Textmode has quit [Read error: Connection reset by peer]
X-Scale has quit [Ping timeout: 265 seconds]
chris_99 has joined #homecmos
<berndj>
kanzure, re "The original 4004 and earlier devices are within the scope of what could be accomplished in a home environment." - really?? i thought even 4004 level tech was a bit beyond home-scale fabbing
<kanzure>
2300 transistors? 10 microns? sure....
<kanzure>
and 10 microns was the *minimum* feature size
<kanzure>
"The Intel 4004 was designed by physically cutting sheets of Rubylith into thin strips to lay out the circuits to be printed"
<_Sync_>
but the problem is getting the reliability
<SpeedEvil>
It depends what you mean.
<berndj>
the geometry isn't what i'm wondering about, it's the yield - related to reagent purity etc
<SpeedEvil>
It is within the scope of things that can reasonably be done in a garage.
<SpeedEvil>
A large garage.
<kanzure>
this would not be mission impossible, though
<berndj>
like won't there be sodium ions *everywhere* ruining things?
<SpeedEvil>
It does not require a massive industrial complex.
<kanzure>
ruining things? use a clean room dude.
<berndj>
or do they not ruin things badly enough at 10um
<kanzure>
many people have clean rooms in their garage (or whatever)
<SpeedEvil>
This is assuming you can properly build out the working space, and purchase reagents that are pure enough
<_Sync_>
they do berndj
<_Sync_>
SpeedEvil: the problem is not the single steps
<_Sync_>
but the whole process
<_Sync_>
that's the problem
<SpeedEvil>
_Sync_: I know.
<SpeedEvil>
_Sync_: the problem is possibly not actually getting it into one garage.
<berndj>
_Sync_, is there just no tolerance at all for some ions? is it a case of 1 sodium ion == dead chip?
<SpeedEvil>
But getting it into one lifetime.
<kanzure>
berndj: what were you reading?
<_Sync_>
berndj: sodium is not really the number one problem
<_Sync_>
usually it is gold
<whitequark>
gold?
<_Sync_>
yes
<whitequark>
why?
<_Sync_>
because it is popular as a contact material
<whitequark>
oh
<_Sync_>
thus you have to be *really* careful not to spread it around your workbenches
<berndj>
kanzure, your message on bitcoin-development re DIY ASIC mining chips
<_Sync_>
a foreign student once forced us to buy a new spindryer
<_Sync_>
because the idiot thought it was a good idea to wash off his plated samples in it
<_Sync_>
even tough it says "NO GOLD IN HERE" very clearly
<_Sync_>
and I'd say it is entirely possible to cram the production into a garage
<_Sync_>
there actually is not that much going on
<berndj>
gold is bad because it acts as a recombination site?
<kanzure>
berndj: it would be more appropriate to do bitcoin wallet manufacturing, not miner manufacturing
<kanzure>
however, you often don't need a wallet in hardware anyway
<kanzure>
unless you want to be sure that your hardware is not compromised (but this is hard because you eventually have to leave the garage)
<berndj>
kanzure, that's true almost by definition - if one could make chips in a garage, one can make better chips in a 100-garage site. more capital = better chips
<kanzure>
why would it matter by which method it is true?
<whitequark>
wouldn't it be easier to make artwork using FOSS, manufacture chips on a real fab, then compare them with your artwork?
<_Sync_>
berndj: it has a high distance from the valence/conduction band
<kanzure>
whitequark: no, decapping takes a loooong time and there's also dopant-level trojans which take an even longer time to find
<whitequark>
not "easy" but "easier".
<kanzure>
it would be easier to do nothing at all
<whitequark>
eh?
<berndj>
whitequark, what does the artwork buy you?
<kanzure>
sigh
<whitequark>
berndj: my point: if you can prove the fab didn't modify anything, you don't have to trust it
<berndj>
whitequark, unmodified artwork proves only that the *artwork* wasn't modified
<SpeedEvil>
Also, there are ways that you can work with compromised hardware.
<SpeedEvil>
There are techniques to do computing on untrusted hardware.
<SpeedEvil>
- at massive performance penalties often
<berndj>
you're talking homomorphic computing and moon math?
<whitequark>
berndj: sure. but producing trusted artwork is not a capital-intensive step
<whitequark>
(and you will have to do it anyway even for your home fab)
<berndj>
oooh, you mean artwork as a general term for the whole chip's layout? i thought you meant silicon doodles
<_Sync_>
it is the same as pcb artwork
<whitequark>
^
<whitequark>
alternative implementation: design a really small trusted base, then assemble it using 4000 series ICs and a large PCB
* whitequark
hides
<SpeedEvil>
Lego!
<SpeedEvil>
Just do all your computing in minecraft, with redstone
<whitequark>
I actually want to take some early chip and replicate it almost exactly on a PCB
<whitequark>
though not enough to actually invest time in that
<whitequark>
but it would be neat
<SpeedEvil>
whitequark: seen the 'big 555' kit?
<whitequark>
put a LED near every transistor
<whitequark>
now THAT is blinkenlights
<_Sync_>
dat clock rate
<whitequark>
SpeedEvil: yeah, something like that, except done with SOT-416 transistors and pick&place
<whitequark>
it would be on the same scale as the non-scaled maskwork, maybe even a bit smaller
<SpeedEvil>
I wish I had good health.
