Switching regulators for Raspberry Pi

[azerty]

There's been some discussion on the topic of replacing the Raspberry Pi's linear regulators with higher efficiency switching regulators. godFather89 has a nifty solution for his Lapdock project, but I wanted to try making more of a drop-in replacement.

As you can see, RG1 and RG2 are replaced with switching regulators...is what I'd like to say, but all you can see from this picture is that RG1 and RG2 look funny, and that my phone's camera has an awful noise reduction algorithm. (Sorry. Maybe I can get a better picture tomorrow.)
So, how much power does it save? Assuming that the 1.8V and core rails draw a trivial amount of current when the processor is idle (a lousy assumption), the biggest offender is the 3.3V LDO (RG2). Since 3.3V is 2/3 of 5V, 1/3 of input power is lost as heat in RG2, and a 33% decrease in power consumption is the asymptote to reach for. Of course this all changes when the processor/RAM are active.
Actual measurement shows that when idle, an unmodified Pi draws 358mA, while the Pi with switching regulators draws 276mA, for a saving of 27%. This translates to 30% longer runtime for battery powered devices, well, if they spend all their time idle.
Is it worth it? Probably not. The LAN9512 is a huge power hog so I suspect a Model A with USB Wifi will probably result in even lower power consumption. But, if you absolutely need to squeeze out every last minute of battery life at any cost, it's worth considering.

[RaTTuS]

has your phone got a macro switch ?
and what ones did you buy [link for the lazy]

[alexeames]

Interesting. Thanks for sharing. Bit less improvement than we thought isn't it? Was it hard to do?

[cowpat]

I was reviewing some of the discussion on the regs tonight, a lot of it seems to hinge on this being a bit hard without any SMT tools.

I don't have any SMT tools - but i do have a nice pair of sidecutters. Other than cosmetic reasons, why not just cut/lift the legs and leave it in situ? - the big metal that needs the heat is just 0v isn't it?

[ksangeelee]

... RG1 and RG2 are replaced with switching regulators...

What chips did you choose to replace the linear regulators?

[azerty]

Hmm...not great, but I think you can get a better idea of what the module looks like.

...and what ones did you buy [link for the lazy]

I should have made it more clear, but there are no switching regulators in the same package as the linear regulators available on the market, otherwise someone would have done it months ago. Basically I made a PCB that "breaks out" a switching regulator and the necessary capacitors/inductor to DPAK/SOT-223 footprints. That said, the 3.3V regulator is a TI TPS62291, and the 1.8V regulator is a Murata LXDC2HL18A.

Was it hard to do?

Replacing the regulators is actually pretty easy, the trick is to snip the leads with a wire cutter, then apply enough heat to the metal tab. Designing, depanelizing and assembling the modules, on the other hand, is quite a bit of work.

I don't have any SMT tools - but i do have a nice pair of sidecutters. Other than cosmetic reasons, why not just cut/lift the legs and leave it in situ? - the big metal that needs the heat is just 0v isn't it?

Correction -- the metal tab is actually the output node, but your method would still work. I imagine it would be easier too since there's less of a size constraint. I did it the way I did it mainly for the challenge.

[alexeames]

Basically I made a PCB that "breaks out" a switching regulator and the necessary capacitors/inductor to DPAK/SOT-223 footprints. That said, the 3.3V regulator is a TI TPS62291, and the 1.8V regulator is a Murata LXDC2HL18A.

You can see that better now. I could have sworn Gert said something at one time about a drop-in replacement? Was it in the power thread or somewhere else?

[AndrewS]

Basically I made a PCB that "breaks out" a switching regulator and the necessary capacitors/inductor to DPAK/SOT-223 footprints.

Impressive!
Any plans to sell these pre-assembled mini-PCBs?

[RaTTuS]

awesome - nicely done

[TonyD]

Great stuff. Any plans to post the design files online?

[Lob0426]

I have been looking at the 5v version of these.
http://www.rutronik.com/uploads/media/R ... 042011.pdf
I have a project where I would like to power from 12v. It would replace the stock unit and let me use 12v input. The input currents needed are pretty high. They cannot be used to replace the existing regulators unfortanely and they are expensive. Cost just under $11 at mouser for the 5v. I have not been able to find LDO versions that could be used in place of the existing RasPi regs.

[AndrewS]

Might be cheaper / easier to use separate 12V->5V and 5V->3.3V regulators?

[Lob0426]

Might be cheaper / easier to use separate 12V->5V and 5V->3.3V regulators?

That regulator could be put in place of the 5v (RG2) reg. It can take an input voltage up 30v and puts out 5v@1amp. I do have some adjustable switching regulators on seperate boards. But this would be a lot smaller footprint and just look better all around. It is only slightly bigger than a 7805.

[AndrewS]

Ahhhh, the old size vs. cost challenge...

[Lob0426]

Ahhhh, the old size vs. cost challenge...

