S3 Feather Hardware, 3.3V voltage generation

[wulflord1981]

Hi,

I ordered 5 boards from jlcpcb, just arrived this week.
Before putting the whole setup together, i checked the voltages on the Board.
If i just connect the 12V voltage to the pure board (No esp32 board) and measure 3.3V pin, there are 9.8V

I checked already all resistors and orientations of the ic´s, but everything seems fine.

The only thing I can not check, is if there is the correct PTS82130 installed.
I cannot see any markings.

Any Ideas?

[JanM321]

CC @cure

[cure]

Hmm, I'm not sure if that's normal, or not. I will try to replicate with one of my boards, I should be able to do that next weekend. I also got my boards from jlcpcb, and the one I've been using has been running flawlessly for ~9 months now.

[wulflord1981]

@cure that would be great.
I already checked/measured a second Board from that batch. Same 9.8V at 3.3V pin.
At work I have an microscope maybe i can read the branding of the TPS ic.
I let you know If i make Progress es.

[NFXUS007]

I was literally about to make the same post until I saw this one on top of the list.

I have the same issue and from the same manufacturer being JLCPCB. My order was made back in late October 2025, but I finally got back around to this project only yesterday only to find out the 5 boards are systematically incorrect in terms of VOUT.

The BOM did say to use C473914 (TPS82130) for U1, however it was replaced by C163210 (TI TPS82130SILT) for some reason (Out of stock? – I can’t remember). These are meant to be functionally identical and interchangeable. All other part numbers assembled are identical to the BOM.

I’m not electronics savvy, but I’m pretty resourceful. I used the _sch and _pcb file in an electronic design viewer (Altium.com – Really handy!!) to understand the pin outs and check for breaks.

What I have seen so far when injecting 12V at Line 3 [12V Input]:

• R1 is showing 242k in one direction and 183k in the other
• R2 is showing 92k in one direction and 89k in the other
• R3 is showing 93K in one direction and 100k in the other.
• FB (Line 6) is giving me about 1.76V, where it’s meant to be 0.8V.
• VOUT is giving me 11.1V and not the expected 3.3V.

This appear to be one of the better results. Another PCB of the same batch has around 175k on R1 and FB (Line 6) is around 4.66V.

I thought these resistors are unidirectional, but is suspect that the resistance is running down another parallel path maybe?

I don't particularly want to be ordering replacement boards if they come with the same/similar voltage issues. Hopefully someone is able to assist?

[wulflord1981]

Hi,

my researches says TSP82130 and TSP82130SILT are the same ic.
If measuring the resistors in circuit you maybe not measure the real value.
You can check the value of the resistor with the print on it.
1003 means 100k for example.
On my boards all R, C, D's are correct.
But Vout is wrong.

[NFXUS007]

All my resistors are correctly labelled for each position, but I don't have anything to measure the capacitors. Diodes just have T4 labelled on them.

Some reading indicates that others have reflowed the solder on the TSP82130 to resolve the issue.

[NFXUS007]

Any one with recent success? I ordered another batch with the original C473914 (TPS82130) for U1. But I'm still getting the same results as before and not out putting 3.3V.

[cure]

@NFXUS007 Apologies for the long delay.

I've now run some tests on my original hardware batch. I have one unit in production (which has been working fine for almost a year now), and four spares. I hooked up a 12V bench supply to my spare boards, then measured between the 3.3V and the ground pin.

On all four spares, when I don't connect a FeatherS3, I measure something close to 12V between the 3.3V pin and ground. When I hook up a FeatherS3, the voltage over the 3.3V and ground pins is around 3.45V, and the FeatherS3 operates normally.

So... at least for my batch of the hardware, the voltage on the floating 3.3V pin does not appear to be representative of when a FeatherS3 is connected.

Have you learned anything else, in the meantime?

[NFXUS007]

Thanks for the clarification!

After your post, I decided to plug in the FeatherS3 and progressively increase the input voltage from 3V all the way to 12V to minimise risk of damage. All of my boards sit around 3.45V and work perfectly! I would definitely suggest adding this into the 'read me' for others to be aware of this quirk.

The other thing that had me thinking that I damaged something was that I was expecting the red LED on the FeatherS3 to light up when powered, similarly to when being connected via USB. This made me start checking voltages, which led me down this path.

Out of curiosity, it is it about the FeatherS3 that needs it to be connected to allow the regulator on the PCB to regulate the output voltage? I would have expected the PCB to regulate power irrespective of a consumer connected.

