the links, but have to look on my other computer for it

[SiliconWizard]

Other option is to look at Gowin, like SiliconWizard suggested, or look at Anlogic. They have the AL3 series that resembles the Altera Cyclone IV series.

Yeah, that was nctnico actually.

Now while we don't know exactly what the OP's requirements are, they said the XC7S15 looked like a match, and from this: "My existing chip has 3000 4-LUTs and is around 90% full.", it would definitely be more than enough (unless the old design used some particular feature of the FPGA that would require a lot more LUTs on the XC7S15). The easiest would have been to just tell us what the FPGA they are currently using is, and what frequency the design runs at.

It's very possible they could even go for a lower-end FPGA than the XC7S15, such as in the iCE40 series or something. Since they mentioned a Lattice part (but which one was it), most of their deprecated parts are relatively low-end stuff from what I remember, so that should be no problem.

Not sure off the top of my head what is EOL at Lattice currently. I'd suspect either the MachXO (not 2 or 3), or old CPLD, or...? Or maybe XP2, is it EOL?

[mon2]

GOWIN has drop in replacements for a mix of the Lattice (and other vendor's) devices:

https://alcom.nl/products/gowin-littlebee-and-aurora-families-as-lattice-semis-machxo2-en-xo3-alternatives

Don't forget to also review Efinix as a possible solution. Available through Digikey.

[pcprogrammer]

Other option is to look at Gowin, like SiliconWizard suggested, or look at Anlogic. They have the AL3 series that resembles the Altera Cyclone IV series.

Yeah, that was nctnico actually.

Sorry, you are right, my bad

[gnuarm]

I'd look elsewhere. Gowin for example. The  XC7S15-1FTGB196C doesn't look like a very high end device.

Not sure why you say that.  It's high end enough.  The 7S6 with less than 4000 6LUTs would be marginal.  I have high confidence the 7S15 with 8,000 6LUTs would do everything I need and more. 

What is your point exactly?

I would like to consider Gowin, but they are Chinese and my customer sells a lot to the US government.  Gowin might not be the best choice.  At one point they were on a list of companies considered to be a bit too close to the Chinese military.  They got taken off the list, but they are still made in China where the Military can control businesses. 

What was that company?  Huawei?  Yeah, that one...

[gnuarm]

Like Someone advised, just download Vivado and try your design to see what it makes of it.

Other option is to look at Gowin, like SiliconWizard suggested, or look at Anlogic. They have the AL3 series that resembles the Altera Cyclone IV series.

The software for it is free, but needs a license key, which can be downloaded from a Sipeed website. When you need it I can provide the links, but have to look on my other computer for it

Both Gowin and Anlogic are Chinese I believe.  That's an automatic 3rd strike for companies selling networking equipment to the US government.

Why are people telling me to run the design through the tools?  I don't need to do that now.  I have a reasonable size estimate and I would not want to use the 7S6 part, even if it would fit.  It is not smart to squeeze a design into a part if you have any thought of reworking the design later.  This is an application where a couple of dollars won't matter, but being able to add to the design later could be a big advantage.  The 7S15 provides that extra space.

I'm just trying to find info on which parts are a reasonable price.  From what I'm seeing, the Xilinx parts may not be a good fit regardless of the number of LUTs.

[gnuarm]

Other option is to look at Gowin, like SiliconWizard suggested, or look at Anlogic. They have the AL3 series that resembles the Altera Cyclone IV series.

Yeah, that was nctnico actually.

Now while we don't know exactly what the OP's requirements are, they said the XC7S15 looked like a match, and from this: "My existing chip has 3000 4-LUTs and is around 90% full.", it would definitely be more than enough (unless the old design used some particular feature of the FPGA that would require a lot more LUTs on the XC7S15). The easiest would have been to just tell us what the FPGA they are currently using is, and what frequency the design runs at.

It's very possible they could even go for a lower-end FPGA than the XC7S15, such as in the iCE40 series or something. Since they mentioned a Lattice part (but which one was it), most of their deprecated parts are relatively low-end stuff from what I remember, so that should be no problem.

Not sure off the top of my head what is EOL at Lattice currently. I'd suspect either the MachXO (not 2 or 3), or old CPLD, or...? Or maybe XP2, is it EOL?

The old chip is LFXP3C-3TN100C and LFXP3E-3TN100C.  Cost was down to $7 each at one point, before they went EOL maybe 8 years ago.  Arrow bought a bunch of stock allowing me to continue to produce them until an AKM factory burned down making their CODEC unavailable.  I managed to buy nearly 9,000 on the grey market for the present run.  Only got ripped off twice.  LOL

The design runs with 33 MHz and 12.288 MHz clocks. 

