Lattice Announces EOL for XP and EC/P Product Lines

[rickman]

This is likely not a big deal to most, but it hurts me a lot. I have
one product in production and it uses an XP device. They are only
giving until November to get your last time buy orders in. I think
Lattice is doing a disservice to themselves as well as the rest of us.
I am very accustomed to extended longevity in FPGAs. This act on the
part of Lattice puts them in a separate camp I think.

I have been looking at the alternatives. The three distinguishing
issues are package, capacity and the need for external configuration
memory. The XP I was using is in a 100 pin QFP which is perfect for the
board, easy to assemble and works with 6/6 design rules and 12 mil hole
diameter. It has 3000 LUTs which are around 80% used and the internal
configuration Flash saves space on the tiny, cramped board.

Mostly the alternatives are other Lattice devices, but none are a
perfect fit. XP2, XO2 and the iCE40 line. The ones that come in the
same package don't have as many LUTs, only 2100 which would require
using a soft CPU to implement the slow functions in fewer LUTs. The
larger parts are in harder to use packages like 0.5 mm BGAs which need
very fine pitch design rules and small drills.

The Xilinx parts are interesting. Spartan 3 devices come in 100 QFPs
and have enough of the "right stuff" inside including multipliers which
I can use. But that external flash needs a spot on the board and I have
to use a 1.2 volt regulator for the core. The XP parts use an internal
regulator and run from 3.3 volts only. Xilinx has a rep for keeping
parts in production for a long, long time, but the S3 line came out in
2005, same as the XP line. Spartan 6 parts give a *lot* more
functionality, but I'd have to use a 256 pin 1.0 mm BGA *and* external
flash *and* the 1.2 volt supply *and* they are twice the price. Maybe
I'll talk to the disties. Maybe they can do something about the price
at least.

I have yet to check out the Altera line. I don't remember them having
anything I liked in a nice package. But that will be something to do
later today. I guess I should check out the Micro-Semi line as well.
It's been a while since I looked hard at their parts and, oh yeah, there
is the PSOC from Cypress. I don't think that was an option at the time
I did this design.

An interesting note is that the Spartan 3 parts came out the same year
as the XP line. Xilinx still sells multiple generations of parts older
than the Spartan 3, so I think it will be a while before they can get
around to obsoleting that line.

--

Rick

 

[Rob Gaddi]

On Tue, 30 Jul 2013 14:37:21 -0400
rickman <gnuarm@gmail.com> wrote:

An interesting note is that the Spartan 3 parts came out the same year
as the XP line. Xilinx still sells multiple generations of parts older
than the Spartan 3, so I think it will be a while before they can get
around to obsoleting that line.

Actually just had this conversation with my Xilinx people. They're not
recommending Spartan 3 for new designs, and are talking
(speculatively) about obsoleting it in 2018.

--
Rob Gaddi, Highland Technology -- www.highlandtechnology.com
Email address domain is currently out of order. See above to fix.

 

[GaborSzakacs]

Rob Gaddi wrote:

On Tue, 30 Jul 2013 14:37:21 -0400
rickman <gnuarm@gmail.com> wrote:

An interesting note is that the Spartan 3 parts came out the same year
as the XP line. Xilinx still sells multiple generations of parts older
than the Spartan 3, so I think it will be a while before they can get
around to obsoleting that line.

Actually just had this conversation with my Xilinx people. They're not
recommending Spartan 3 for new designs, and are talking
(speculatively) about obsoleting it in 2018.

And yet, Xilinx recently updated the Spartan 3 data sheet to remove

the "not recommended for new designs" banner and indicated that they
in fact *are* recommended for new designs.

Still all of these manufacturers are at the mercy of their foundries
and have to pull the plug on devices that can no longer be manufactured
due to the process going away at UMC, TSMC, ...

At this point it's hard to say whether the FPGA manufacturer's previous
track record on supporting old devices is any indication of future
performance.

Another point on Xilinx parts in small packages - I seem to remember
that Lattice gave you more usable IO in the same package / pin count
than Xilinx. So the fact that you could get a Spartan 3 in a TQ100
doesn't necessarily mean it will have enough IO to replace the Lattice
XP device.

The other obvious options are:

1) Try to estimate your future usage of this part and schedule that LTB.

