If think you can reduce the requirement for output current of the LDO to around 1.5A max. That is the highest value I know of for 486 class CPUs. Don't know the spec for the SXL2 atm but highly doubt it beeing much higher. Meaning you could use other LDOs with enough margin in dropout voltage at this current.

Also, a voltage drop of >100mV over load seems a bit odd. Maybe the dropout voltage rose with increasing current and input voltage wasn't sufficient for set output voltage. But I'm just guessing here.

pentiumspeed wrote on 2022-12-01, 23:01:

Have you confirmed that adapter's ground plane is not oscillating? For this, oscilloscope's ground clipped to the motherboard's ground and your probe on the adapter's ground plane?

Here's what the ground looks like connected per your suggestion. The setup:

Low freq view:

High freq view (similar scale to previous scope images):

And zoomed in:

However...

However, that bump still exists without the replica SXL2 interposer installed.

Here it is with an Am386 installed and clipping as shown:

Note that I have swapped the motherboard to one I have more than one of. I didn't want to ruin my DTK Symphony 361/362 board.

I'll get to the other suggestions when I have time.

feipoa wrote on 2022-12-02, 07:55:

If you have DIP-14 crystal oscillators in the 81-89 MHz range, we might be able to achieve 85 MHz on some select CPUs, but I don't have any such oscillators and they are (I think) uncommon.

if you play with arduino get yourself $5 "Si5351A I2C Clock Generator Breakout Board 8 KHZ to 160 MHz", if not $30 "SI5351 High Frequency Clock Signal Generator + STM32 Main Controller TFT Display" includes some kind of controller and display. Will allow you to control clock however you want.

feipoa wrote on 2022-12-02, 07:55:

rasz_pl wrote on 2022-12-02, 06:52:

Yes. It would be illumination to see ~2 ohm power resistor (I assume TI486SXL2-66 is consuming 5-10W?) instead of CPU, connected thru couple of CPU socket power pins, not directly to regulator output.

From my notes, I had used 5 ohm for 670 mA and 10 ohm for 350 mA. I'll have to check what I have, but I could four 20-ohm resistors in parallel between VCC3 and GND pushed into the PGA socket. Or maybe six 30-ohm resistors.

Whatever power current CPU consumes is fine.

feipoa wrote on 2022-12-02, 07:55:

Zip file is here: Re: Custom interposer module for TI486SXL2-66 PGA168 to PGA132 - HELP!

🙁 vcc is a flood fill, afaik caps arent effective/doing very much this way
PS: oh man, kicad still as user friendly as gimp 😀 Had to resort to Altium to see flood fills. Turns out Zones by default after install was set to 0 😮

feipoa wrote on 2022-12-02, 07:55:

I don't quite follow the suggestion here. I understand 'desolder regulator' and 'bend OUT leg out' so that it doesn't go into the via, then solder the other 4 pins back as normal, and don't solder OUT back to the via? Instead, only connect OUT to the adjacent tantalum with a wire? I could do that. I have a lot to try....

Yes, so the LDO output doesnt go into the PCB but instead is externally wired to tantalum cap. Capacitor pads need to be over the current path between LDO and CPU, just like they look in the schematic. Same goes for the Feedback pin (adj) of the regulator, it cant be just plopped wherever 🙁. Lift the pot and wire pin 1 (VCC) with another wire to LDO Output.

VCC pins seem to all be on the outside, so they could all be pulled to the outside one by one, then small caps and connection to main VCC net. 22 VCC CPU pins, put them into groups of three and we end up with 7 small caps around the upper socket. Looks doable.

Ground is fine, after all its pass-thru pins from the motherboard.

We could go pwm and use a murata module?

Reconfigure for an external power supply and use calamitylime’s inline power supply.

See pcb projects link in sig

I am a little confused by the two differing remarks concerning bending the VRM's pins. Previously you talked about disconnecting and bending the OUT pin (also known as pin 4, VCC3, or OUT),

rasz_pl wrote on 2022-12-02, 12:29:

Now that I look at the output cap its also stitched to power plane instead of being between output and load? If thats the case desolder regulator, bend output leg out, solder it back, solder wire going from output leg to Cout.

But in your latest comment, you talk about disconnecting pin 1, which is the ENABLE pin, but on the PCB is tied directly to IN (also known as pin 2, VCC5, or IN).

rasz_pl wrote on 2022-12-02, 12:29:

Yes, so the LDO output doesnt go into the PCB but instead is externally wired to tantalum cap. Capacitor pads need to be over the current path between LDO and CPU, just like they look in the schematic. Same goes for the Feedback pin (adj) of the regulator, it cant be just plopped wherever :(. Lift the pot and wire pin 1 (VCC) with another wire to LDO Output.

