PCB Layout for BGAs

On Wednesday, January 11, 2023 at 5:26:48 AM UTC-5, David Brown wrote:
> On 10/01/2023 23:17, gnuarm.del...@gmail.com wrote: 
> > 
> > Yes, sorry, they do make a few multiplier chips with FPGA tiles. I 
> > was referring to parts that I might be able to use. They have a 
> > couple of 8 kLUT parts, only one in a package that I could use. I 
> > can pick between a 0.8 mm ball pitch, or 0.65 mm. Not really excited 
> > about either, even though there's a bit of inventory of the 256 ball, 
> > 0.8 mm part. But no insight into future deliveries. 
> > 
> > This looking for usable parts gets old fast, and I've been doing it 
> > for over a year now. When I find the guy responsible for this 
> > shortage, I'm going to give him a piece of my mind! 
> >
> The reason you can parts in high-density packages, but not low-density 
> packages, is that there are lots of people such as yourself who are so 
> reluctant to use the small pitch devices. (This is not criticism - you 
> have solid reasons for preferring larger pitch devices, as do many 
> others.) Big manufacturers often prefer smaller pitch and higher 
> density, as it can lead to lower overall costs for their products, even 
> if design is more costly and the pcbs are more expensive. 
> 
> There have been component supply issues for several years now, with only 
> gradual improvement in many areas. But there is a general pattern of 
> somewhat higher availability in smaller pitch parts.

The very fine pitch parts are used to save space on the board.  Some applications, like cell phones, simply require it.  Not sure it saves any money, really.  If you save on board size, you pay that back for finer pitch and laser drilled holes. 

-- 

Rick C.

-++ Get 1,000 miles of free Supercharging
-++ Tesla referral code - https://ts.la/richard11209

On Wednesday, January 11, 2023 at 5:42:23 AM UTC-5, David Brown wrote:
> On 10/01/2023 23:44, gnuarm.del...@gmail.com wrote: 
> > On Tuesday, January 10, 2023 at 1:35:24 PM UTC-5, Michael Schwingen 
> > wrote: 
> >> On 2023-01-09, gnuarm.del...@gmail.com <gnuarm.del...@gmail.com> 
> >> wrote: 
> >>> 
> >>> I'm concerned about adding cost for the boards, cost for the 
> >>> assembly and just an easy road forward. I spend the last two 
> >>> years building 8,000 units when the CODEC factory burnt down. The 
> >>> customer knows about this issue, but the previous CM turned flaky 
> >>> on me and all but stopped delivering product. 
> >>> 
> >>> I have a new CM, but I don't want to go through production 
> >>> problems again. 
> >> 0.8mm BGA should be no problem for any reputable CM - fine-pitch 
> >> QFP is usually more trouble. 
> > 
> > Part of my problem is a lack of having designed with BGAs before. I 
> > can find footprint recommendations, but they are different for every 
> > manufacturer. It didn't occur to me that this might be because even 
> > though they have the same pitch and ball count, they may not have the 
> > same ball size. 
> > 
> > The two primary choices right now are a 196 ball, 1.0 mm pitch and 
> > 256 ball, 0.8 mm pitch. Can you share the design rules you used for 
> > these parts? 
> >
> The board stackup, routing and bypassing recommendations from FPGA 
> manufacturers are basically bollocks. I believe it is primarily a 
> matter of being able to fob off complaints and support requests by 
> saying "Did you follow our layout application notes, impossible though 
> they may be? If not, it's not /our/ fault that you have problems." 
> 
> OK, that's a bit of an exaggeration, but you can ignore the suggestions 
> of 16 layers with 8 different power planes and a dozen different 
> capacitor sizes mounted directly below the device. 

I see the opposite.  When FPGA makers offer routing suggestions, they often provide one for routing of 100% of I/O pins, and another, using fewer layers, routing a portion of the I/O pins.  So clearly they are trying to optimize cost of the boards for the user.   No sign of CYA. 


> Yes, there are complications for BGA layouts. And I'm afraid you are 
> going to have to do some research, some learning, and some discussions 
> with both PCB manufacturers (or their proxies) and board builders. 
> 
> For the same pitch of BGA, there can be different sized balls, and 
> different sized pads on the underside of the BGA device which will 
> affect the shape of the ball after soldering. 