2) Stick you head in the sand and deal with the grey market for parts
until you can't get any more, then redesign. (This seems to be the
approved method here)

--
Gabor

 

[Rob Gaddi]

On Tue, 30 Jul 2013 15:03:54 -0400
GaborSzakacs <gabor@alacron.com> wrote:

Rob Gaddi wrote:
On Tue, 30 Jul 2013 14:37:21 -0400
rickman <gnuarm@gmail.com> wrote:

An interesting note is that the Spartan 3 parts came out the same year
as the XP line. Xilinx still sells multiple generations of parts older
than the Spartan 3, so I think it will be a while before they can get
around to obsoleting that line.

Actually just had this conversation with my Xilinx people. They're not
recommending Spartan 3 for new designs, and are talking
(speculatively) about obsoleting it in 2018.

And yet, Xilinx recently updated the Spartan 3 data sheet to remove
the "not recommended for new designs" banner and indicated that they
in fact *are* recommended for new designs.

Still all of these manufacturers are at the mercy of their foundries
and have to pull the plug on devices that can no longer be manufactured
due to the process going away at UMC, TSMC, ...

At this point it's hard to say whether the FPGA manufacturer's previous
track record on supporting old devices is any indication of future
performance.

Another point on Xilinx parts in small packages - I seem to remember
that Lattice gave you more usable IO in the same package / pin count
than Xilinx. So the fact that you could get a Spartan 3 in a TQ100
doesn't necessarily mean it will have enough IO to replace the Lattice
XP device.

The other obvious options are:

1) Try to estimate your future usage of this part and schedule that LTB.

2) Stick you head in the sand and deal with the grey market for parts
until you can't get any more, then redesign. (This seems to be the
approved method here)

--
Gabor

The thing I'm finding really concerning with Xilinx at the moment is
that they've got this big investment in yet another entirely new
toolchain (Vivado), and they're saying it's the way of the future. And
it doesn't even support Spartan 6, let alone anything older.

I switched years ago from X to A when my continuing problems with ISE
finally became too much to deal with. I applauded the decision to
scrap ISE's dodgy old codebase and take a new crack at it. But if the
software they're pushing going forward doesn't support a given chip,
then I can't possibly consider that chip to be going forward with them.

--
Rob Gaddi, Highland Technology -- www.highlandtechnology.com
Email address domain is currently out of order. See above to fix.

 

[Jon Elson]

rickman wrote:

The Xilinx parts are interesting. Spartan 3 devices come in 100 QFPs
and have enough of the "right stuff" inside including multipliers which
I can use. But that external flash needs a spot on the board and I have
to use a 1.2 volt regulator for the core. The XP parts use an internal
regulator and run from 3.3 volts only. Xilinx has a rep for keeping
parts in production for a long, long time, but the S3 line came out in
2005, same as the XP line. Spartan 6 parts give a *lot* more
functionality, but I'd have to use a 256 pin 1.0 mm BGA *and* external
flash *and* the 1.2 volt supply *and* they are twice the price. Maybe
I'll talk to the disties. Maybe they can do something about the price
at least.

Spartan 3AN has internal flash. I don't recall if there is a 100-pin

version, I am using the 144-pin version in a couple products.
I refuse to go to BGAs until there are no leaded parts remaining
available.

Jon

 

[GaborSzakacs]

Jon Elson wrote:

rickman wrote:


The Xilinx parts are interesting. Spartan 3 devices come in 100 QFPs
and have enough of the "right stuff" inside including multipliers which
I can use. But that external flash needs a spot on the board and I have
to use a 1.2 volt regulator for the core. The XP parts use an internal
regulator and run from 3.3 volts only. Xilinx has a rep for keeping
parts in production for a long, long time, but the S3 line came out in
2005, same as the XP line. Spartan 6 parts give a *lot* more
functionality, but I'd have to use a 256 pin 1.0 mm BGA *and* external
flash *and* the 1.2 volt supply *and* they are twice the price. Maybe
I'll talk to the disties. Maybe they can do something about the price
at least.

Spartan 3AN has internal flash. I don't recall if there is a 100-pin
version, I am using the 144-pin version in a couple products.
I refuse to go to BGAs until there are no leaded parts remaining
available.

