yikes

Youch!

I had a similar problem with an LM2842. The "pin 1" indicator is a slight difference in texture along one edge. Impossible to see if you get anything on it. Amusingly, I was replacing it because mine also wouldn't come up right - it was floating around 1.3v and not switching. Turns out - and this is NOT DOCUMENTED - the switcher won't start if anything else is powering that rail, even a tiny bit. At least they had a drop-in replacement[*] that didn't have that problem.

Nice PCB !

Hmm, I just designed the LM2842 into my RIS-796A, where it powers a 12V fan. Seems to work fine for that job.

It goes in the front of this box:

It's always a nail biter to bring up a big board and hope that no fatal mistakes were made. The supplies are up now and we're jtag'ing ARM code. The next big event will be to successfully configure the FPGA.

My main power supply is a kilowatt of 48-volt MeanWell. The first thing on this board is to switch that down to +12, with an ancient LM2576-HV, and then run everything else off the +12. The regulators get three deep.

The 2576 switches at the nosebleed rate of 52 KHz.

The first switcher I designed ran at 24 KHz, because I could hear 22K at that time.

Wow. TI has had some real lemons...

TPS40210 is the last one I stepped on. Has a current fault function, which happens to be within the error amp's range for some perverse reason. That is, the current sense range is e.g. 0-100mV nominal (proportional to error amp output), with a fixed 150mV threshold that flips a latch and discharges the soft start cap (basically restarting from cold, so you wait a soft-start cycle before going it again). But the error amp doesn't saturate at a voltage corresponding to 100mV threshold, it goes over 150mV. So it dutifully saturates, commanding full startup current as a converter is supposed to, and faults its retarded ass, then sits there in a fit of hiccuping...

Fix I employed, remove the SS cap so it restarts ~immediately (within 100s of us). This gives about half (200-300mA) the startup current compared to nominal capacity (~500mA). Fortunately the expected load is very light, so it should behave...

Tim

Maxim "solved" that problem by making this i2c isolator rotation-symmetrical and then they just don't put a dot indicator on the chip.

They don't tell you that explicitly though AFAIK, they like you to puzzle over why there isn't one and then figure out how clever they are on your own.

Easy fix: Never Buy Maxim.

The TPS54302 data sheet shows a square pin1 index that doesn't exist on actual chips. Different batches are marked differently too. All are laser markings, barely visible if the light is just right.

ICs should never have been rotationally symmetric. That was a historical blunder.

They could come in a few variety of shapes, like Tetris pieces. You could really get dense-packed boards that way.

Hmm, maybe that's worth a patent.

I used the LM2842 because I had used it successfully in other projects, so had a working reference design to paste in. But this latest project had two other power rails, +5 and +12, and something was leaking some of that power into the +3.3 rail.

We had one today, too. We use an opto to power-up a mezzanine board that draws only about 5mA @ 3.3 VDC. We wanted to turn the board off to save power budget when it runs on battery backup. (Hardly worth the effort, but every little bit helps!)

Long story short: The EagleCAD library had the transistor side of the OPTO pin assignment reversed. Looks great on the paper schematic, not on the Gerbers!

Luckily, they make suitable opto's with the alternate pinout, so we only need to adjust the BOM and back-annotate the error.

I've seen LED matices with that feature.

I got a pair of (the red version of) them in a kitset and there were no markings for pin 1 on the board, I traced out the wiring and then noticed the symmetry.

looking at page 10 I'm not seeing that, the VDD pins on each side aren't rotationally symmetrical.

I've been bitten by this one, too. One is the index and two other ones are marks from ejector pins that push out the package from the mould. They might be in the package drawing or not, it depends.

Assembly houses I've used have never asked about this - it's probably business as usual and noticed by pick and place camera.

Oh you're right. I guess I mentally transposed the VDD pins with the upper grounds.

Ok well there's still no orientation marking on the chip, I guess they're just jerks. Should have figured Maxim wouldn't be "clever"

Lots of small signal dual MOSFETs are like that, e.g.:

I put a lot of N-channel and P-channel like that in several designs. I had to explain more than three times to production that there was no pin

The time consumed in such conversations can make it quite costly.

These chips are picked from a tray and in that case the Universal machine assumes that they are oriented correctly. The operator put them in wrong.

I've alerted QC to give particular attention to these ST chips in the future, and our P+P operator is sufficiently embarassed.

The FLIR thermal imager was worth the absurd price we paid for it. It answers the often vexing question "where is the current going?"

Not in 10K+ runs. The time (= cost) on the pick and place machine for 1 instead of 2 components already made up for that. I just got a call once when a new design was entered into the machine.

Also, it saves enormously in set up cost to minimize in different components, so use the same value as often as possible. Example: resistor networks, we used 100K's of 4 single resistors in a 0804 package. The buying cost per package is way below the pick and place cost for a package so it is a lot cheaper than using separate resistors. I just designed with 'mostly' 10K (logic, pull-up) and 50R (DDR terminators). Example:

One process that I find annoying is to review all the R and C values in a design to try to reduce the number of BOM line items and feeder reels.

Using quad r-packs is always tempting, but they lock down the ratios.

I have a couple of charts of all the resistor values and ratios that can be made using a quad resistor pack.