Low ripple low noise small power supply design

[VEGETA]

I can hook up resistors, but I wanted electronics load since it is kinda better.

Here there are no scopes like this except in universities and no one sells them.

[BrianHG]

I can hook up resistors, but I wanted electronics load since it is kinda better.

You would need 3 of them.  Or a triple channel one...

[VEGETA]

I have made a small change which is L2 and L3 of the EMI filter before the switchers... look like we got the wrong footprint! 0603 in kicad got a very small part which is not like the 06s in JLCPCB...

long story short: I changed it to SMD,2.5x2.0x1.2mm package (slightly bigger size) and it is now 2.2uH instead of 1uH. I updated the PCB and now doing the panel again

here is JLCPCB part: https://lcsc.com/product-detail/Power-Inductors_Sumida-252012CDMCDDS-2R2MC_C351245.html/?href=jlc-SMT

it is 3 amps so I guess it is fine.

[BrianHG]

I have made a small change which is L2 and L3 of the EMI filter before the switchers... look like we got the wrong footprint! 0603 in kicad got a very small part which is not like the 06s in JLCPCB...

long story short: I changed it to SMD,2.5x2.0x1.2mm package (slightly bigger size) and it is now 2.2uH instead of 1uH. I updated the PCB and now doing the panel again

here is JLCPCB part: https://lcsc.com/product-detail/Power-Inductors_Sumida-252012CDMCDDS-2R2MC_C351245.html/?href=jlc-SMT

it is 3 amps so I guess it is fine.

Yes, for filtering the input, that inductor will clear as the one for the 5v will max out at ~2 amps and less for the 3.3v side.  If it was for the switcher's output, that coil would saturate and you would loose efficiency and add ripple to your output.

[VEGETA]

I have made a small change which is L2 and L3 of the EMI filter before the switchers... look like we got the wrong footprint! 0603 in kicad got a very small part which is not like the 06s in JLCPCB...

long story short: I changed it to SMD,2.5x2.0x1.2mm package (slightly bigger size) and it is now 2.2uH instead of 1uH. I updated the PCB and now doing the panel again

here is JLCPCB part: https://lcsc.com/product-detail/Power-Inductors_Sumida-252012CDMCDDS-2R2MC_C351245.html/?href=jlc-SMT

it is 3 amps so I guess it is fine.

Yes, for filtering the input, that inductor will clear as the one for the 5v will max out at ~2 amps and less for the 3.3v side.  If it was for the switcher's output, that coil would saturate and you would loose efficiency and add ripple to your output.

the way I understood it is this:

5v*2 amps max = 10 watts. 10 watts/12v = 0.83 amps input current.
3.3v*3 amps max or so = about 10 watts = 0.83 amps input current.

but of course, switching current is gonna be more.

is it this case or not?

[BrianHG]

I have made a small change which is L2 and L3 of the EMI filter before the switchers... look like we got the wrong footprint! 0603 in kicad got a very small part which is not like the 06s in JLCPCB...

long story short: I changed it to SMD,2.5x2.0x1.2mm package (slightly bigger size) and it is now 2.2uH instead of 1uH. I updated the PCB and now doing the panel again

here is JLCPCB part: https://lcsc.com/product-detail/Power-Inductors_Sumida-252012CDMCDDS-2R2MC_C351245.html/?href=jlc-SMT

it is 3 amps so I guess it is fine.

Yes, for filtering the input, that inductor will clear as the one for the 5v will max out at ~2 amps and less for the 3.3v side.  If it was for the switcher's output, that coil would saturate and you would loose efficiency and add ripple to your output.

the way I understood it is this:

5v*2 amps max = 10 watts. 10 watts/12v = 0.83 amps input current.
3.3v*3 amps max or so = about 10 watts = 0.83 amps input current.

but of course, switching current is gonna be more.

is it this case or not?

No, you got it right, except needing to add the loss and quisient current of the switcher.
I was maximizing the output current to 3 amps on both 6v and 4.3v outputs.

So, 6v * 2 amp = 12 watts, or 1 amp at 12v.  Or, consider worst case 80% efficiency at high load, and then that source current goes up to ~1.2amps.  I did my ~math with ~3 amps at 6v feeding the linear regulator outputting 5v at 3 amps.  This would mean that @3amps, the mosfet will radiate ~3 watts of heat.

Remember, your switcher's output is +1v and whatever you draw from the linear side at -1v from there still has the same current draw on the switcher +1v side.