I haven't done a survey to check this yet.  Do you know this for a fact?  


> Pad size on the pcb has 
> different options. You have a key decision between solder mask defined 
> and non-solder mask defined pads, which affects mechanical strength, 
> thermal stability, solder paste masks, routeability, and manufacturing 
> requirements. And BGA soldering has different requirements in 
> production than non-BGA devices. 
> 
> 
> I have no doubt that this is something you can master quite quickly - 
> it's not /that/ hard. But it's not something you can learn just by a 
> thread on a newsgroup.

It's not "hard", it's "hard" to find the information for layout recommendations from each FPGA vendor.  I'm going to need to put together a compendium of layout information, before I can compare vendors.  The vendors may make it easy for me, based on availability and pricing.  Xilinx is not in the running unless I can get someone there to give assurance of better supply in six months.  Right now I'll have to buy every part in inventory of several combinations of speed and temperature, to build the order I have coming.  

-- 

Rick C.

+-- Get 1,000 miles of free Supercharging
+-- Tesla referral code - https://ts.la/richard11209

On 2023-01-10, gnuarm.del...@gmail.com <gnuarm.deletethisbit@gmail.com> wrote:
>
> Part of my problem is a lack of having designed with BGAs before.  I can
> find footprint recommendations, but they are different for every
> manufacturer.  It didn't  occur to me that this might be because even
> though they have the same pitch and ball count, they may not have the same
> ball size.

> The two primary choices right now are a 196 ball, 1.0 mm pitch and 256
> ball, 0.8 mm pitch.  Can you share the design rules you used for these
> parts?

I have a 529 pin BGA with 0.8mm pitch. SMD pads for the BGA are 0.4mm, vias
are also 0.4mm with 0.2mm drill. Using these rules, a via fits nicely
between 4 BGA pads.

I have plugged/plated vias in order to put 0402/0201 capacitors underneath
the BGA, but if you can place the capacitors outside the BGA area, normal
vias should do.

>> Talk to your PCB manufacturer about the details before doing the final 
>> layout - there is some fine tuning (eg. drill size, annular ring, spacing) 
>> where different PCB manufacturers have different preferences regarding which 
>> rules will yield good results - when doing do, 0.8mm BGA should be possible 
>> at modest PCB costs. 
>
> You mean my CM who orders the PWBs?  Yeah, I've tried asking before and
> they say they would need a design so they could get a quote.  I know, that
> sounds lame, but I used four different CMs over the last decade and they
> have all said the same thing.  They don't have design rules, that's for me
> to know.

OK, if you do not order the PCBs yourself, you have to forward this through
your CM. You will probably have to prepare a sample design (just the BGA
area with fanout), produce gerbers, and have them ask for feedback. Same
about the layer stackup if you need controlled impedances.

cu
Michael
-- 
Some people have no respect of age unless it is bottled.

On Wednesday, January 11, 2023 at 1:06:01 PM UTC-4, Michael Schwingen wrote=
:
> On 2023-01-10, gnuarm.del...@gmail.com <gnuarm.del...@gmail.com> wrote:=
=20
> >
> > Part of my problem is a lack of having designed with BGAs before. I can=
=20
> > find footprint recommendations, but they are different for every=20
> > manufacturer. It didn't occur to me that this might be because even=20
> > though they have the same pitch and ball count, they may not have the s=
ame=20
> > ball size.=20
>=20
> > The two primary choices right now are a 196 ball, 1.0 mm pitch and 256=
=20
> > ball, 0.8 mm pitch. Can you share the design rules you used for these=
=20
> > parts?
> I have a 529 pin BGA with 0.8mm pitch. SMD pads for the BGA are 0.4mm, vi=
as=20
> are also 0.4mm with 0.2mm drill. Using these rules, a via fits nicely=20
> between 4 BGA pads.=20

You would need to use 5 mil trace and space to get between the pads.  That =
doesn't sound too bad.  Via to pad is 6.5 mil, again good. =20