Knowing all that info, what is the price of XC7S15-1FTGB196C, given an annual purchase of 1,000 units? 

Just in case I can use the Gowin parts, I got a budgetary price on the GW1N-LV9LQ100 and GW1N-UV9LQ100 at qty 1,000.  That was before I was asked to consider a dual board which would need a larger package I expect.  I can fit a LQ144 on the larger dual board.  That would provide enough I/Os. 

[SiliconWizard]

Oh OK, it's not even a XP2 then but a XP. (I didn't even knew about them.)
Given your requirements the only additional thing we may need to give you proper alternative(s) is: the number of IOs you need and what amount of embedded RAM (if any) your design uses.

[gnuarm]

Oh OK, it's not even a XP2 then but a XP. (I didn't even knew about them.)
Given your requirements the only additional thing we may need to give you proper alternative(s) is: the number of IOs you need and what amount of embedded RAM (if any) your design uses.

You are truly an engineer.  I'm not asking anyone to find an alternative part for me.  I was asking about how to get a price on a part that I could have some confidence in. 

But that's ok.  I have not locked into any part.  In fact, I've got a new thought, that I'm going to post in a new thread, since it's a bit off this topic.  Thanks for the effort though.  If you want to make more suggestions, the original design used pretty much all 62 I/Os in the Lattice XP3 device in the QFP100 package.  Some of those go away, so I think we can make do with 100 I/Os, but 120 would make me more comfortable.  I also don't want to use a BGA with smaller than 1 mm ball pitch.  A QFP144 might work if the I/O count is high enough.  On the double board we'll have the room a pair of QFP100s were occupying plus a bit more as some level shifters are going away.

[dawnclaude]

What do you think about Efinix FPGAs? T8Q144I4 seems like a good match for your IO and LUT requirements with currently 22k stocked at Digikey at $8 a piece.

https://www.digikey.com/en/products/detail/T8Q144I4/2134-T8Q144I4-ND/11591372

They are supposedly US based ( ? )

[gnuarm]

What do you think about Efinix FPGAs? T8Q144I4 seems like a good match for your IO and LUT requirements with currently 22k stocked at Digikey at $8 a piece.

In that package, you can get two cores, the T8 and the T20.  The T8 seems a bit small for my needs.  Current chip is 3k LUTs times two is 6k LUTs, plus room for growth is pushing on 7,400 LUTs.  Then there is the uncertainty of fit because the LUT/FF cells are used for routing!  I worked with Concurrent Logic devices which were that way.  They used a lot of cells for routing.

https://www.digikey.com/en/products/detail/T8Q144I4/2134-T8Q144I4-ND/11591372

They are supposedly US based ( ? )

The T20Q144C3 looks a bit more interesting and is still only $5.50 @ 1k through LCSC.  These parts are stocked well at Digikey if not LCSC, but at much higher prices.

I seem to recall you have to buy a dev kit to test drive their software.

[up8051]

Unfortunately, the LCSC stopped stocking Gowin and Efinix chips and their prices were much better than DigiKey.

Maybe you need to do an e-mail action asking to bring these producers back into the offer.
I sent such an e-mail with a request to restore the Gowin and Efinix chips for sale by LCSC

[NorthGuy]

I was asking about how to get a price on a part that I could have some confidence in.

Not possible. Unless you get a time machine and go back in time.

[Gribo]

If the Xilinx part fits your requirement, I would have placed an order now, even with 50 weeks lead time. At 1K/year, you won't get special treatment, and if your order doesn't get pushed back, consider yourself lucky.

[nctnico]

If the Xilinx part fits your requirement, I would have placed an order now, even with 50 weeks lead time. At 1K/year, you won't get special treatment, and if your order doesn't get pushed back, consider yourself lucky.

That doesn't make any sense. You will need parts and some assurance you have those parts in time for your production run. Betting on Xilinx is a very bad idea IMHO.

I'm under the impression that only newer parts are available in small quantities. Remember Xilinx has been taken over by AMD and AMD does not really have a good track record keeping chips available. Especially for the lower end / low cost parts for low volume customers. Been there, done that. Currently there are so many companies jumping onto the FPGA bandwagon (enabled by the free / Opensource Yosys software). For some projects I'm designing Xilinx out and I'm not going to use Xilinx FPGAs for any new projects if possible. Some of my customers are considering discontinueing products that use Xilinx FPGAs that have unspecified lead times (technically the products are already discontinued because they can not be made). New products -without Xilinx products- will be ready sooner. Xilinx has been good to me for several decades but I guess it is time to say goodbye.

[gnuarm]

If Xilinx is out, who exactly, is in? 

I don't know if I want to fight the battle to use Gowin.  It doesn't help that they were on a US list of Communist Chinese Military Companies at one time.  My customer sells to the government a lot. 