Jon

I'm pretty sure that the 144-pin package is the smallest with flash.
In any case it's not a big win over an external SPI flash part. The
difference in footprint between 100 TQFP and 144 TQFP is more than
the flash footprint. Not to mention there's a price premium for that
multi-die package.

--
Gabor

 

On Tuesday, July 30, 2013 9:37:21 PM UTC+3, rickman wrote:

This is likely not a big deal to most, but it hurts me a lot. I have

one product in production and it uses an XP device. They are only

giving until November to get your last time buy orders in. I think

Lattice is doing a disservice to themselves as well as the rest of us.

I am very accustomed to extended longevity in FPGAs. This act on the

part of Lattice puts them in a separate camp I think.



I have been looking at the alternatives. The three distinguishing

issues are package, capacity and the need for external configuration

memory. The XP I was using is in a 100 pin QFP which is perfect for the

board, easy to assemble and works with 6/6 design rules and 12 mil hole

diameter. It has 3000 LUTs which are around 80% used and the internal

configuration Flash saves space on the tiny, cramped board.



Mostly the alternatives are other Lattice devices, but none are a

perfect fit. XP2, XO2 and the iCE40 line. The ones that come in the

same package don't have as many LUTs, only 2100 which would require

using a soft CPU to implement the slow functions in fewer LUTs. The

larger parts are in harder to use packages like 0.5 mm BGAs which need

very fine pitch design rules and small drills.



The Xilinx parts are interesting. Spartan 3 devices come in 100 QFPs

and have enough of the "right stuff" inside including multipliers which

I can use. But that external flash needs a spot on the board and I have

to use a 1.2 volt regulator for the core. The XP parts use an internal

regulator and run from 3.3 volts only. Xilinx has a rep for keeping

parts in production for a long, long time, but the S3 line came out in

2005, same as the XP line. Spartan 6 parts give a *lot* more

functionality, but I'd have to use a 256 pin 1.0 mm BGA *and* external

flash *and* the 1.2 volt supply *and* they are twice the price. Maybe

I'll talk to the disties. Maybe they can do something about the price

at least.



I have yet to check out the Altera line. I don't remember them having
anything I liked in a nice package.

If "nice" = 100 pin QFP, then yes, except for ancient Cyclone-I, Altera does not have anything nice.
But if 144 pin QFP is also o.k. then there are relatively modern Cyclone III devices. Voltage and the rest is more or less the same as Xilinx.

MAX2/MAX5 are not for you - too few LUTs.

But that will be something to do

later today. I guess I should check out the Micro-Semi line as well.

It's been a while since I looked hard at their parts and, oh yeah, there

is the PSOC from Cypress. I don't think that was an option at the time

I did this design.



An interesting note is that the Spartan 3 parts came out the same year

as the XP line. Xilinx still sells multiple generations of parts older

than the Spartan 3, so I think it will be a while before they can get

around to obsoleting that line.



--



Rick

 

[Jon Elson]

GaborSzakacs wrote:

I'm pretty sure that the 144-pin package is the smallest with flash.
In any case it's not a big win over an external SPI flash part. The
difference in footprint between 100 TQFP and 144 TQFP is more than
the flash footprint. Not to mention there's a price premium for that
multi-die package.

Right, unless I have a pretty strong reason to use the 3AN, I use

the Spartan 3, and the SST flash chips, which are insanely cheap.
I wrote my own programmer code for those. Spartan 2E needed some
interface fooling around to command the memory to start dumping at
location zero, but the 3A knows how to do it by setting some config
pins.

Jon

 

[rickman]

On 7/30/2013 3:03 PM, GaborSzakacs wrote:

Rob Gaddi wrote:
On Tue, 30 Jul 2013 14:37:21 -0400
rickman <gnuarm@gmail.com> wrote:

An interesting note is that the Spartan 3 parts came out the same
year as the XP line. Xilinx still sells multiple generations of parts
older than the Spartan 3, so I think it will be a while before they
can get around to obsoleting that line.

Actually just had this conversation with my Xilinx people. They're not
recommending Spartan 3 for new designs, and are talking
(speculatively) about obsoleting it in 2018.

And yet, Xilinx recently updated the Spartan 3 data sheet to remove
the "not recommended for new designs" banner and indicated that they
in fact *are* recommended for new designs.

