TonyD wrote:Another one to consider is Seeedstudio. Their Fusion PCB service costs $40 for 10-pieces of 2 Layer 10cm*10cm.
+1 for seeedstudio. I have had probably 20 orders from them and they have all been good.
Probably not suitable for BGA track widths but have done lots of 4 layer 8x8mil boards that always work. They have the best price for 4 layer boards I have ever found. Goldphenix can't touch them for price.
Personally I usually use OSH Park, prices are fantastic and the quality is good. They do 4 layer too though the smallest trace/gap is 6 mil, 13 mil hole. I am still working on a board design for a Dev Board based on the U6A/U8A on the 96-FBGA package, certainly is pushing OSH Park to their limits but I am pretty confident it will work out (Should only cost about ~$6 a board), certainly no issue using TQFP.
I have also used Elecrow and iTeadStudio but neither are good enough for BGA, but would be fine for TQFP and only cost about $20 for 10 boards.
I think XMOS should reconsider their BGA packaging.
As can be seen from the posts above most people are just rambling on about their favorite
cheapo DIY board service few of which are capable of making boards suitable for BGA .
I don't see anybody mentioning the boards they HAVE had made for BGA because probably few people have.
The trace widths and via sizes needed are too small for these batch board companies.
BGA requires expensive fine pitch boards which are a waste of money if only one chip is BGA and the rest are SOIC. TQFP packages would be far more sensible and are even possible to hand assemble for prototyping etc. Unless you have the money and contacts to have BGA boards made and someone that can place the parts for you prototyping is going to be tough. No prototyping means fewer people will be using XMOS parts for real world projects.
One of the things that really got me was the port assignments as I was laying out my PCB, how come 32A is such a huge port with no other secondaries? Why only 1 8 bit port? It seems to me an 8-bit port would be quite a commodity to have at least 2 on a single tile, followed by the fact that hardware clocked ports only work on falling edge, when I came up upon these limits myself in some of the very first project concepts I came up with using the xcore chip I almost wanted to scrap designing my dev board as it made me realize the usefulness of these chips might be hindered enough to consider them not usable in a product.
With that said I do love what xcore can bring to the table, but there are some limitations that seriously make me doubt it's feasibility in a product, case in point the 8 bit ports, I want to drive 2 DACs but the speed I need to drive them at makes it near impossible to do with software port control, but I can do it with hardware port buffering, but oh wait, only a single 8 bit port, I CANT hardware drive the second DAC. Then I find out the DAC I am using is rising edge clocked, so no, I can't use either 8-bit PORT. What I would of gained by using xcore I lost in hardware functionality, and I minus well just use STM32F4 because it will be fast enough to handle the port needs and has multiple 8 bit ports, and can handle rising or falling edge.
That does in-fact bring me to the next problem which is the BGA, sure I can work with the 96FBGA with the current companies I use, but I don't know a company that can handle anything tighter then that for less then a few hundred dollars, making it near impossible to do small time products with it.
Will I use xcore in a product? Maybe if my hardware port needs will work on it as I need it too and the multithreading gives me an advantage over another chip, but unfortunately ALL of the projects I have in my mind I thought it would be good for are better suited to something like an STM32F4 simply because of port structure above all else.