thanks to AI demand, Lattice cannot make enough FPGAs to fulfill orders on iCE40's (and probably ECP5's) with less than a year of lead time. this is likely to cause supply chain issues for Glasgow revD

words cannot describe my contempt for this shit, and my disdain for anyone still pushing it

(there is no shortage or supply chain issues with the silicon itself. rather, the problem here is the resin used to make a part of the packaging. in theory it would be possible to just glue a die to the PCB and bond it out during assembly, in practice this would be prohibitively expensive as far as I know)

@whitequark
This whole house of fab-less cards was destined for a collapse. "Let's put a tiny island under constant political and military threat at the epicenter of the entire digital supply chain. What could possibly go wrong?"

Not like we haven't been here before, either:

https://www.computerworld.com/article/1456590/mitsubishi-to-resume-making-chip-packaging-resin-after-quake.html

Mitsubishi to resume making chip packaging resin after quake

Computerworld
@RealGene oh, 2011 is before I was involved in electronics production in any real capacity
@whitequark Surely there are plenty of spare old foundries to make them? AI chips really need EUV these days.

@lynne it's not about the silicon. the silicon is fine

it's about the little piece of printed circuit board that redistributes signals from the bond wires to the balls under the BGA

@whitequark *that's* the bottleneck?
I'd have thought that this would have been vertically integrated into the production.
@lynne you would think so, but there's a worldwide shortage of the stuff now
@lynne specifically of the epoxy resin, which is only manufactured in a few places in the world
@lynne @whitequark funnily enough apparently one of the major producers of this stuff is a company primarily making sauces and other seasoning, which at some point branched out into making packaging materials as a side thing...
@HeNeArXn @lynne oh like Ball jars?
@whitequark @lynne seems more resins and such, branching out from the chemistry they had developed for food science. This company: https://www.ajinomoto.com
@HeNeArXn @whitequark the global hardware industry is reliant on the company that makes MSG; the global software industry is reliant on furries. there's a connection here
@HeNeArXn @whitequark @lynne Oh hah, yeah I have some of their Hondashi.
@HeNeArXn @lynne @whitequark i should walk up to lattice and tell them if the teriyaki epoxy is out we‘re also fine with garlic flavor
@whitequark there we go with FPGA shortage again  for more absurd reasons this time...
@ariac i wonder if TQFPs are still being packaged
@whitequark and i wonder if the price of used wire bonders already got higher.

i've been coincidentally asked about it today
@whitequark is this only affecting BGAs at the moment? Or QFN, QFP, etc as well?
@whitequark is the part available in WLCSP? Even if it means a more expensive PCB for early units, a polyimide RDL should not be affected by a BT shortage
@whitequark (also the consequences of a BT shortage are going to be immense it's gonna hit LGA modules, BGAs, SiPs...)
@azonenberg @whitequark It looks like the closest package for the 8K is the CB132, which is just a high density BGA package and might not have enough IO for the glasgow anyways. I think that this is a curse of the original ICE40s being a bit old, unfortunately.

@whitequark

in theory it would be possible to just glue a die to the PCB and bond it out during assembly, in practice this would be prohibitively expensive as far as I knowConsidering that most calculators have chips packaged this way, I assume it's because of the setup costs, not the per-unit costs, right?

@whitequark Check out the newer Lattice Nexus parts - 9k to 100k logic cells. In stock at Mouser.

https://www.mouser.com/c/?q=LFD2NX-9

@0h00000000 @whitequark

EDIT: So there is (preliminary?) support for Nexus in Yosys and nextpnr-oxide, which is great news. But I didn't find quickly to reach information on feature completeness and maturity.

The whole idea behind Glasgow: Create the bitstream in-situ using the highly efficient Yosys toolchain and a nice to use, even scriptable HDL.

@0h00000000 turns out there's many more ECP5's than anticipated, so we might go with a 12F