Still all of these manufacturers are at the mercy of their foundries
and have to pull the plug on devices that can no longer be manufactured
due to the process going away at UMC, TSMC, ...

At this point it's hard to say whether the FPGA manufacturer's previous
track record on supporting old devices is any indication of future
performance.

Before making any decisions I will do my due diligence as well as have
any decision approved by my customer. They will be designing my board
into their new product, so they are free to make the decision for me.
Actually, 2018 might work for me if not for my customer. I expect I'll
be fully retired in 5 more years.

Another point on Xilinx parts in small packages - I seem to remember
that Lattice gave you more usable IO in the same package / pin count
than Xilinx. So the fact that you could get a Spartan 3 in a TQ100
doesn't necessarily mean it will have enough IO to replace the Lattice
XP device.

In this case the count is higher in the Spartan 3 part and is a *lot*
higher in nearly any other part since most won't be in the same package.

The other obvious options are:

1) Try to estimate your future usage of this part and schedule that LTB.

Not mine to estimate. I tried buying just 10 boards ahead and ended up
with 10 rev 1.1 boards after we did the 2.0 design. The demand is
*very* lumpy as my customer puts it. We got orders this year for more
units than we have sold in the last five...

2) Stick you head in the sand and deal with the grey market for parts
until you can't get any more, then redesign. (This seems to be the
approved method here)

No, this is too important to me any my customer. We will work it out
one way or the other. Thanks for your comments.

--

Rick

 

[rickman]

On 7/30/2013 5:11 PM, Theo Markettos wrote:

already5chosen@yahoo.com wrote:
If "nice" = 100 pin QFP, then yes, except for ancient Cyclone-I, Altera
does not have anything nice.

But if 144 pin QFP is also o.k. then there are relatively modern Cyclone
III devices. Voltage and the rest is more or less the same as Xilinx.

There's some Cyclone IVs in 144ish QFP too.

I would love to use a 144 pin QFP, but they don't fit on the board. It
is a very narrow daughtercard, only 0.85" wide and the 144 QFP is 0.86"
wide without counting the pad over hang. A 17x17mm 256 BGA is the best
of the rest so far. It has a 1 mm pitch although I haven't looked at
what it takes to get vias between the balls. I expect I will need to be
willing to push below the comfort zone of 6/6 mil space/trace and maybe
below 10 mil drill.

One nice thing about the 256 BGA is that I only need to reach the power
and ground pins and the outer ring or two to get my 62 I/Os. I can
ignore the rest of the balls.

--

Rick

 

[rickman]

On 7/30/2013 3:50 PM, Jon Elson wrote:

rickman wrote:


The Xilinx parts are interesting. Spartan 3 devices come in 100 QFPs
and have enough of the "right stuff" inside including multipliers which
I can use. But that external flash needs a spot on the board and I have
to use a 1.2 volt regulator for the core. The XP parts use an internal
regulator and run from 3.3 volts only. Xilinx has a rep for keeping
parts in production for a long, long time, but the S3 line came out in
2005, same as the XP line. Spartan 6 parts give a *lot* more
functionality, but I'd have to use a 256 pin 1.0 mm BGA *and* external
flash *and* the 1.2 volt supply *and* they are twice the price. Maybe
I'll talk to the disties. Maybe they can do something about the price
at least.

Spartan 3AN has internal flash. I don't recall if there is a 100-pin
version, I am using the 144-pin version in a couple products.
I refuse to go to BGAs until there are no leaded parts remaining
available.

No S3AN in 100 pin QFP, this may be because it is a dual die
configuration or more likely they just don't expect to sell a lot of
this size. I haven't bothered to check the price either. They do have
a FTG256 which might do the job. I expect I can find the space for a
very small 8 pin flash part if I go QFN. Again, I need to dig more.
I've just started looking...

Actually my main concern with the external flash is the whole JTAG
programming at the factory and/or lab thing, but there are likely many
ways to deal with that including having them programmed before assembly.
It is just that I've been using these Flash FPGAs for some time now
and I'm very used to them.

--

Rick

 

[rickman]

On 7/30/2013 6:02 PM, Jon Elson wrote:

GaborSzakacs wrote:


I'm pretty sure that the 144-pin package is the smallest with flash.
In any case it's not a big win over an external SPI flash part. The
difference in footprint between 100 TQFP and 144 TQFP is more than
the flash footprint. Not to mention there's a price premium for that
multi-die package.