Where did you get your pad size numbers?  Your via pad only gives you 4 mil=
 annular ring.  That sounds a bit tight.  To make that a 5 mil annular ring=
 would shorten the 6.5 mil via to pad space to 5.5 mil, still good.  Why di=
d you choose a 0.4 mm pad?=20


> I have plugged/plated vias in order to put 0402/0201 capacitors underneat=
h=20
> the BGA, but if you can place the capacitors outside the BGA area, normal=
=20
> vias should do.
> >> Talk to your PCB manufacturer about the details before doing the final=
=20
> >> layout - there is some fine tuning (eg. drill size, annular ring, spac=
ing)=20
> >> where different PCB manufacturers have different preferences regarding=
 which=20
> >> rules will yield good results - when doing do, 0.8mm BGA should be pos=
sible=20
> >> at modest PCB costs.=20
> >=20
> > You mean my CM who orders the PWBs? Yeah, I've tried asking before and=
=20
> > they say they would need a design so they could get a quote. I know, th=
at=20
> > sounds lame, but I used four different CMs over the last decade and the=
y=20
> > have all said the same thing. They don't have design rules, that's for =
me=20
> > to know.
> OK, if you do not order the PCBs yourself, you have to forward this throu=
gh=20
> your CM. You will probably have to prepare a sample design (just the BGA=
=20
> area with fanout), produce gerbers, and have them ask for feedback. Same=
=20
> about the layer stackup if you need controlled impedances.

I asked my CM the general question of their BGA assembly experience and an =
estimated cost increment for going from a 100QFP to the 196 ball, 1.0 mm pi=
tch BGA and 256 ball, 0.8 mm BGA.  We'll see what they come up with.  If th=
ey can give me a dollar figure, they should be able to give me dimensions t=
hey are comfortable working with.=20

Thanks for discussing this with me.=20

--=20

Rick C.

+-+ Get 1,000 miles of free Supercharging
+-+ Tesla referral code - https://ts.la/richard11209

On 11/01/2023 13:49, gnuarm.del...@gmail.com wrote:
> On Wednesday, January 11, 2023 at 5:42:23 AM UTC-5, David Brown
> wrote:
>> On 10/01/2023 23:44, gnuarm.del...@gmail.com wrote:
>>> On Tuesday, January 10, 2023 at 1:35:24 PM UTC-5, Michael
>>> Schwingen wrote:
>>>> On 2023-01-09, gnuarm.del...@gmail.com
>>>> <gnuarm.del...@gmail.com> wrote:
>>>>> 
>>>>> I'm concerned about adding cost for the boards, cost for the 
>>>>> assembly and just an easy road forward. I spend the last two 
>>>>> years building 8,000 units when the CODEC factory burnt down.
>>>>> The customer knows about this issue, but the previous CM
>>>>> turned flaky on me and all but stopped delivering product.
>>>>> 
>>>>> I have a new CM, but I don't want to go through production 
>>>>> problems again.
>>>> 0.8mm BGA should be no problem for any reputable CM -
>>>> fine-pitch QFP is usually more trouble.
>>> 
>>> Part of my problem is a lack of having designed with BGAs before.
>>> I can find footprint recommendations, but they are different for
>>> every manufacturer. It didn't occur to me that this might be
>>> because even though they have the same pitch and ball count, they
>>> may not have the same ball size.
>>> 
>>> The two primary choices right now are a 196 ball, 1.0 mm pitch
>>> and 256 ball, 0.8 mm pitch. Can you share the design rules you
>>> used for these parts?
>>> 
>> The board stackup, routing and bypassing recommendations from FPGA 
>> manufacturers are basically bollocks. I believe it is primarily a 
>> matter of being able to fob off complaints and support requests by 
>> saying "Did you follow our layout application notes, impossible
>> though they may be? If not, it's not /our/ fault that you have
>> problems."
>> 
>> OK, that's a bit of an exaggeration, but you can ignore the
>> suggestions of 16 layers with 8 different power planes and a dozen
>> different capacitor sizes mounted directly below the device.
> 
> I see the opposite.  When FPGA makers offer routing suggestions, they
> often provide one for routing of 100% of I/O pins, and another, using
> fewer layers, routing a portion of the I/O pins.  So clearly they are
> trying to optimize cost of the boards for the user.   No sign of
> CYA.
> 

Fair enough.  Certainly you want to look at all the information you can 
here - you just have to be aware that some of it will be conflicting, 
and some of it will be overkill.  I read somewhere (a long time ago, and 
I've forgotten the details) of someone who initially made their design 
following application notes for bypass capacitors.  Then to save costs, 
they depopulated about 90% of these capacitors, basically at random. 
There were no measurable differences in signal integrity, EMC results, 
or any functionality.

