With the new config the pin header on the latest (not yet uploaded design) is 0.3556mm closer to cpu socket than the one you have.

the mio and so are now closest to socket. it required moving them to vias

Edit: place save

Sphere478 wrote on 2022-12-13, 01:28:

90* header is a good idea

I am missing something here.
Can you guys clarify where this is going ?
Thanks.

pshipkov wrote on 2022-12-13, 07:39:

I am missing something here. Can you guys clarify where this is going ? Thanks. […]
Show full quote

Sphere478 wrote on 2022-12-13, 01:28:

90* header is a good idea

I am missing something here.
Can you guys clarify where this is going ?
Thanks.

You could use either type.

-

The latest place save above:

I added ground vias around the edge to better stitch together the perimeter of the ground plane. I believe the ground plane should be pretty solid again.

I wish I could make the 5v plane a little more robust.

Need to add silk over the new vias still.

The design as it is does not prevent us from using p shaped pins if one finds it feasible anyway.
It does not need any modes. Or am i missing the point ?

feipoa wrote on 2022-12-13, 02:39:

rasz_pl, thanks for your help here, but I don't really follow. It has been some time for me, but if I recall, we need something like 2.5x the bandwidth of the frequency under observance even begin to resolve it. There's a name for that law/rule, but I forget it. The signal I am looking at is 0.7 - 5.0 KHz in in the 50 mV range. In this case, wouldn't we want to use the x1 probe so that we are not attenuating our voltage 10x to the scope? And we are well withing the 6 MHz range of this probe. I think this is the message presented in that eevblog youtube post you directed me towards.

It all depends on
-if you know what you are doing
-what you want to measure
This is not without its own problems tho, you might end up measuring signals that dont exist, are aliases of other signals, artefacts of the method or tool used, etc

feipoa wrote on 2022-12-13, 02:39:

Regarding antenna's and picking up stray signals - when I probe my DC variable supply with a 3' long unshielded cable, I do not see any waveforms.

was that cable also going over cpu interposer?

feipoa wrote on 2022-12-13, 02:39:

Here yellow is VCC3 with standard probe on x10, blue is VCC3 with x10 and low inductance GND probe. Is it because we have cut the voltages seen by the scope 10-fold that we are seeing this extra noise in the yellow signal, when compared with x1 standard probe? Low inductance probe cut out a lot of noise:

That noise was never there to begin with. Metrology is its own branch of science. Sadly its not as simple as press probes receive bacon.
Pages 8-10 "MR. MURPHY’S GALLERY OF HIGH SPEED AMPLIFIER PROBLEMS" https://www.analog.com/media/en/technical-doc … otes/an47fa.pdf Dont worry about the title, LDO is also an op amp if you look deep enough 😀

feipoa wrote on 2022-12-13, 02:39:

Low inductance GND probe also made the waveform more resolvable, it is really 5 KHz?

It might also not exist 😀 First it was 800Hz, then 2.6KHz, now 5KHz. It might just be artifact of the way you are measuring and the tool used. You are effectively measuring rf circuit (40MHz), nothing is obvious.

feipoa wrote on 2022-12-13, 02:39:

I then put CH1-yellow back to x1, but keep CH2-blue at x10 w/low inductance. The Vpp noise measured between the two is fairly agreeable. It seems to me that without a low inductance GND probe, it is better to use a standard probe in x1 for these measurements if you are after Vpp:

Switching to 1x limits BW to 6MHz filtering all the EMC interference picked by bad probing technique. In effect you are measuring double wrong, but two wrongs cancel each other out leaving you with non obviously garbage result. Im sorry that I cant provide a golden quick fix/method. This is a PhD hard problem. Maybe look for a local gray beard HAM to help you? I know only enough to mostly be able to recognize when im doing something wrong 🙁

feipoa wrote on 2022-12-13, 02:39:

Next, I use the low inductance GND probe in x1 mode and tell the scope it is now x1. What is the true noise of Vcc3 then? Is it the low inductance probe at x1 32 mV, or in x10 mode at 72 mV:

the fact this seemingly 5KHz beating somehow magically goes down when using more precise mode should tell you something is not right here. Might be looking at a ghost, might be influence of different probe impedance (real LDO oscillation due to too low output esr).

feipoa wrote on 2022-12-13, 02:39:

Still with 50 mV scale, I put the low inductance probe back on x10 mode. Is the real noise in VCC3 68 mV or 30 mV? From what the eevblog youtuber mentioned, it seems like x10 is attenuating the voltage too much and the scope cannot realise these small voltages properly. Please correct my understanding if I am not following:

if that was the case 10x mode would be showing lower (attenuated) noise level, at least In theory assuming tool used (scope, probe) is calibrated correctly and works properly.

feipoa wrote on 2022-12-13, 02:39:

I won't have time to watch the FFT videos until past midnight, but I have quickly grabbed what I think you were asking for. FFT/Hanning is part of the Math menu, not the probe menu. I took one shot at 1x zoom and Vrms, the other at 10x zoom and dbVrms.