Efinix might be a choice, but they have a goofy routing using logic functions, which clearly means you need more LUTs than any other logic family.  I can't test drive their tools without buying an eval board, so no test designs for now. 

Anlogic website is all Chinese.  Even if Google will translate that for me, I'm not sure I can use a product that can't decide if it is an eagle or a falcon.  It would also help if I could download a data sheet, or maybe not. 

[nctnico]

There is Cologne Chip from Germany:
https://www.colognechip.com/programmable-logic/gatemate/
This company has been around for several decades. Digikey has their CCGM1A1 device in stock.

[gnuarm]

There is Cologne Chip from Germany:
https://www.colognechip.com/programmable-logic/gatemate/
This company has been around for several decades. Digikey has their CCGM1A1 device in stock.

I looked at them before.  I don't remember what the issue is, but they were DoA for some reason.   A quick looks shows a single part, in a single package (rather large, 324BGA) at a high price.  They have some sort of a LUT tree that I'm pretty sure I was not impressed with. 

Has anyone here used them?  It doesn't strike me as a part to use when concerned about supply.

[EverydayMuffin]

What about Microchip IGLOO2?

M2GL010-TQG144 is 12,000 Logic Elements.

1680pcs in stock with Microchip.

https://www.microchipdirect.com/product/M2GL010-TQG144?productLoaded=true

[gnuarm]

What about Microchip IGLOO2?

M2GL010-TQG144 is 12,000 Logic Elements.

1680pcs in stock with Microchip.

https://www.microchipdirect.com/product/M2GL010-TQG144?productLoaded=true

I've never been a fan, without any particular reason I can recall.  I do know that the previous design didn't have room for larger parts (21 mm wide), so that excludes a lot of FPGAs without going to finer pitch BGAs. 

I don't recall the details on the IGLOO2 parts, but there's often an issue with the equivalence of LUTs. Also, this part is a bit pricier, $15 @ 1k qty. 

Maybe I need to take a fresh look at their parts. 

Wow!  They can ship 1,600 now, but... "Additional quantities estimated to ship by 16-Apr-2024"  They know how to put the "time" in lead-time!

[EverydayMuffin]

I've never been a fan, without any particular reason I can recall.  I do know that the previous design didn't have room for larger parts (21 mm wide), so that excludes a lot of FPGAs without going to finer pitch BGAs. 

I don't recall the details on the IGLOO2 parts, but there's often an issue with the equivalence of LUTs. Also, this part is a bit pricier, $15 @ 1k qty. 

Maybe I need to take a fresh look at their parts. 

Wow!  They can ship 1,600 now, but... "Additional quantities estimated to ship by 16-Apr-2024"  They know how to put the "time" in lead-time!

The TQG144 package is 20x20mm, worth considering that there is also a small space saving because these FPGAs are non-volatile and don't require an external flash chip.

A Microchip Logic Element is LUT4 so the M2GL010 which has 12,084 LUT4 (roughly equivalent to 8,000 LUT6).

Looks like Mouser have 2,227pcs in stock of the more expensive Industrial temperature grade part.

https://www.mouser.ie/ProductDetail/Microchip-Technology-Atmel/M2GL010-TQG144I?qs=pU29NIZ4ZwCkB%252BTNyuGkrw%3D%3D

[gnuarm]

I've never been a fan, without any particular reason I can recall.  I do know that the previous design didn't have room for larger parts (21 mm wide), so that excludes a lot of FPGAs without going to finer pitch BGAs. 

I don't recall the details on the IGLOO2 parts, but there's often an issue with the equivalence of LUTs. Also, this part is a bit pricier, $15 @ 1k qty. 

Maybe I need to take a fresh look at their parts. 

Wow!  They can ship 1,600 now, but... "Additional quantities estimated to ship by 16-Apr-2024"  They know how to put the "time" in lead-time!

The TQG144 package is 20x20mm, worth considering that there is also a small space saving because these FPGAs are non-volatile and don't require an external flash chip.

A Microchip Logic Element is LUT4 so the M2GL010 which has 12,084 LUT4 (roughly equivalent to 8,000 LUT6).

Looks like Mouser have 2,227pcs in stock of the more expensive Industrial temperature grade part.

https://www.mouser.ie/ProductDetail/Microchip-Technology-Atmel/M2GL010-TQG144I?qs=pU29NIZ4ZwCkB%252BTNyuGkrw%3D%3D

The bigger space saving is from the fact that QFP parts have no obstructions under them (usually) which frees up routing.  BGAs require a via at most pins and QFN parts have a big heat slug on the bottom forcing vias to the outside.  I've never seen a QFN that I thought would use less space on a PCB than the equivalent QFP.  Not sure where BGAs fall as I've not used many.