> 
>> Yes, there are complications for BGA layouts. And I'm afraid you
>> are going to have to do some research, some learning, and some
>> discussions with both PCB manufacturers (or their proxies) and
>> board builders.
>> 
>> For the same pitch of BGA, there can be different sized balls, and 
>> different sized pads on the underside of the BGA device which will 
>> affect the shape of the ball after soldering.
> 
> I haven't done a survey to check this yet.  Do you know this for a
> fact?

Yes.

BGA balls are attached to circular pads on the underside of the BGA 
package, and the size of these pads can be different for different 
packages with the same pitch.  In general, you get the mechanically 
strongest bond when the pads on the pcb (or the opening in the solder 
mask, for solder mask defined pads) is the same size.  But that does not 
mean you /always/ want them to be the same as there are other factors in 
the trade-offs, and it's quite rare that mechanical strength is 
critical.  (If you are gluing on a large heatsink, without screws, and 
then mounting the board upside down in a high vibration environment, 
you'll have different requirements from a "normal" usage.)

> 
> 
>> Pad size on the pcb has different options. You have a key decision
>> between solder mask defined and non-solder mask defined pads, which
>> affects mechanical strength, thermal stability, solder paste masks,
>> routeability, and manufacturing requirements. And BGA soldering has
>> different requirements in production than non-BGA devices.
>> 
>> 
>> I have no doubt that this is something you can master quite quickly
>> - it's not /that/ hard. But it's not something you can learn just
>> by a thread on a newsgroup.
> 
> It's not "hard", it's "hard" to find the information for layout
> recommendations from each FPGA vendor.  I'm going to need to put
> together a compendium of layout information, before I can compare
> vendors.  The vendors may make it easy for me, based on availability
> and pricing.  Xilinx is not in the running unless I can get someone
> there to give assurance of better supply in six months.  Right now
> I'll have to buy every part in inventory of several combinations of
> speed and temperature, to build the order I have coming.
> 

That's the unfortunate reality these days.  Find out what you can get 
hold of, check if it looks good enough, then buy the stock.  There's no 
point in finding out that vendor X has good layout and manufacturing 
information, or vendor Y has good toolchains, if you can only get parts 
from vendor Z.  (This is not news to you, of course - I'm just 
sympathising.)

On 2023-01-12, gnuarm.del...@gmail.com <gnuarm.deletethisbit@gmail.com> wrote:
>> I have a 529 pin BGA with 0.8mm pitch. SMD pads for the BGA are 0.4mm, vias 
>> are also 0.4mm with 0.2mm drill. Using these rules, a via fits nicely 
>> between 4 BGA pads. 
>
> You would need to use 5 mil trace and space to get between the pads.  That
> doesn't sound too bad.  Via to pad is 6.5 mil, again good.

Trace width in the BGA area is 0.11mm (for data lines).

> Where did you get your pad size numbers?  Your via pad only gives you 4
> mil annular ring.  That sounds a bit tight.
> To make that a 5 mil annular
> ring would shorten the 6.5 mil via to pad space to 5.5 mil, still good. 
> Why did you choose a 0.4 mm pad?

That is the minimum given by our PCB manufacturer - small via pads allow for
bigger traces where needed (power traces, despite using a 8-layer PCB).

That is the area where you can fine tune after discussion with your PCB
manufacturer. Some may like a bigger annular ring, some may prefer smaller
ring and more pad-to-trace clearance. 

https://www.nxp.com/docs/en/package-information/PBGAPRES.pdf

has some information about the BGA pad design. Our BGA has 0.45mm pads on
the BGA side, so the 0.4mm pads are on the lower end of the recommended
range.