Right, unless I have a pretty strong reason to use the 3AN, I use
the Spartan 3, and the SST flash chips, which are insanely cheap.
I wrote my own programmer code for those. Spartan 2E needed some
interface fooling around to command the memory to start dumping at
location zero, but the 3A knows how to do it by setting some config
pins.

This is the stuff I'm concerned about with the external flash. But it
is just a mater of figuring out how to do it and then doing that... Too
bad I don't have space for an MCU. The design could use one, but if it
happens it will be *inside* the FPGA. Just no board space at all.

In fact, I'm skipping Altera for the moment and skipping over to
MicroSemi and Cypress to see if their combination CPU/Logic devices
might do the job well and let me eliminate the stereo CODEC to (another
part that could go obsolete at any time). I seem to recall that the
Cypress part might be just the ticket but the MicroSemi part runs some
$50 at the low point. The current Lattice part is running under $10.

Although, if the CODEC could be absorbed inside an MCU, there is some
board space at that point, about 6.5mm sq. But how many MCUs can do 16
bit, CD quality audio on two channels? Maybe Analog Devices, but their
MCU parts don't give me confidence in their longevity. It also has to
read the external flash and dump it into the FPGA so those tiny audio
DSPs are out.

--

Rick

 

[Theo Markettos]

already5chosen@yahoo.com wrote:

If "nice" = 100 pin QFP, then yes, except for ancient Cyclone-I, Altera
does not have anything nice.

But if 144 pin QFP is also o.k. then there are relatively modern Cyclone
III devices. Voltage and the rest is more or less the same as Xilinx.

There's some Cyclone IVs in 144ish QFP too.

Theo

 

[Rob Doyle]

On 7/30/2013 11:37 AM, rickman wrote:

Spartan 6 parts give a *lot* more functionality, but I'd have to use
a 256 pin 1.0 mm BGA *and* external flash *and* the 1.2 volt supply
*and* they are twice the price. Maybe I'll talk to the disties.
Maybe they can do something about the price at least.

The smaller Spartan 6 parts do come in a 144 pin TQFP package. Too small?

Rob.

 

On Wednesday, July 31, 2013 12:40:20 AM UTC+2, rickman wrote:

On 7/30/2013 5:11 PM, Theo Markettos wrote:

already5chosen@yahoo.com wrote:

If "nice" = 100 pin QFP, then yes, except for ancient Cyclone-I, Altera

does not have anything nice.



But if 144 pin QFP is also o.k. then there are relatively modern Cyclone

III devices. Voltage and the rest is more or less the same as Xilinx.



There's some Cyclone IVs in 144ish QFP too.



I would love to use a 144 pin QFP, but they don't fit on the board. It

is a very narrow daughtercard, only 0.85" wide and the 144 QFP is 0.86"

wide without counting the pad over hang. A 17x17mm 256 BGA is the best

of the rest so far. It has a 1 mm pitch although I haven't looked at

what it takes to get vias between the balls. I expect I will need to be

willing to push below the comfort zone of 6/6 mil space/trace and maybe

below 10 mil drill.

http://bmc.bu.edu/bmc/asd/tester/xapp157.pdf

says 6/6 and 12mil drill is doable


-Lasse

 

In fact, I'm skipping Altera for the moment

If you have some reasonable time that you can bridge with your LTB parts, I think you should really contact your local Altera FAE to see if he has some news for you.

Personally, I like Quartus the most by far when comparing it with the software from the other vendors (just look at SignalTap, for example). (I have no Vivado experience, however...)

Regards,

Thomas

www.entner-electronics.com

 

[Jon Elson]

rickman wrote:

No S3AN in 100 pin QFP, this may be because it is a dual die
configuration or more likely they just don't expect to sell a lot of
this size. I haven't bothered to check the price either. They do have
a FTG256 which might do the job. I expect I can find the space for a
very small 8 pin flash part if I go QFN. Again, I need to dig more.
I've just started looking...

The smallest Spartan 3A is under $10, the 3AN is about $15.

Actually my main concern with the external flash is the whole JTAG
programming at the factory and/or lab thing, but there are likely many
ways to deal with that including having them programmed before assembly.
It is just that I've been using these Flash FPGAs for some time now
and I'm very used to them.