Hopefully this weekend I can finally connect it to the AC and join the others who had their success with a completed project.

[cure]

Out of curiosity, it is it about the FeatherS3 that needs it to be connected to allow the regulator on the PCB to regulate the
output voltage? I would have expected the PCB to regulate power irrespective of a consumer connected.

Yeah, it was a bit surprising to me as well, apparently the U1 needs some load to regulate the voltage down to 3.3V. A Zener diode could be added to the design to make it always 3.3V, even when the pin is floating.

Did you make further progress?

[dav20011]

I ordered hardware-tiny and, testing with jumper leads (as my DevKit was delayed in shipping), the 3.3V rail immediately cooked my ESP32's flash. After my own testing the cause is clear.

The TPS82130 is not at fault. Both hardware-tiny and hardware-FeatherS3 are missing the required output capacitor (the original hardware variant uses a self-contained TSRN1-2433 module and is unaffected). The TPS82130 integrates only the inductor and needs an external low-ESR ceramic output cap - see TI datasheet Fig 8-1, C2 = 22 µF on VOUT. On both boards the only bulk cap (C1) is on the +12V input. The +3V3 net has no capacitor at all.

Without it the DCS-Control loop can't regulate. The design only works because a directly-socketed ESP board's own 3V3 decoupling caps happen to sit close enough to act as the output cap. In my case ~15 cm jumper leads added enough inductance to isolate those caps from VOUT, so the regulator ran effectively uncompensated and the rail over-volted into the ESP32.

Evidence (all at 7 V input via a bench PSU):

• No load: VOUT swings ~9 V pp (−1.5 to +7.5 V) at 1.75 MHz, basically the unfiltered output.
• ESP32 attached via jumpers: still ~2.9 V pp with 5.4 V peaks, well above the ESP's 3.6 V max. (note: the picture primarily shows the PSM ripple, the ~2 MHz ripple from the previous picture is the noise band during the voltage rise here)
  
• At 10 V no load the board draws ~100 mA and the TPS runs hot (~60 °C)

Adding a 22 µF cap at the regulator's output should fix all issues including the original 9.8V observation. The choice to not use a cap was deliberate (see this comment). From an engineering perspective this is very unusual as it makes the circuit unreliable unless exactly the correct operating conditions are met. As an electrolytic capacitor is the wrong part for this job anyway, replacing it with a ceramic one would have both solved the part availability and improved the design.

[Desses1984]

does this means that the files on the Github are bad? I just ordered 5 boards using the "hardware-FeatherS3" (May 2025)
gerber , bom and and cpl files..

[dav20011]

does this means that the files on the Github are bad? I just ordered 5 boards using the "hardware-FeatherS3" (May 2025) gerber , bom and and cpl files..

From a convential design perspective? I would say yes. This would definitely be flagged in a review in the industry. The design violates very explicit requirements from TI (input cap is not low-ESR ceramic, output cap is not close the regulator) and relies even on the specific layout of another board (the inductance depends on the length of the traces, pin headers, etc.) which may or may not undergo multiple hardware revisions. It's simply very error prone.

From a hobbyist perspective? I will measure with an oscilloscope in the intended constellation and if the voltage spikes are low enough to not kill the flash again this is good enough for me. I don't think this will cause critical issues like fires. Before anything serious happens the TPS will likely just run into its thermal protection. Note that measurements with a multimeter are pretty meaningless here. Sadly, I can't provide measurements with the DevKit yet, Mouser currently has long delays on this hardware. Also, it should considered that the original decision for this was based on the DevKit, the FeatherS3 might behave differently.

Regarding a fix: Adding a capacitor to the design is quite trivial. For existing boards, it is possible to solder a capacitor to the board with wires. As the indoor unit vibrates a bit via the copper tubes I suggest glueing such a fix down with epoxy (don't put glue on the TPS).

[cure]

@dav20011 thanks for this analysis! I added hardware-FeatherS3, which is an iteration on hardware-tiny. I'm interested in improving the design. Looking at the points you flagged, a question regarding the input cap:

input cap needs to be low-ESR ceramic: I think this was already fixed in hardware-FeatherS3: C1 is different from the two earlier designs, it is a Taiyo Yuden TMK325BJ106KN-T, which is a ceramic cap. Would you agree?

I'll see about adding a low-ESR ceramic 22 µF cap at the regulator's output.