> I asked my CM the general question of their BGA assembly experience and an
> estimated cost increment for going from a 100QFP to the 196 ball, 1.0 mm
> pitch BGA and 256 ball, 0.8 mm BGA.  We'll see what they come up with.  If
> they can give me a dollar figure, they should be able to give me
> dimensions they are comfortable working with.

I would expect pick & place to be easier for the 0.8mm BGA than the TQFP.
Cost increase will probably happen at the PCB level (small annular ring, or
more expensive surface finish - TQFP may work with HASL, BGA needs a flatter
finish. However, ENIG is not that expensive nowadays.)

cu
Michael
-- 
Some people have no respect of age unless it is bottled.

On 12/01/2023 14:36, Michael Schwingen wrote:
> On 2023-01-12, gnuarm.del...@gmail.com <gnuarm.deletethisbit@gmail.com> wrote:

>> I asked my CM the general question of their BGA assembly experience and an
>> estimated cost increment for going from a 100QFP to the 196 ball, 1.0 mm
>> pitch BGA and 256 ball, 0.8 mm BGA.  We'll see what they come up with.  If
>> they can give me a dollar figure, they should be able to give me
>> dimensions they are comfortable working with.
> 
> I would expect pick & place to be easier for the 0.8mm BGA than the TQFP.
> Cost increase will probably happen at the PCB level (small annular ring, or
> more expensive surface finish - TQFP may work with HASL, BGA needs a flatter
> finish. However, ENIG is not that expensive nowadays.)
> 

Yes, BGAs can often be easier to place than TQFP's - you have a bigger 
pitch, and they "float" to the correct place even if there is a slight 
placement error.

On the other hand, you need better control of the soldering parameters, 
and they are harder if you have a board that has awkward heat flow - 
many high components nearby, or big thermal masses.  And it is harder to 
check connectivity and good quality soldering.

A good production facility will have tools to help here.  They will do 
the first boards with temperature probes between the balls, and X-Ray to 
check the quality of the soldering.  Make sure you have a production 
house that is not scared to give you feedback - many far eastern places 
will just do their best with what you give them, and never tell you how 
to improve your layout.

Re-work is, obviously, far more difficult with BGAs.

On Sat, 7 Jan 2023 09:49:24 -0800 (PST), "gnuarm.del...@gmail.com"
<gnuarm.deletethisbit@gmail.com> wrote:

>A small board with a 100QFP is being redesigned for a new FPGA due to obsolescence.  Gowin makes a 100QFP device that would be a good fit, but my customer has said "no" to the 100% Chinese brand... US government customers, ya know!  
>
>So now I'm looking at a BGA.  I don't want to get into fine PCB design rules, so 1.0 mm ball pitch is my preference.  The only devices I can find that fit on the board have 196 or 256 pins.  But the real problem is availability.  
>
>Digikey has a few of the XC7S15-1FTGB196I and more a scheduled for delivery in April.  Add in the various speed and temperature flavors trickling in (mostly in April) and I should be ok for the initial delivery in August... if I can get my hands on those.  I don't know if Digikey factors in the backlog orders in these counts.  
>
>Mouser shows great inventory of Efinix parts, particularly the T13 and T20 in a 0.8 mm 256 pin BGA, 10s of thousands in stock.  But I'd rather work with a 1.0 mm BGA.  Oddly enough, LCSC shows part numbers, but zero inventory. 
>
>Anyone work with 0.8 mm BGAs?  What PWB feature dimensions did you use?  Did this impact the PWB cost?  

The 0.8 mm 256-ball T20 isn't bad...

https://www.dropbox.com/s/xjqgj2pz9mdhtma/P941_FPGA.jpg?raw=1

The BGA pads are 16 mils. 8 mil drills on the BGA vias. 6 mil traces
mostly, except for the 50 ohm monsters. No big deal these days. Works
great.

We considered a T8 for a simpler application, but its 0.5 mm ball
pitch looked nasty.

The efinix tool chain looks like it was developed in someone's garage,
which is actually praise. It's free and simple and just works without
200 gbyte downloads and doing battle with FlexLM.