The 100QFP on the current board has 8 element resistor packs on the other side of the board.  They fit well. 

They are there to assure a valid state of the I/Os while Config is held asserted and until the chip comes out of configuration.  Often, it is hard to know what the chips are doing until they've started up.  FPGA makers tend to spread this info around, rather than having a single section discussing it.

[josuah]

There look like being some GW1N FPGAs on Mouser at the moment: https://www.mouser.fr/c/semiconductors/programmable-logic-ics/fpga-field-programmable-gate-array/?m=GOWIN%20Semiconductor&instock=y

Get them while they're hot!

Although highly priced!
I wonder how the Sipeed Tang Nano 9k managed to be sold for $5.90 https://www.seeedstudio.com/Sipeed-Tang-Nano-FPGA-board-powered-by-GW1N-1-FPGA-p-4304.html

And still today, the board is less expensive than the naked chip.

[gnuarm]

There look like being some GW1N FPGAs on Mouser at the moment: https://www.mouser.fr/c/semiconductors/programmable-logic-ics/fpga-field-programmable-gate-array/?m=GOWIN%20Semiconductor&instock=y

I have a quote from a disti around $6 for the GW1N-9 parts in QFP100 and QFP144 packages, about $1 difference in price, qty 10,000.  Delivery, 16 weeks.

My concern is that my customer won't want to use parts from mainland China because they sell a lot to the government.  It's actually rather complicated.

[gnuarm]

I was doing a quick scan of the FPGA landscape today just to check current status and found the XC7S15-1FTGB196C at Mouser was expecting a delivery in January!  My heart was pounding, until I noticed that was January of 2024! 

This is getting to be a bit like waiting for rain to come, or the next Tesla model to come out, or politicians to become trustworthy! 

[NorthGuy]

I was doing a quick scan of the FPGA landscape today just to check current status and found the XC7S15-1FTGB196C at Mouser was expecting a delivery in January!  My heart was pounding, until I noticed that was January of 2024! 

Don't worry, AMD(Xilinx) sent an email recently where they said that all Spartan 7 and Artix 7 will continue to be available until 2035. That's for your piece of mind If anyone only knew what they mean by "available" ...

[SiliconWizard]

Yes, the "available" word has gotten a whole new meaning starting with 2020.

Whatever promise they make only holds until there is a case of force majeure, and those have become the norm rather than the exception, so don't hold your breath.

[nctnico]

I was doing a quick scan of the FPGA landscape today just to check current status and found the XC7S15-1FTGB196C at Mouser was expecting a delivery in January!  My heart was pounding, until I noticed that was January of 2024! 

Don't worry, AMD(Xilinx) sent an email recently where they said that all Spartan 7 and Artix 7 will continue to be available until 2035. That's for your piece of mind If anyone only knew what they mean by "available" ...

That will probably be old stock as everyone has designed them out by now and they can't sell what they produce next year or the year after. I'm ditching Xilinx in favour of FPGA vendors that had and have devices available to buy right now. Probably save quite a bit of money in the process as well.

[gnuarm]

I was doing a quick scan of the FPGA landscape today just to check current status and found the XC7S15-1FTGB196C at Mouser was expecting a delivery in January!  My heart was pounding, until I noticed that was January of 2024! 

Don't worry, AMD(Xilinx) sent an email recently where they said that all Spartan 7 and Artix 7 will continue to be available until 2035. That's for your piece of mind If anyone only knew what they mean by "available" ...

That will probably be old stock as everyone has designed them out by now and they can't sell what they produce next year or the year after. I'm ditching Xilinx in favour of FPGA vendors that had and have devices available to buy right now. Probably save quite a bit of money in the process as well.

And which FPGA vendors would those be?  I can get parts from Gowin, but they may not be acceptable to my customer, who is concerned about stability of production, since Chinese companies could be cut off if tensions rise.  Otherwise, a search has revealed nothing, from any company, other than four Igloo parts from the former Actel and a stray Atmel part that is on the verge of EOL. 

[nctnico]

If Gowin stops, just move to a different vendor. Just make sure that you can move your project to a different vendor easely. There is a lot of choice nowadays triggered by the opensource synthesis tools which takes away the huge investment to write synthesis tools. The devices from  Cologne Chip look very interesting and they are based in Germany.

One of my customers is basically dead in the water due to Xilinx not delivering chips. That is way worse than having to depent on a Chinese manufacturer. And I strongly doubt Chinese manufacturers are going anywhere. In fact, the chips from Chinese manufacturers seem to be least affected by chip shortage so they look like a very safe bet to me when the sh!t hits the fan. Gowin -for example- seems to have their products manufactured in Taiwan by TSMC. The chances of anyone or any country messing with TSMC are zero.

