<_whitenotifier-c>
[starshipraider] azonenberg 0b15709 - Missing change from last commit
<azonenberg>
welp, after spending way too much time debugging, i have a respin with four changes being ordered
<azonenberg>
one is cosmetic, two are relatively minor DFM fixes
<azonenberg>
and one killed the amplifier on this board :p
<azonenberg>
good news: the respun board is going to cost me less than $3 at oshpark
<azonenberg>
bad news: the amplifier is almost $50
<azonenberg>
plus ~2 weeks of delay and the other parts, and a new stencil.
<azonenberg>
and it's not even the $50 it's that i only had 5 of them
<azonenberg>
so the next one better work well enoguh for me to at least get some waveforms out of it
<azonenberg>
i ordered more but they're not coming until october at best
<monochroma>
what killed the amp?
<azonenberg>
monochroma: i connected the active-low power down strap to +2.5V
<azonenberg>
except the amplifier's "ground" was at -2.5V for dual supply operation
<azonenberg>
so it was effectively +5V
<azonenberg>
and apparently despite it being designed for 5V single supply operation, input max on the powerdown pin is 2.1 - 3.3V logic high with a nominal 2.8V on die pullup
<azonenberg>
sorry i meant, the logic level is
<azonenberg>
absolute max is +3.6V
<azonenberg>
with +5V on it, i apparently blew out the protection diode bad enough that the +2.5V supply rail was measuring like -1V
<azonenberg>
and with the board off the +2.5V rail measured about 2 ohms to -2.5V
<azonenberg>
i reworked the board and put the same chip back on it, but the internal pullup was fried and the chip was off with powerdown floating
<azonenberg>
with powerdown hard pulled to ground (+2.5V vs chip ground) power consumption of the chip increased but output still flatlined no matter what i threw at the input
<GenTooMan>
azonenberg, I suppose a nice big (slow?) input isolation resistor would not help much?
<azonenberg>
GenTooMan: it's a bit late for any of that :p
<azonenberg>
The chip is fried and i have a respin ordered
<GenTooMan>
azonenberg, I am aware just the "FYI this might have helped" still may be good to put the isolation resistor in regardless. :D
<azonenberg>
I usually use 0R resistors for straps
<azonenberg>
i omitted them in this design as it's cost optimized
<azonenberg>
had i used a 1K or something maybe the bug would not have been fatal to the chip
<azonenberg>
s/cost optimized/size optimized/
<GenTooMan>
I guess size issues will do it.
<azonenberg>
yeah i mean i remember doing the initial layout of the -3.5V supply which uses a WLCSP buck-boost converter
<azonenberg>
and thinking "why are these resistors so enormous?"
<azonenberg>
they were 0402s
<azonenberg>
the entire supply fits on the back side of a u.fl
<azonenberg>
and i still wanted it to shrink
<azonenberg>
the production rev of this board will likely use a HDI stackup with blind filled via-in-pad
<azonenberg>
maybe buried vias as well
<azonenberg>
shrink as many passives as i can from 0402 to 0201
<azonenberg>
i think the ICs are already as small as they can get (UDFN and WLCSP)
<azonenberg>
the connector footprint can definitely shrink a bunch, that's already on the todo
<GenTooMan>
well it's unlikely using flip chip direct bonding will make it cost effective enough.
<GenTooMan>
you aren't making 100k of these
<azonenberg>
This is also an active differential probe though, targeting 6 GHz BW
<azonenberg>
it's competing with commercial products with high 4 / low 5 digit price tags
<azonenberg>
So it's worth using fairly extreme design rules and PCB techniques if they will make it better
<azonenberg>
if it ends up costing $200 instead of $100 per unit in BOM/assembly, it's still a massive cost savings for the end user
<azonenberg>
who cares if you end up saving $7800 instead of $7900 on the cost of the probe? Lol
<GenTooMan>
just presenting possibilities as one person astutely said "all things are possible, just not practical" so in this case ascertaining what's practical is key I suppose.
<GenTooMan>
integrated passives into the PCB might help ... FYI
<azonenberg>
lol
<azonenberg>
now *that* is expensive. realistically, wirebonding is probably not goign to happen
<azonenberg>
mostly because i dont think it will actually help much
<azonenberg>
WLCSP etc are already the size of the die