[VEGETA]

So according to this, I think about doing a future enhancement in BOM by using smaller and noticeably cheaper power inductors instead of that big one to reduce total cost. However this is in the future.

getting 2 of that small 2.2uH inductor instead of the big 10uH switching one seems ok in terms of current since each one tolerates 3 amps which makes it 6 amps total. However, 2.2+2.2 = 4.4uH instead of 10uH. I am afraid this won't remove ripple as much. getting 5 of it will be more than 10uH and still maybe less or similar footprint but more expensive.

I won't bother with this now though. Right now I am contacting JLCPCB for the panels.

the damn 22uF 1206 caps costs about 182$ (quantity of 4000 xD) for the 10 panels (250 boards). Big inductors about 77$.

[BrianHG]

So according to this, I think about doing a future enhancement in BOM by using smaller and noticeably cheaper power inductors instead of that big one to reduce total cost. However this is in the future.

getting 2 of that small 2.2uH inductor instead of the big 10uH switching one seems ok in terms of current since each one tolerates 3 amps which makes it 6 amps total. However, 2.2+2.2 = 4.4uH instead of 10uH. I am afraid this won't remove ripple as much. getting 5 of it will be more than 10uH and still maybe less or similar footprint but more expensive.

I won't bother with this now though. Right now I am contacting JLCPCB for the panels.

the damn 22uF 1206 caps costs about 182$ (quantity of 4000 xD) for the 10 panels (250 boards). Big inductors about 77$.

What does the datasheet say about the 10uh?  Don't you need 10uh to filter the ~2MHz?

Careful.  The filter to supply the switchers is different than the inductors at the switcher's output.  The inductors at the switcher's output needs to support the DC load current + drive input switching current without saturating.  This is why in the datasheet for the 4 amp switcher, they recommend a 6-8amp inductor when you draw the full 4 amps.

As for the source power inductors, you only need around 1-2.2uh to remove the high frequency spikes generated from the edge of the switching mosfet inside the switcher ICs.  The 10uf/22uf ceramics before and after are there to short as much as possible at high frequencies to the GND plane.

[VEGETA]

check the datasheet: http://aosmd.com/res/data_sheets/AOZ1284PI.pdf

it recommends 22uH but we chose 10uH. I think it is enough since I don't think final output current is gonna be that big... like 4 amp continuous.

[BrianHG]

check the datasheet: http://aosmd.com/res/data_sheets/AOZ1284PI.pdf

it recommends 22uH but we chose 10uH. I think it is enough since I don't think final output current is gonna be that big... like 4 amp continuous.

You went to 10uH because you increased the frequency.

[BrianHG]

Take a look at the trace I highlighted in RED.
Isn't that a feedback trace?
Isn't that long aster the resistor dividers, going under the second switcher's inductor?
Shouldn't the feedback divider resistors be near the feedback input pin?

[VEGETA]

Take a look at the trace I highlighted in RED.
Isn't that a feedback trace?
Isn't that long aster the resistor dividers, going under the second switcher's inductor?
Shouldn't the feedback divider resistors be near the feedback input pin?

is it necessary now to change it? I mean, we didn't notice that all along past reviews for some reason.

[BrianHG]

Take a look at the trace I highlighted in RED.
Isn't that a feedback trace?
Isn't that long aster the resistor dividers, going under the second switcher's inductor?
Shouldn't the feedback divider resistors be near the feedback input pin?

is it necessary now to change it? I mean, we didn't notice that all along past reviews for some reason.

It wasn't noticed because I cannot tell from just images as I didn't layout the components myself.
It is not necessary to move, it is just common practice when driving a feedback input that any high impedance signal is kept as short as possible for minimal noise.
It is your choice as the change wont affect the circuit's function.

[VEGETA]

Take a look at the trace I highlighted in RED.
Isn't that a feedback trace?
Isn't that long aster the resistor dividers, going under the second switcher's inductor?
Shouldn't the feedback divider resistors be near the feedback input pin?

is it necessary now to change it? I mean, we didn't notice that all along past reviews for some reason.

It wasn't noticed because I cannot tell from just images as I didn't layout the components myself.
It is not necessary to move, it is just common practice when driving a feedback input that any high impedance signal is kept as short as possible for minimal noise.
It is your choice as the change wont affect the circuit's function.

TBH, I lost focus from the various revisions I have done to the board, so couldn't notice these resistors to be the feedback ones. the irony is that the 2nd switcher is done better.