Could you please clarify your suggesting using this naming convention:
Pin - Function
1 - EN
2 - IN
3 - GND
4 - OUT
5 - ADJ

Thanks a lot.

feipoa wrote on 2022-12-02, 21:24:

I am a little confused by the two differing remarks concerning bending the VRM's pins. Previously you talked about disconnectin […]
Show full quote

I am a little confused by the two differing remarks concerning bending the VRM's pins. Previously you talked about disconnecting and bending the OUT pin (also known as pin 4, VCC3, or OUT),

rasz_pl wrote on 2022-12-02, 12:29:

Now that I look at the output cap its also stitched to power plane instead of being between output and load? If thats the case desolder regulator, bend output leg out, solder it back, solder wire going from output leg to Cout.

But in your latest comment, you talk about disconnecting pin 1, which is the ENABLE pin, but on the PCB is tied directly to IN (also known as pin 2, VCC5, or IN).

rasz_pl wrote on 2022-12-02, 12:29:

Lift the pot and wire pin 1 (VCC) with another wire to LDO Output.

pin 1 of potentiometer, sorry for being ambiguous

LDO pin 4 bend outwards, connected with solid thick wire directly to Cout VCC
R1 pot pin 1 lifted and connected to same wire, this one can be thin

Idea is they should be connected together before going into the flood fill 3V power plane, still not ideal but better than now. You can also experiment with additional caps soldered on top of Cout, Imo the ones in the middle of socket arent doing much.

rasz_pl wrote on 2022-12-03, 06:48:

pin 1 of potentiometer, sorry for being ambiguous […]
Show full quote

feipoa wrote on 2022-12-02, 21:24:

I am a little confused by the two differing remarks concerning bending the VRM's pins. Previously you talked about disconnectin […]
Show full quote

I am a little confused by the two differing remarks concerning bending the VRM's pins. Previously you talked about disconnecting and bending the OUT pin (also known as pin 4, VCC3, or OUT),

rasz_pl wrote on 2022-12-02, 12:29:

Now that I look at the output cap its also stitched to power plane instead of being between output and load? If thats the case desolder regulator, bend output leg out, solder it back, solder wire going from output leg to Cout.

But in your latest comment, you talk about disconnecting pin 1, which is the ENABLE pin, but on the PCB is tied directly to IN (also known as pin 2, VCC5, or IN).

rasz_pl wrote on 2022-12-02, 12:29:

Lift the pot and wire pin 1 (VCC) with another wire to LDO Output.

pin 1 of potentiometer, sorry for being ambiguous

LDO pin 4 bend outwards, connected with solid thick wire directly to Cout VCC
R1 pot pin 1 lifted and connected to same wire, this one can be thin

Idea is they should be connected together before going into the flood fill 3V power plane, still not ideal but better than now. You can also experiment with additional caps soldered on top of Cout, Imo the ones in the middle of socket arent doing much.

Thanks for the clarification. This is what I was assuming you meant. Let me know if you still think it worth testing this after presenting the following. Since I needed to remove the VRM, I decided to run the interposer w/CPU installed using a benchtop variable DC power supply. I ran this test in 1x mode. The rest of the motherboard is powered by my AT power supply.

I set the variable supply for 4.0 V, which was the minimum voltage this CPU needed to run at 80 MHz. The scope showed only around 3.3 Vrms:

I added a jumper cable between the scope/power supply/interposer ground to the molex, this didn't help much, so I just upped the power supply's voltage to 4.8 V so that the scope showed about 4.0 Vrms. This matches the conditions of the previous tests.

Here are the results closer up. The low frequency noise w/external linear variable power supply showed a 192 mV p-p. The noise demonstrated previously using the selected MIC29302WT voltage regulator was 200 mV. The oscillation frequency of the noise is around 2.5 KHz, which is a bit higher. This may be due to swapping the motherboard or oscillator. I'd have to check with the MIC29302WT soldered on again.

And here is the high frequency noise. The frequency is half the crystal oscillator value, as expected and the amplitude around 80 mV, or about half the amplitude as the interposer with MIC29302WT soldered on.

In short, the overall amplitude of the noise observed when using a benchtop power supply to power the SXL2 CPU was the same as with the MIC29302WT soldered on. These results seem to take blame away from the MIC29302WT regulator. And perhaps point more towards either the PCB's trace-trace interference, the situation with the power planes on the PCB, or perhaps the cap type or placement?