<SpeedEvil>
One of the projects on my stack is an automated 'wirebonder'
<SpeedEvil>
Which automatically point-point wires with enamelled copper wire and LASER and solder SMD components.
<SpeedEvil>
Possibly in a reducing atmosphere, maybe just inert.
<whitequark>
SOT416 transistors go up to like 300mA
<_Sync_>
what's the point SpeedEvil
<whitequark>
assuming a 1mA per LED, it's reasonable even in the worst case
<SpeedEvil>
Not nearly so much if dead-bugged with no PCB. But still quite usable
<SpeedEvil>
_Sync_: very dense, rapid production.
<whitequark>
LASER?
<SpeedEvil>
_Sync_: Is it going to be comparable with mass-production pick and place - no
<_Sync_>
if there would be a need for that it'd be a already on the market
<_Sync_>
if I can get a board turned in under 24h inhouse, that's usually fine
<SpeedEvil>
_Sync_: That's not quite true.
<SpeedEvil>
_Sync_: If you can get it down to a price point below where any of the vendors that might have considered doing it have gone - through advances in technology, or accepting more crapness, it may be at least saleable.
<SpeedEvil>
There is nothing that stopped 3d printers being sold in the 80s.
<_Sync_>
we have had a dual litho 3d printer since the early 90s
<whitequark>
bofh__: yeah, I remember when that hit hackaday
<whitequark>
was about to find and post the link
<_Sync_>
if there would *really* a point in doing so there'd be someone buying it
<_Sync_>
but as I said, I can get a board spun in under 24h and then pop it into my production like
<_Sync_>
there is no big point
<_Sync_>
and usually you only need that for involved designs where the pcb is of more concern than the rest
<SpeedEvil>
I think it would be a fun thing to do, with at least niche applications, and possibly saleable in small volumes for more than I could make it at. That's pretty much all.
<SpeedEvil>
I'm not arguing that it's better in any way than a nice PCB and assembly facility.
chris_99 has quit [Remote host closed the connection]
<bofh__>
whitequark: personally I found the most fascinating part of that is reading a bit more and realising just how much more popular ECL used to be
<_Sync_>
my point is, if you are trying to make a rapid prototyping tool it has to meet certain parameters
<_Sync_>
and you are not going to hit them with copper wire
<_Sync_>
bofh__: not suprising, as it was fast
chris_99 has joined #homecmos
chris_99 has quit [Changing host]
chris_99 has joined #homecmos
<bofh__>
_Sync_: more surprised that it's relatively uncommon nowdays
<whitequark>
the USSR ECL families were quite convenient for some tasks
<whitequark>
for example, boiling water
<_Sync_>
negative voltages and current
<_Sync_>
they LOVE power
<bofh__>
_Sync_: heck, they used to use it in CPU designs... the alpha ev6 / Tsunami chipset was almost entirely ECL... only way to get 833MHz back in 1997
<whitequark>
how many kW did it dissipate?
<bofh__>
whitequark: the manual rated it for ~940 BTU of heat
<bofh__>
no joke
<bofh__>
iirc the exact number was 939BTU
<whitequark>
BTU per what?
<whitequark>
940 BTU is approximately 1 MJ
<bofh__>
oh, per hr
<whitequark>
275W?
<_Sync_>
I don't think so bofh__
<bofh__>
130W / cpu * 2 is what they were rated for
<_Sync_>
iirc IBM had a SOI chipset running at 850
<bofh__>
so that number falls right in line
<_Sync_>
which was completely cmos
<bofh__>
I just found it hilarious that the CPU document gave an actual heat output value
<_Sync_>
very important number
<bofh__>
_Sync_: what year was that from? cool, did not know about that
<_Sync_>
96 or something
<bofh__>
whitequark: fun thing about ECL is that the power draw doesn't vary much based on CPU load (or really at all)
<_Sync_>
In September 1998, Samsung announced they would fabricate a variant of the Alpha 21264B in a 0.18 µm fully depleted silicon-on-insulator (SOI) process with copper interconnects that was capable of achieving a clock frequency of 1.5 GHz. This version never materialized.
<_Sync_>
ah
<_Sync_>
that was the time
<_Sync_>
yeah
<_Sync_>
iirc IBM had some power cpu at 8xxMHz in the same timeframe
<_Sync_>
I'd have to dig through my notes
<_Sync_>
they basically took a gen2 POWER, put it through an SOI process and it ran faster
<bofh__>
not surprised
<_Sync_>
they also tried strained silicon and Si-C back then but apparently they all died because they did not understand how it worked
<SpeedEvil>
SOI is coming back!
<SpeedEvil>
At least allegedly
<_Sync_>
lots of back and forth on that
<_Sync_>
I think it's not gonna happen soon
<_Sync_>
because of dat cooling
<whitequark>
use a diamond substrate? :D
<whitequark>
I've heard that's investigated for LEDs
<_Sync_>
I have one of the first blue leds here in a block of plexi
<_Sync_>
well
<_Sync_>
point is, stuff needs to be cheap
<_Sync_>
cpus get optimized quite a few digits right of the decimal point