[asmi]

If Gowin stops, just move to a different vendor. Just make sure that you can move your project to a different vendor easely. There is a lot of choice nowadays triggered by the opensource synthesis tools which takes away the huge investment to write synthesis tools. The devices from  Cologne Chip look very interesting and they are based in Germany.

That is only possible if your designs are relatively simple. If your design requires DDR3/4 memory, uses multi-GB serial links, or PCIE, or other advanced features, forget about vendor-neutrality. And no opensource tools will help you with that.

[nctnico]

If Gowin stops, just move to a different vendor. Just make sure that you can move your project to a different vendor easely. There is a lot of choice nowadays triggered by the opensource synthesis tools which takes away the huge investment to write synthesis tools. The devices from  Cologne Chip look very interesting and they are based in Germany.

That is only possible if your designs are relatively simple. If your design requires DDR3/4 memory, uses multi-GB serial links, or PCIE, or other advanced features, forget about vendor-neutrality. And no opensource tools will help you with that.

The solution is complicated but quite comparable with what I'm facing: take all that out of FPGA territory. In a lot of designs stuff is pushed into an FPGA just because it is kind of a catch-all. But it doesn't need to be; if you partition a design differently, you should be able to take large portions of complicated things out of the FPGA and use off-the-shelve chips for those functions. One of the designs I'm working on already went from a $2500 FPGA to a $100 FPGA + $100 companion chips. Next phase is to take the FPGA and split it into a $20 FPGA + four $5 chip. See how the cost savings are adding up as well? At one of my previous employers we split a design for a single FPGA into multiple, smaller FPGAs and ended up with a board that saved 60% on the costs for FPGAs.

[asmi]

The solution is complicated but quite comparable with what I'm facing: take all that out of FPGA territory. In a lot of designs stuff is pushed into an FPGA just because it is kind of a catch-all. But it doesn't need to be; if you partition a design differently, you should be able to take large portions of complicated things out of the FPGA and use off-the-shelve chips for those functions. One of the designs I'm working on already went from a $2500 FPGA to a $100 FPGA + $100 companion chips. Next phase is to take the FPGA and split it into a $20 FPGA + four $5 chips. See how the cost savings are adding up as well? At one of my previous employers we split a design for a single FPGA into multiple, smaller FPGAs and ended up with a board that saved 60% on the costs for FPGAs.

Again, that is only possible if your design is relatively slow and doesn't utilize modern high-speed interfaces or features. And you also pay for that by increased complexity of entire design (== higher R&D cost), longer time to market, and lower reliability (more chips == more things to fail == lower reliability). If your design uses any more-or-less modern features like fast DDR memory, you are out of luck.

[nctnico]

The solution is complicated but quite comparable with what I'm facing: take all that out of FPGA territory. In a lot of designs stuff is pushed into an FPGA just because it is kind of a catch-all. But it doesn't need to be; if you partition a design differently, you should be able to take large portions of complicated things out of the FPGA and use off-the-shelve chips for those functions. One of the designs I'm working on already went from a $2500 FPGA to a $100 FPGA + $100 companion chips. Next phase is to take the FPGA and split it into a $20 FPGA + four $5 chips. See how the cost savings are adding up as well? At one of my previous employers we split a design for a single FPGA into multiple, smaller FPGAs and ended up with a board that saved 60% on the costs for FPGAs.

Again, that is only possible if your design is relatively slow and doesn't utilize modern high-speed interfaces or features. And you also pay for that by increased complexity of entire design (== higher R&D cost), longer time to market, and lower reliability (more chips == more things to fail == lower reliability). If your design uses any more-or-less modern features like fast DDR memory, you are out of luck.

Many of the modern lower cost FPGAs have DDR memory and high speed interfaces nowadays. It wouldn't make sense to bring a new product to market that doesn't have high speed interfaces. I'm not talking about substituting a modern FPGA with an old CPLD device... 

[asmi]

Many of the modern lower cost FPGAs have DDR memory and high speed interfaces nowadays. It wouldn't make sense to bring a new product to market that doesn't have high speed interfaces. I'm not talking about substituting a modern FPGA with an old CPLD device... 