The 3AN can be programmed by JTAG, the SST serial EPROM I use on

the 3A is not JTAG, although a flexible programmer or tester
could easily be "taught" the protocol. I chose this device
so I could have field-replaceable firmware. I had to make an
SO-8 to DIP converter board though, as the SST chip is only
available in a couple SMT packages.

Jon

 

[Gabor]

On 7/30/2013 9:03 PM, Rob Doyle wrote:

On 7/30/2013 11:37 AM, rickman wrote:
Spartan 6 parts give a *lot* more functionality, but I'd have to use
a 256 pin 1.0 mm BGA *and* external flash *and* the 1.2 volt supply
*and* they are twice the price. Maybe I'll talk to the disties.
Maybe they can do something about the price at least.

The smaller Spartan 6 parts do come in a 144 pin TQFP package. Too small?

Rob.

Apparently the 144 TQ package is too big (physically). And once you
look at a 256-ball 1mm BGA you could find any number of devices
including those from Lattice (XP2?). For internals, the smallest
Spartan 6 is about the size of the original XP part he was using.

As to price, we never pay anything near list for Xilinx parts, but
we don't get the same steep discount on Spartan 6 as we do on other
series.

--
Gabor

 

[rickman]

On 7/30/2013 7:44 PM, langwadt@fonz.dk wrote:

On Wednesday, July 31, 2013 12:40:20 AM UTC+2, rickman wrote:
On 7/30/2013 5:11 PM, Theo Markettos wrote:

already5chosen@yahoo.com wrote:

If "nice" = 100 pin QFP, then yes, except for ancient Cyclone-I, Altera

does not have anything nice.



But if 144 pin QFP is also o.k. then there are relatively modern Cyclone

III devices. Voltage and the rest is more or less the same as Xilinx.



There's some Cyclone IVs in 144ish QFP too.



I would love to use a 144 pin QFP, but they don't fit on the board. It

is a very narrow daughtercard, only 0.85" wide and the 144 QFP is 0.86"

wide without counting the pad over hang. A 17x17mm 256 BGA is the best

of the rest so far. It has a 1 mm pitch although I haven't looked at

what it takes to get vias between the balls. I expect I will need to be

willing to push below the comfort zone of 6/6 mil space/trace and maybe

below 10 mil drill.


http://bmc.bu.edu/bmc/asd/tester/xapp157.pdf

says 6/6 and 12mil drill is doable


-Lasse

Thanks for the link, that saved me a little bother. But I'll still need
to run through the calcs myself. A quick check says this works. Still,
I'm not crazy about using a 256 ball part when I only need 62 I/O... lol
Better than no choice at all.

--

Rick

 

[rickman]

On 7/30/2013 9:50 PM, Jon Elson wrote:

rickman wrote:


No S3AN in 100 pin QFP, this may be because it is a dual die
configuration or more likely they just don't expect to sell a lot of
this size. I haven't bothered to check the price either. They do have
a FTG256 which might do the job. I expect I can find the space for a
very small 8 pin flash part if I go QFN. Again, I need to dig more.
I've just started looking...

The smallest Spartan 3A is under $10, the 3AN is about $15.

Actually my main concern with the external flash is the whole JTAG
programming at the factory and/or lab thing, but there are likely many
ways to deal with that including having them programmed before assembly.
It is just that I've been using these Flash FPGAs for some time now
and I'm very used to them.

The 3AN can be programmed by JTAG, the SST serial EPROM I use on
the 3A is not JTAG, although a flexible programmer or tester
could easily be "taught" the protocol. I chose this device
so I could have field-replaceable firmware. I had to make an
SO-8 to DIP converter board though, as the SST chip is only
available in a couple SMT packages.

Yeah, I am supposed to provide JTAG programmability through the
equipment this daughtercard is plugged into. That is, I provide the
JTAG port, it is up to them to do the software to program it. That is
one of my concerns with an external seral prom. May not be easy to do
in an 8 pin package... But if they can learn the protocol, maybe that
would work too. We'll see. After 5 years we still have not required
this functionality. But I'll be pushing for increased capability in the
new version to allow it to be sold into new areas. So remote updates
may be more important then.

--

Rick