 

On Saturday, January 14, 2023 at 12:39:49 AM UTC-4, John Larkin wrote:
> On Sat, 7 Jan 2023 09:49:24 -0800 (PST), "gnuarm.del...@gmail.com"
> <gnuarm.del...@gmail.com> wrote:=20
>=20
> >A small board with a 100QFP is being redesigned for a new FPGA due to ob=
solescence. Gowin makes a 100QFP device that would be a good fit, but my cu=
stomer has said "no" to the 100% Chinese brand... US government customers, =
ya know!=20
> >=20
> >So now I'm looking at a BGA. I don't want to get into fine PCB design ru=
les, so 1.0 mm ball pitch is my preference. The only devices I can find tha=
t fit on the board have 196 or 256 pins. But the real problem is availabili=
ty.=20
> >=20
> >Digikey has a few of the XC7S15-1FTGB196I and more a scheduled for deliv=
ery in April. Add in the various speed and temperature flavors trickling in=
 (mostly in April) and I should be ok for the initial delivery in August...=
 if I can get my hands on those. I don't know if Digikey factors in the bac=
klog orders in these counts.=20
> >=20
> >Mouser shows great inventory of Efinix parts, particularly the T13 and T=
20 in a 0.8 mm 256 pin BGA, 10s of thousands in stock. But I'd rather work =
with a 1.0 mm BGA. Oddly enough, LCSC shows part numbers, but zero inventor=
y.=20
> >=20
> >Anyone work with 0.8 mm BGAs? What PWB feature dimensions did you use? D=
id this impact the PWB cost?
> The 0.8 mm 256-ball T20 isn't bad...=20
>=20
> https://www.dropbox.com/s/xjqgj2pz9mdhtma/P941_FPGA.jpg?raw=3D1=20

I can't really see much detail.  It looks like there are virtually no pads =
on the vias under the BGA.  What size are they?=20


> The BGA pads are 16 mils. 8 mil drills on the BGA vias. 6 mil traces=20
> mostly, except for the 50 ohm monsters. No big deal these days. Works=20
> great.=20

Yeah, 0.8 mm pad centers are doable, but I don't know where the line is for=
 higher pricing on the PWB.  The via pads seem to be pushing the technology=
 line at JLCPCB.  Not that I'm using them, but if they can do it, pretty mu=
ch anyone should be able to do it.  They build 0.45 mm via pads and 0.2 mm =
drills (5 mil annular ring and 8 mil drill), but charge extra for a 0.4 mm =
via pads (4 mil annular ring). =20


> We considered a T8 for a simpler application, but its 0.5 mm ball=20
> pitch looked nasty.=20

I didn't price the T8, because they use the logic cells for routing in a wa=
y they don't explain, so no way to factor it in.  The T12 would be gravy fo=
r my design I expect, but it's only $1 more for the T20, so why not?  If it=
 saves a day of work, it's a break even for 1,000 units.  If it enables a f=
uture expansion, it's worth much more than that!  Both parts seem to have t=
he same pin out, including I/O counts, so switching between them should onl=
y be a recompile.=20


> The efinix tool chain looks like it was developed in someone's garage,=20
> which is actually praise. It's free and simple and just works without=20
> 200 gbyte downloads and doing battle with FlexLM.

The large downloads are from the support for the many, many products the bi=
g three FPGA companies sell.  Don't expect Efinix tools to continue to be s=
mall... and they aren't really free.  You have to buy a board.  That's more=
 than I've paid for tools from FPGA vendors. =20

I'd really like to use the Gowin parts (LQFP100).  But the customer is hink=
y about parts from a Chinese company.  They sell stuff to the US Government=
.=20

--=20

Rick C.

++- Get 1,000 miles of free Supercharging
++- Tesla referral code - https://ts.la/richard11209

On Fri, 13 Jan 2023 21:20:50 -0800 (PST), "gnuarm.del...@gmail.com"
<gnuarm.deletethisbit@gmail.com> wrote:

>On Saturday, January 14, 2023 at 12:39:49 AM UTC-4, John Larkin wrote:
>> On Sat, 7 Jan 2023 09:49:24 -0800 (PST), "gnuarm.del...@gmail.com"
>> <gnuarm.del...@gmail.com> wrote: 
>> 
>> >A small board with a 100QFP is being redesigned for a new FPGA due to obsolescence. Gowin makes a 100QFP device that would be a good fit, but my customer has said "no" to the 100% Chinese brand... US government customers, ya know! 
>> > 
>> >So now I'm looking at a BGA. I don't want to get into fine PCB design rules, so 1.0 mm ball pitch is my preference. The only devices I can find that fit on the board have 196 or 256 pins. But the real problem is availability. 
>> > 
>> >Digikey has a few of the XC7S15-1FTGB196I and more a scheduled for delivery in April. Add in the various speed and temperature flavors trickling in (mostly in April) and I should be ok for the initial delivery in August... if I can get my hands on those. I don't know if Digikey factors in the backlog orders in these counts. 
>> > 
>> >Mouser shows great inventory of Efinix parts, particularly the T13 and T20 in a 0.8 mm 256 pin BGA, 10s of thousands in stock. But I'd rather work with a 1.0 mm BGA. Oddly enough, LCSC shows part numbers, but zero inventory. 
>> > 
>> >Anyone work with 0.8 mm BGAs? What PWB feature dimensions did you use? Did this impact the PWB cost?
>> The 0.8 mm 256-ball T20 isn't bad... 
>> 
>> https://www.dropbox.com/s/xjqgj2pz9mdhtma/P941_FPGA.jpg?raw=1 
>
>I can't really see much detail.  It looks like there are virtually no pads on the vias under the BGA.  What size are they? 

The BGAVIAs are 12.5 mil OD with 8 mil drills. The other vias on the
board STANDARDVIA and POWERVIA are bigger.

I have seen vias with no annullar ring, just a trace falling into a
hole, but the PCB houses don't like that.

Filled via-in-pad would be cool but that's complex and expensive. As
is buried vias.

>
>
>> The BGA pads are 16 mils. 8 mil drills on the BGA vias. 6 mil traces 
>> mostly, except for the 50 ohm monsters. No big deal these days. Works 
>> great. 
>
>Yeah, 0.8 mm pad centers are doable, but I don't know where the line is for higher pricing on the PWB.  The via pads seem to be pushing the technology line at JLCPCB.  Not that I'm using them, but if they can do it, pretty much anyone should be able to do it.  They build 0.45 mm via pads and 0.2 mm drills (5 mil annular ring and 8 mil drill), but charge extra for a 0.4 mm via pads (4 mil annular ring).  

We use US suppliers for production boards, and they seem to think this
6-layer board is within the normal range. One advantage to using a big
FPGA (256 balls in this case) is that you don't have to go deep to hit
enough balls, so may save a PCB layer or two. The T20-256 is a nice
part and Digikey has 29,000 in stock.

Another project used a 484 ball Zynq and we used almost every ball.
Lots of different power pours too. That took 10 layers. Another recent
board has a 400-ball ZYNQ with a few unused PS pins and fits on 8
layers.

The ZYNQ has analog inputs but, crazily, they are all differential so
they make you ground a perfectly good i/o pin for every analog input
that you want.

>
>
>> We considered a T8 for a simpler application, but its 0.5 mm ball 
>> pitch looked nasty. 
>
>I didn't price the T8, because they use the logic cells for routing in a way they don't explain, so no way to factor it in.  The T12 would be gravy for my design I expect, but it's only $1 more for the T20, so why not?  If it saves a day of work, it's a break even for 1,000 units.  If it enables a future expansion, it's worth much more than that!  Both parts seem to have the same pin out, including I/O counts, so switching between them should only be a recompile. 
>
>
>> The efinix tool chain looks like it was developed in someone's garage, 
>> which is actually praise. It's free and simple and just works without 
>> 200 gbyte downloads and doing battle with FlexLM.
>
>The large downloads are from the support for the many, many products the big three FPGA companies sell.  Don't expect Efinix tools to continue to be small... and they aren't really free.  You have to buy a board.  That's more than I've paid for tools from FPGA vendors. 

$150! That's in the noise, and an eval board is good anyhow.

 
>
>I'd really like to use the Gowin parts (LQFP100).  But the customer is hinky about parts from a Chinese company.  They sell stuff to the US Government. 

Yeah, we have a lot of aerospace customers and avoid Chinese parts.