Well which "lower cost" FPGA has DDR4 interface? Or even DDR3 that works above "pedestrian" 400 MHz? I know of none. The cheapest FPGA I know of which is capable of such feat is good old Kintex-7 K70.
Also each vendor has it's own DDR memory controler with interface that is not compatible with any other vendor's interface. So "transplanting" design from one vendor to another is quite expensive and long undertaking. The cost of controller itself is also a factor, though for commercial projects it's usually overshadowed by the R&D cost to re-target your design from other vendor's controller.
But DDR is not the worst of it, as slower interface speed sometimes can be compensated for to some degree by utilizing wider bus. When it comes to high-speed serial links, that is a real horror story, as most other vendors require commercial license to even use those transceivers, not to mention high cost of any relevant IPs (like HDMI, DisplayPort, PCI Express, 10/25/40/100G Ethernet, etc.), and incompatible user interfaces. Even when it comes to relatively simple LVDS links, 7 series can achieve over 1.25 Gbps on a single pair as per spec, which is far above and beyond what their competitors can, most of which top out at 800 Mbps. Again, it can be worked around with by using wider bus, but it's not always possible - like for example when you interface with external devices using industry-standard protocols like LVDS video, or HDMI.
Lastly, I've been looking at some point for a CPU SoC with publicly available manuals which would offer a high-speed connection that can be used to connect to FPGA, but the best I was able to find back then was a single PCI Express 2.0 link with only 5 Gpbs of theoretical bandwidth, which was nowhere near enough for my requirements at a time. Not sure if things are any better now, but would be interesting to know if it would be cheaper than using overpriced MPSoCs.

[gnuarm]

If Gowin stops, just move to a different vendor.

LOL  You make it sound like that is trivial.  It would require a new layout of the board, which means a lot of expensive testing has to be redone. 

Just make sure that you can move your project to a different vendor easely.

That's like when a manual tells you to do something "carefully", whatever the heck that means.  There ain't no moving to a different FPGA vendor "easily". 

There is a lot of choice nowadays triggered by the opensource synthesis tools which takes away the huge investment to write synthesis tools. The devices from  Cologne Chip look very interesting and they are based in Germany.

I looked at them and found nothing of interest.  Again, my customer would almost certainly not want me to use that part.  I'm thinking if they reject the Gowin devices, I will give them a Xilinx part number in a 196 pin, 1 mm spaced BGA and ask them to supply the parts.  They have talked more than once about their ability to throw their weight around to get what they need.  I can't even get samples.

One of my customers is basically dead in the water due to Xilinx not delivering chips. That is way worse than having to depent on a Chinese manufacturer. And I strongly doubt Chinese manufacturers are going anywhere. In fact, the chips from Chinese manufacturers seem to be least affected by chip shortage so they look like a very safe bet to me when the sh!t hits the fan. Gowin -for example- seems to have their products manufactured in Taiwan by TSMC. The chances of anyone or any country messing with TSMC are zero.

There are many concerns about buying from Chinese vendors.  Buying from most of the far East is not so problematic, but China can be cut off for political reasons.  Then there's also the issue of buying parts that will go into systems bought by the US Government.  Anyone heard of Huawei?

[gnuarm]

If Gowin stops, just move to a different vendor. Just make sure that you can move your project to a different vendor easely. There is a lot of choice nowadays triggered by the opensource synthesis tools which takes away the huge investment to write synthesis tools. The devices from  Cologne Chip look very interesting and they are based in Germany.

That is only possible if your designs are relatively simple. If your design requires DDR3/4 memory, uses multi-GB serial links, or PCIE, or other advanced features, forget about vendor-neutrality. And no opensource tools will help you with that.

It's not possible at all.  At minimum it requires a board spin, which requires new EMI testing, etc.   No, a part has to be picked up front that will be available.  In fact, I've been told to expect a 10 year lifetime of the product and to pick parts appropriately. 

[nctnico]

They have talked more than once about their ability to throw their weight around to get what they need.  I can't even get samples.

Good luck with that. I've heard that story more than once but have not seen any parts appear out of thin air. The assembler I'm using also does a lot of work for a major supplier of electric distribution parts. It doesn't help. Heck, not even the major car manufacturers can get parts. All you do while waiting for Xilinx parts to appear is fooling around and wasting time wishing life is better. If you want to move products out of the door in this new world, then you need to adapt and get work done. The sooner you realise that, the sooner you can make money.

And no, EMC testing is not an issue where it comes to replacing a part that has zero incluence on EMC behaviour. A quick test at most and off you go. Unless ofcourse your layout guy is grossly incompetent. On top of that, a product that has been designed with replacing parts in mind at some point is much easier to modify later on compared to a design that is tightly knit together. Components can become obsolete even though promises have been made for long term support. Recently a module that one of my customers demanded to use in a design got obsoleted without even the chance of doing a last time buy. I always strongly object to using modules BTW.

There are many concerns about buying from Chinese vendors.  Buying from most of the far East is not so problematic, but China can be cut off for political reasons.  Then there's also the issue of buying parts that will go into systems bought by the US Government.  Anyone heard of Huawei?

And how about all the test equipment that is made in China? Can't you use that for development either?