You should start by grabbing whole max spectrum span your scope supports, 100MHz? This will tell you whats out there and which frequencies noise dominates.

Sphere478 wrote on 2022-12-13, 08:35:

I added ground vias around the edge to better stitch together the perimeter of the ground plane. I believe the ground plane should be pretty solid again.

on the outside will prevent radiating out, while it might be more beneficial in the middle between tracks
layer int2 and int3 still neighboring running tracks in parallel directly above each other. Can you either swap int1 with int2 or int3 with int4?

Sphere478 wrote on 2022-12-13, 08:35:

I wish I could make the 5v plane a little more robust.

I wish you would stop worrying about it ;P. Plot twist - remove 5V altogether including socket pins, use new free real-estate (https://www.youtube.com/watch?v=cd4-UnU8lWY) for moar ground and finally properly clearanced cap pads, add external 5V input from molex 😀

I don’t agree about the 5v plane unless we plan to power it externally. But you can continue to try to make your case. I am trying to be open minded. I am listening, agreeing isn’t guaranteed :p

Good call on the layer swap. done.

🤔 thoughts?

🤔

progress save^🤔

-

Would it be better to reverse fan polarity? accidental fan connection to it? 5v or gnd through fan to memw safer?

Other than 5V 😜 and imo excessive 5V caps looks good. Without 5V caps would be one solid middle ground island for swapping deeper inside to provide even more isolation. But its just nitpicking, I doubt signal integrity is a problem when first prototype worked great.

Sphere478 wrote on 2022-12-14, 06:08:

Would it be better to reverse fan polarity? accidental fan connection to it? 5v or gnd through fan to memw safer?

both bad, ground less bad

I was figuring.

Not really a good way to dummy proof it though. Earlier I spaced them out so it was impossible. No room now 🤣.

rasz_pl, attached are the FFT results cropped to the periods of interest w/x10 probe. Expecting something at 60 Hz, 5 KHz, 40 MHz, and 80 MHz and this is pretty much what we get.

full bandwidth:
40 MHz: 3.75 Vrms
80 MHz: 7.5 Vrms
90 MHz: 1.75 Vrms

low freq:
60 Hz: 25 mVrms
5 KHz: 15 Vrms

I ran the variable DC supply again with 1 metre shielded cables connected to the scope probe. I then moved the tip of the scope probe around the motherboard to see which component is generating the 5 KHz. When I move the probe tip over the FPU, BIOS, UM82C206F, or UM82c482, I see the same 5 KHz waveform on the scope. Vpp increase with increasing distance to the component, e.g. from 100 mV to 175 mV. When I move the probe tip over the SXL2, 80 Mhz oscillator, or UM82C481 I do not see the waveform. Isn't there a crystal on these motherboards used to set the data/time which runs in the KHz range?

I also looked at the variable DC supply's noise at output (w/out 1 metre cable) and its noise is about the same as the interposer at regulator, or 55-70 mV Vpp with x10 low inductance probe.

Sphere478 wrote on 2022-12-14, 07:57:

I was figuring.

Not really a good way to dummy proof it though. Earlier I spaced them out so it was impossible. No room now 🤣.

RETVRN to the molex KK connector for the fan

Excellent investigative work with the scope feiopa.

maxtherabbit wrote on 2022-12-14, 13:07:

Sphere478 wrote on 2022-12-14, 07:57:

I was figuring.

Not really a good way to dummy proof it though. Earlier I spaced them out so it was impossible. No room now 🤣.

RETVRN to the molex KK connector for the fan

I don’t believe we have room for it anymore

give this a day or two. if no changes order for prototyping.

I will be surprised if anyone uses a 90 degree header. It makes the unit a bit cumbersome. There's also a chance it hits something on the motherboard, like an adjacent reset/speaker/LED connector on the motherboard. As my best performing SXL2 chip has the heatsink offset by 1mm in the affected direction, I will solder the cable to the via rather than use or install the header.

Anyone who has played around enough with 386 CPUs can attest to having inserted them in the incorrect orientation, no matter how much attention they thought they paid to the proper orientation prior to insertion. Once installed incorrectly, if not powered off in time, something breaks.

Once the SXL2 is installed in the interposer, the top view of the pin-1 triangle marker is no longer visible. The most important white triangle market is near the trimmer. I've drawn it in as shown. Your first thought might be 'but there's triangles on the bottom!' haha, yeah, but once you flip the chip over to install it, you'd be amazed and how it often people still mess up orientation installation.

Yeah there doesn’t seem to be much that I can do about that triangle without making the PCB bigger

I think just place it under the trimmer. Some of it will still show.

If someone wants to make the voltage fixed and not use the trimmer, it will really be visible then.

feipoa wrote on 2022-12-15, 02:39:

I think just place it under the trimmer. Some of it will still show.

If someone wants to make the voltage fixed and not use the trimmer, it will really be visible then.

I could print resistor values for fixed voltages if you want, there.