It is better to move them, but now I need to re-pannelize the board which is the real headache. I am gonna try though if I had time.

However, I think we are ok with this being the way it is since it is a switching regulator and even if some little noise got in the signal, it won't matter. The output of this regulator will have ripple due to being a switcher, and it is well above the post-regulator so no problem if it got a tiny extra.

I have learned a lot doing this project, thanks to you.

Here is a quick and not-so-accurate pricing for 125 boards (5 panels):

- boards assembled = 290$.
- shipping them = ~60$ or so.
- aoz1284 switchers (not in stock now) = 75$.
- 1000uf caps = 75$
- other through-hole connectors = 150$.
- mosfets = 40$ or so.
- top-side inductor = 25$.
- heatsinks = ~100$
- extra materials (thermal tape, packaging, solder, etc...) = 100$
- handling and customs = 100$

total = 1015$ -> make it 1000$.

Price for a complete ready-to-ship product = 8$. I planned to sell it for 50$ or similar, maybe 40$ at least.

I still have a lot to do like getting these through hole connectors, especially that stupid 6 pin molex one.

Plus, figuring out where to get the 3d printed connector part like this:

https://www.thingiverse.com/thing:4576696
https://www.thingiverse.com/thing:2779041

which is important to tightly fit the DC jack connector. I tried thinking about a small PCB instead of this but it is not good. How much does this thing cost in your experience? I could buy say 1000 piece at once, or at least 300.

[VEGETA]

As for the 3D printed part, I got an offer from local 3D printing service company. For 1000 part, I pay 250$ means 0.25$ per part which is hugely cheap! I will go with them for sure without thinking too much.

JLCPCB noticed a problem that only the first board had proper ground layer while the others are not... hence I must re-do it which is an opportunity to do the small fix you mentioned.

[VEGETA]

I've done the modification.

[BrianHG]

I've done the modification.

Nice, but separate this 1 trace from the connector's pad. it became so close and this is a stress point on the PCB as you have off PCB wires pulling at a random direction on the power connector pads.

[VEGETA]

Done.

finally the last ever change!

waiting JLCPCB to know how can I proceed. Also, the switcher is still out of stock!

[BrianHG]

finally the last ever change!

[VEGETA]

finally the last ever change!

Right? hhh

on a parallel timeline:

year 2020: vegeta starts the project.
Year 3020:

Vegeta: I think now we are ready.
BrianHG: hmm... look at that trace. I think we need to move this and that.

 

____

I will keep you posted on what happens further on.

[BrianHG]

2 mistakes... You have no connection to the bottom right red VIA and you lost the second top left red arrow via with it's connection to the same net.

That connection can be over an amp, so, 1 via to transfer the 12v power from the top layer to the bottom switcher isn't enough.  Earlier versions of the PCB, we had all 3 vias in parallel to do the job.

( Earlier version where they are wired: https://www.eevblog.com/forum/beginners/small-mosfet-for-capacitance-multiplier-in-a-small-psu/msg3352100/#msg3352100 )

[VEGETA]

done

[VEGETA]

BTW, I was checking what we have chosen for the mosfet and found this:

https://lcsc.com/product-detail/Transistors-NPN-PNP_ON-Semiconductor-ON-MJD44H11T4G_C38780.html

it is an NPN transistor not a MOSFET. DPAK package and pins are compatible.

Pin1 -> gate.
Pin2 -> input.
Pin3 -> output.

remind me of our choice if you remember.

[BrianHG]

Ok, now,

Red arrows, too close to the edge of the PCB.
Move C33 in a little.
As for the trace on the left, you can move C11 to the right a little and the big long +5v trace can be pushed in quite a bit and made vertical straight to the grid.

The 2 purple arrow VIAs are not needed.

The green arrow trace is a little close to your screw drill hole, just push it up a minuscule notch.

[BrianHG]

We switched to the mosfet awhile back.

We engineered a linear 'Source-follower' amplifier meaning the voltage we put into the 'gate' comes out the 'source' - the VGS drop.  The 'drain' has the + supply in this scenario.

I do believe that the 'Gate' is also pin 1 and the 'Source' is pin 3.

If the LM358 was strong enough to drive the base of the transistor, and in a lower power scenario it is, a linear 'emitter-follower' amp would achieve the same functionality with a -0.5v drop from base to emitter.

Dont forget to fix that trace on the left too close to the edge.

(Arrrrrrrrgggggg -> The page count is over 9000!!!!)