[gnuarm]

They have talked more than once about their ability to throw their weight around to get what they need.  I can't even get samples.

Good luck with that. I've heard that story more than once but have not seen any parts appear out of thin air. The assembler I'm using also does a lot of work for a major supplier of electric distribution parts. It doesn't help. Heck, not even the major car manufacturers can get parts. All you do while waiting for Xilinx parts to appear is fooling around and wasting time wishing life is better. If you want to move products out of the door in this new world, then you need to adapt and get work done. The sooner you realise that, the sooner you can make money.

And no, EMC testing is not an issue where it comes to replacing a part that has zero incluence on EMC behaviour.

That's absurd.  It's not the part that emits any real EMI, it's the board.  To change the FPGA is a major redo of the board layout, likely with different power supplies as well.  It would be up to me to certify that nothing significant changed, and there's no way I'm doing that.  If they don't want to retest, that's their problem since it's in their product anyway. 

But just to be clear, I won't have a choice.  If they say no Gowin parts, I pick something else.  So this discussion is of no value.

A quick test at most and off you go. Unless ofcourse your layout guy is grossly incompetent.

I don't know what to say to you.  That's not how EMI works.

On top of that, a product that has been designed with replacing parts in mind at some point is much easier to modify later on compared to a design that is tightly knit together.

Tightly knit???  That's of no value in this discussion since you know nothing of the design.

Components can become obsolete even though promises have been made for long term support. Recently a module that one of my customers demanded to use in a design got obsoleted without even the chance of doing a last time buy. I always strongly object to using modules BTW.

Sure, there are no guarantees in life.  But we have to go through the motions and get what assurances we can.  I'm pretty sure Xilinx and Altera have an idea of which sub families are going to be supported longer term.  It's not like they never think of this.  Heck, you can still buy Spartan 3 parts.  They said many years ago, to design in "this flavor" if you want to make boards long term.

There are many concerns about buying from Chinese vendors.  Buying from most of the far East is not so problematic, but China can be cut off for political reasons.  Then there's also the issue of buying parts that will go into systems bought by the US Government.  Anyone heard of Huawei?

And how about all the test equipment that is made in China? Can't you use that for development either?

You aren't making sense.  Maybe you have not read the many reports of spying on networks by Huawei equipment? 

[asmi]

To whom it may concern - there are quite a bunch (312 at the time of publishing) of XC7S15-2CSGA225's in stock on Mouser right now: https://www.mouser.ca/ProductDetail/Xilinx/XC7S15-2CSGA225I?qs=unwgFEO1A6ttMK67MnC55Q%3D%3D Grab some if you need them!
This might actually be a good replacement for device in the OP - it's slightly smaller (13x13 mm) and has the same 100 user IO pins.

[nctnico]

They have talked more than once about their ability to throw their weight around to get what they need.  I can't even get samples.

Good luck with that. I've heard that story more than once but have not seen any parts appear out of thin air. The assembler I'm using also does a lot of work for a major supplier of electric distribution parts. It doesn't help. Heck, not even the major car manufacturers can get parts. All you do while waiting for Xilinx parts to appear is fooling around and wasting time wishing life is better. If you want to move products out of the door in this new world, then you need to adapt and get work done. The sooner you realise that, the sooner you can make money.

And no, EMC testing is not an issue where it comes to replacing a part that has zero incluence on EMC behaviour.

That's absurd.  It's not the part that emits any real EMI, it's the board.  To change the FPGA is a major redo of the board layout, likely with different power supplies as well.  It would be up to me to certify that nothing significant changed, and there's no way I'm doing that.  If they don't want to retest, that's their problem since it's in their product anyway. 

But just to be clear, I won't have a choice.  If they say no Gowin parts, I pick something else.  So this discussion is of no value.

A quick test at most and off you go. Unless ofcourse your layout guy is grossly incompetent.

I don't know what to say to you.  That's not how EMI works.

On top of that, a product that has been designed with replacing parts in mind at some point is much easier to modify later on compared to a design that is tightly knit together.

Tightly knit???  That's of no value in this discussion since you know nothing of the design.

Sorry, but these are really poor arguments. An FPGA is sitting somewhere on a circuit board. Typically surrounded by other compoments that deal with I/O and outside world signals. If you leave room on the board to use a different component, it is not much of a problem to switch to a different component. This is PCB design 101 and I have done that in several occasions. However, if you pack the components together so they are tightly knit together, you'll lose any flexibility to alter the design.