I re-routed the VRM's OUT pin and trimmer pin1 per the latest suggestion, but it didn't help any. I also changed Cin to that of a 10 uF ceramic rather than 10 uF tantalum. No change in noise.

are we supposed to be comparing against Final_Scope_40x1_MHz_LOW_freq_AMPLITUDE.JPG ?

it does look tad lower, 5% maybe within measurement error
hmm this

and

both beat at ~2KHz?

rasz_pl wrote on 2022-12-03, 12:49:

are we supposed to be comparing against Final_Scope_40x1_MHz_LOW_freq_AMPLITUDE.JPG ?

Yes, we're looking at the 1x multiplier mode noise. However, I have swapped motherboards since that photo was taken. There was too much goofing around on a rare Symphony board for my liking, so I'm using one of those MS-3131 UMC 481/482 boards we fixed on eevblog many years ago. This is why the frequency is a bit different now, that is, different motherboards and different crystals yield slightly different low frequency waveforms. Noise is still in the 200 mV ballpark.

To re-establish a baseline, I will need to run the MIC29302WT on this UMC 481/482 board once I'm thru with the LP38503TSX-ADJ.

I then swapped the MIC29302WT re-routing configuration for the LP38503TSX-ADJ because they are pin compatible except for one N/C pin. I suppose if someone wants to use the LP over the MIC, it is probably fine, but you'll want a 1K 0805 resistor and a 10 K-ohm PV36 trim pit, rather than the 121-ohm resistor and 500 ohm PV36 trim pot used with the MIC. Here are the results. Noise at 160 mV. It might be better or it might not. I will need to check again with MIC29302WT on this motherboard, however maybe the re-route hack waveform captured earlier is sufficient? If so, maybe we lost 40 mV.

rasz_pl wrote on 2022-12-03, 12:49:

both beat at ~2KHz?

Between the MIC29302WT and the external power supply? Yes.

Now im stumped. I theocratized this low frequency noise was inherent to this particular LDO, but its evidently motherboard dependent.
Hmm, can it be memory controller pausing cpu to refresh memory? This would result in regular low frequency load jumps.
How to test this theory? I wonder, bios keeps running after detecting no ram because boards generate those annoying 'no ram' beep codes. Wonder what the scope would look like if you pulled all the ram.
This still doesnt sound all that plausible given frequencies, 800-2000Hz, refresh cycles are usually every 64/8192 = 7.8 microsecond (~130KHz). Is there anything else in PC halting CPU at ~1KHz?

At least now we know two things
-I was wrong, this LDO is not oscillating
-"problem", if you want to call it that (after all it works just fine), is with CPU decoupling and follows cpu bus load

So how do we fix it.

Do I flip the center cap orientation?

Add two more caps below on front

Maybe this is a good thing..? Maybe all those ground planes I used are picking up stray EMI just like I intended and it is just ending up on the ground plane.

Can you somehow tell if that is the case?

Idk, like the thing works. In fact it works just as well at least as a oem unit… I’m not sure what to make of this power plane noise thing…

Btw, your bench psu test sounds like a farce. You had serious voltage drop it sounds like. Do it again with shorter, thicker wires please.

Also, please scope your bench top psu under equal resistive load as a control.

Hard to say anything concrete by just looking from the side, but usually adding a ferrite bead in serial to Vcc and something like 10-to-100mF capacitor to ground before or after it ?

pshipkov wrote on 2022-12-03, 16:56:

Hard to say anything concrete by just looking from the side, but usually adding a ferrite bead in serial to Vcc and something like 10-to-100mF capacitor to ground before or after it ?

inductor would only work if noise was coming from power supply (inductors block AC)

Sphere478 maybe something like this? yes it would make it wider on 3 sides to accommodate extra bypass caps :

feipoa looking at photo from the side all the VCC3 pins are not obscured by bottom socket at all, bypass caps can be soldered directly to those pins externally on the bottom. 7-8 small ceramic caps.

but as it works I dont expect any design respins 😀

Okay, but hear me out.

Try improving the voltage drop on that bench psu, with voltage drop like that it’s not going to hold any kinda voltage, that’s nuts.

Also, maybe the power planes are noisy because they are picking up stray EMI from the signals. And if so, isn’t that the whole point? Could that be a good thing? I basically designed this interposer with so many ground planes that the traces may as well be coaxal cables 🤣 (which should be good) and I have doubts that the oem interposers went through this much trouble. But maybe they did.

A lithium battery near the interposer would hold that voltage pretty well. They have low resistance. Like 4 of them in parallel. As a control.

In any case, it seems that we are probably over analyzing this. A real problem would have real world effects and with 80mhz result I remind you that we succeeded.

But I share feiopa’s sentiment that, if we can make it better, let us.