Where it comes to EMI: Typically you'll have at least a 4 layer board with a solid ground plane + power planes if you use an FPGA. Each signal to and from the FPGA can have an extremely small loop area because the return path (ground or power plane) is right under the trace. So unless you set the drive strength to maximum and/ or route traces over splits in a plane and/or have high speed signals going into external cables, it is safe to say that a different FPGA won't affect EMI in any way. One of the reasons FPGAs have adjustable drive strengths is to help reduce potential EMI issues. I'm not saying not to re-test but the re-test can be a very quick (and thus cheap) measurement. Nothing to sweat about there. From my experience you really need to screw up the PCB design of a board to make the board itself radiate too much. Where it comes to EMC compliance, wiring is way more problematic.

Power supplies is also straightforward. The FPGA you are after is not a super high-end device so currents are low. If you keep the power supplies close to the FPGA, you can replace these if necessary but likely you can keep the same ones set to slightly different voltages. Bonus points for designing the power supply solution with some flexibility in mind.

You aren't making sense.  Maybe you have not read the many reports of spying on networks by Huawei equipment?

This is borderline thin-foil hat conspiracy theorism. Explain how you'd go about spying through a low end FPGA in a random design? At some point you just have to be realistic about the actual risks.

[gnuarm]

They have talked more than once about their ability to throw their weight around to get what they need.  I can't even get samples.

Good luck with that. I've heard that story more than once but have not seen any parts appear out of thin air. The assembler I'm using also does a lot of work for a major supplier of electric distribution parts. It doesn't help. Heck, not even the major car manufacturers can get parts. All you do while waiting for Xilinx parts to appear is fooling around and wasting time wishing life is better. If you want to move products out of the door in this new world, then you need to adapt and get work done. The sooner you realise that, the sooner you can make money.

And no, EMC testing is not an issue where it comes to replacing a part that has zero incluence on EMC behaviour.

That's absurd.  It's not the part that emits any real EMI, it's the board.  To change the FPGA is a major redo of the board layout, likely with different power supplies as well.  It would be up to me to certify that nothing significant changed, and there's no way I'm doing that.  If they don't want to retest, that's their problem since it's in their product anyway. 

But just to be clear, I won't have a choice.  If they say no Gowin parts, I pick something else.  So this discussion is of no value.

A quick test at most and off you go. Unless ofcourse your layout guy is grossly incompetent.

I don't know what to say to you.  That's not how EMI works.

On top of that, a product that has been designed with replacing parts in mind at some point is much easier to modify later on compared to a design that is tightly knit together.

Tightly knit???  That's of no value in this discussion since you know nothing of the design.

Sorry, but these are really poor arguments. An FPGA is sitting somewhere on a circuit board. Typically surrounded by other compoments that deal with I/O and outside world signals. If you leave room on the board to use a different component, it is not much of a problem to switch to a different component. This is PCB design 101 and I have done that in several occasions. However, if you pack the components together so they are tightly knit together, you'll lose any flexibility to alter the design.

You know nothing of the design, yet you insist that you can tell me how to make it work with multiple different parts.

Why are you like this? 

I'm pretty sure I've explained that I can't replace the FPGA without the customer requiring retesting at several levels.  In fact, that's exactly what we are doing now.  Replacing the FPGA and a CODEC chip.  This now requires the board to be recertified.  YOUR experiences with YOUR customers is irrelevant to the matter. 

Where it comes to EMI: Typically you'll have at least a 4 layer board with a solid ground plane + power planes if you use an FPGA. Each signal to and from the FPGA can have an extremely small loop area because the return path (ground or power plane) is right under the trace. So unless you set the drive strength to maximum and/ or route traces over splits in a plane and/or have high speed signals going into external cables, it is safe to say that a different FPGA won't affect EMI in any way. One of the reasons FPGAs have adjustable drive strengths is to help reduce potential EMI issues. I'm not saying not to re-test but the re-test can be a very quick (and thus cheap) measurement. Nothing to sweat about there. From my experience you really need to screw up the PCB design of a board to make the board itself radiate too much. Where it comes to EMC compliance, wiring is way more problematic.

You are an engineer and seem to be only capable of thinking like an engineer.  The issues involved are simply that the traces on the board have changed.  NO ONE, including YOU, can say for certain that small changes in the routing won't alter the emissions enough to no longer be in compliance.

Power supplies is also straightforward. The FPGA you are after is not a super high-end device so currents are low. If you keep the power supplies close to the FPGA, you can replace these if necessary but likely you can keep the same ones set to slightly different voltages. Bonus points for designing the power supply solution with some flexibility in mind.

You aren't making sense.  Maybe you have not read the many reports of spying on networks by Huawei equipment?

This is borderline thin-foil hat conspiracy theorism. Explain how you'd go about spying through a low end FPGA in a random design? At some point you just have to be realistic about the actual risks.

Again, this is something you know nothing about.  There are experts in the field who know what can, and has been done in devices from China. 

Why do you keep talking as if you know things, that you clearly don't?