To that end, feiopa, add caps to the pins themselves. Everyfreakingwhere. Show us what works and I will try to make it. But before all that we need to be damned sure the bench psu control test was valid. So put the scope on that psu with a resistor and analize it to see if it is good enough to use, and do the test again with very thick very short wires

And try a lithium battery.

The tests may reveal that it is indeed the interposer design, and that is fine but we need to be sure our test methodology is sound.

I may be able to redesign the center portion to have more caps near the perimeter. And also flip vcc3 180 or maybe 4 spokes instead of 1

We may be able to shoehorn some caps onto the actual edge of the pcb kinda like bumps. But it’ll look really weird 🤣.

But adding them to the surface of the edge is gonna cut out a fair amount of conductive cross section there.

Perhaps two cut vias with a edge cut down into between them could recess the caps flush into the pcb only taking up the thickness of the cap.

Still gonna loose a fair amount of conductive cross section to this though. Including clearances, etc. and these are gonna be very small caps btw

oh yeah, those caps I drew on the outside could be on the inside, I think this is how that commercial one is routed, every 2-3 power pins first go to the pad in the middle with a cap and only then connect to main 3V rail.

Too bad that there are so many traces packed in the middle, otherwise we could do all sorts of crazy stuff like vertically mounting them straight through the pcb 🤣. That would be cool.

Just double checking, no one has a busted oem unit that they are willing to sand down and photograph the layers do they?

rasz_pl wrote on 2022-12-03, 19:15:

inductor would only work if noise was coming from power supply (inductors block AC) […]
Show full quote

pshipkov wrote on 2022-12-03, 16:56:

Hard to say anything concrete by just looking from the side, but usually adding a ferrite bead in serial to Vcc and something like 10-to-100mF capacitor to ground before or after it ?

inductor would only work if noise was coming from power supply (inductors block AC)

Sphere478 maybe something like this? yes it would make it wider on 3 sides to accommodate extra bypass caps :
lolpass.jpg

feipoa looking at photo from the side all the VCC3 pins are not obscured by bottom socket at all, bypass caps can be soldered directly to those pins externally on the bottom. 7-8 small ceramic caps.

but as it works I dont expect any design respins :)

It is slowly becoming more apparent that the issue somehow stems from the PCB or its cap placement. I think unless we can show some massive improvement with the existing alpha boards using whatever hackery, there would be little value in redesigning the PCB - unless someone else is paying for them. lol.

Rasz_pl, Yes, there are several points for which I could solder some caps to the bottom pints. I hope 0805 fits here. I'll see what sizes I have. What capacitance, 47 pF, 10 nF, 47 nF, 100 nF? And should I remove the centre region caps? I currently have a 10 uF cap for Cin because the LP VRM requests below 500 mOhm ESR.

Sphere478 wrote on 2022-12-03, 16:28:

Do I flip the center cap orientation?

No. Checked before I got started. I also took scope photos without those caps installed.

Sphere478 wrote on 2022-12-03, 16:28:

Idk, like the thing works. In fact it works just as well at least as a oem unit… I’m not sure what to make of this power plane noise thing…

Not quite. Were you able to read the wall of photos? Using the same CPU, the OEM unit could overclock with 0.05 V less than the replica. I wouldn't call it a problem.

Sphere478 wrote on 2022-12-03, 16:28:

Btw, your bench psu test sounds like a farce. You had serious voltage drop it sounds like. Do it again with shorter, thicker wires please.

I see no value to gain from this. Using two different VRMs will have the shortest "leads" possible and the same oscillations occur. I am already using my shortest cable, which is probably falling apart at the clip unions due to age. However, if I were to hack together a 5" cable, what would it show? It would probably show less voltage drop, but the oscillations would still be there. Could a 2-ohm wire resistance cause oscillations like this, no, but they would create that deltaV.

Sphere478 wrote on 2022-12-03, 16:28:

Also, please scope your bench top psu under equal resistive load as a control.

Sphere478 wrote on 2022-12-03, 21:30:

we need to be damned sure the bench psu control test was valid. So put the scope on that psu with a resistor and analize it to see if it is good enough to use, and do the test again with very thick very short wires

I'm pretty sure you are barking up the wrong tree here. I did this 5 years ago, and I my mind was much sharper back then. Refer to the resistive load measurements presented earlier for trials of different VRMs. If it is absolutely do or die for you, I'll run it again, but I feel it is a waste of effort.

The last thing I will do, after adding direct pin caps, is compare the MIC against the LP on this motherboard. If there's even a 50 mV improvement, we can use the LP instead, but they don't have a mounting hole option for the tab.