Ah so you want me to use the easy old LM2596 adjustable regulator boards, that I already have, instead of taking the iron to one of my Pi's! So where in the heck is the fun of that? Hack it together with soldered wires and boards instead of the Elegant, plug it into the stock Micro USB (albeit it will be 12v instead of 5v). And at the miniscule cost of like 27% the total cost of the board? Shame AndrewS, Shame!

[Douglas W. Jones]

Several people here have misstated the functions of RG1 and RG2 so let's get that out of the way first. Reading the schematic carefully:

RG2 is a NCP1117-3V3 regulator to drop 5V to 3.3V, dissapating at most 800mW. This implies that the maximum current on the 3.3V supply is under 470mA.

RG1 is a NCP1117-1V8 regulator to drop 3.3V to 1.8V. The schematics don't say enough to let me guess the current on the 1.8V rail, but it's obviously less than the 3.3V rail, since that's where it gets it's share of the power. Also, the wattage is apparently much lower.

Anyway, if I were redesigning the board and the parts weren't too expensive, I'd substitute an MCP1603-330 switching regulator for RG2. This thing is from 80 to 90 percent efficient for current draws from 1 to 500mA (peak efficiency is at around 100mA. This buck regulator needs a 4.7uH inductor, you can get these as surface mount dice, and I think the MCP1603 plus the inductor could fit in the same pc board footprint as the current RG2. I am not sure any additional capacitors would be required, since the current schematic surrounding RG2 has plenty in the places where the MCP1603 datasheet asks for them.

Doug Jones
jones@cs.uiowa.edu

PS: While I am listed as coauthor on one Microchip applications note, I have no affiliation with the company and my suggestion of their regulator is only because a web search found that part first. I am sure other parts are available.

[Lob0426]

@ Douglas W. Jones: I was one those people. RG2 is indeed 3.3 volts. I ordered a 5v (RECOM R-78C5.0- 1.0) DC/DC converter to supply RG2 with 5v. I also ordered (RECOM R-78C3.3- 1.0) to replace RG2. I will have to build a board, then wire it to the RasPi. I plan on putting 12v into that board. The R-78C3.3 will work to replace the Linear with a switching, but you then have no 5v for your USB. Also you need to input 7v to feed the R-78C3.3. So then I need a R-78C5.0 to supply 5v which needs 8v input. The easier fix for 12v application to RasPi is the LM2596 boards that I have. But who wants easy.

The solution you have is to replace the Linear with Switching regulators which is the original topic. This would lower the consumption of the board for battery, solar and other low power applications. I hope someone a bit more electronics savvy can move forward with a regulator switch. It would be a great tutorial. The 500ma should be more than sufficient especially since RG1 would be replaced also.

[carlosfm]

@ Douglas W. Jones: I was one those people. RG2 is indeed 3.3 volts. I ordered a 5v (RECOM R-78C5.0- 1.0) DC/DC converter to supply RG2 with 5v. I also ordered (RECOM R-78C3.3- 1.0) to replace RG2. I will have to build a board, then wire it to the RasPi. I plan on putting 12v into that board. The R-78C3.3 will work to replace the Linear with a switching, but you then have no 5v for your USB.

You should feed the RPi with 5V, or else you will need another 5V reg for the USB ports.
Presently I use a 12V switching PSU, which I have tweaked here and there to reduce the noise, then an RC filter feeding a 7805, which has a big heatsink, then it enters the Pi though a 2.5mm transformer plug (I don't like the Micro USB plug, it was a bad idea IMO, it doesn't handle the necessary current), then a ferrite bead directly to TP1 (which bypasses F3).
The USB polyfuses were replaced with SMD ferrite beads and then 10uF ceramics to ground, on the power pins of each USB output.
These 10uF ceramics were removed from the analog output, because ceramics are some of the worst types for coupling caps (the other ones are tantalum).
The analog outputs have proper non-polarized 10uF electrolytics.
I also have heatinks on the CPU, the network chip, and RG2.
Why all this?
Well, in my book you need a stable, silent PSU, and good capacitance where needed.
The 7805 and the other onboard regs are not brilliant, far from that, but switching regulators are usually noisier - it very much depends on the implementation, and space is at a premium here.
Also, cool running improves stability and long time reliability.
I have my Pi running stable at 1100Mhz (CPU), 500Mhz (GPU) and 550Mhz (Ram).
This would not be possible with a lowsy mobile phone charger - those have copious amounts of ripple and the output voltage is not stable.

[Lob0426]

@ Douglas W. Jones: I was one those people. RG2 is indeed 3.3 volts. I ordered a 5v (RECOM R-78C5.0- 1.0) DC/DC converter to supply RG2 with 5v. I also ordered (RECOM R-78C3.3- 1.0) to replace RG2. I will have to build a board, then wire it to the RasPi. I plan on putting 12v into that board. The R-78C3.3 will work to replace the Linear with a switching, but you then have no 5v for your USB.

You should feed the RPi with 5V, or else you will need another 5V reg for the USB ports.
Presently I use a 12V switching PSU, which I have tweaked here and there to reduce the noise, then an RC filter feeding a 7805, which has a big heatsink, then it enters the Pi though a 2.5mm transformer plug (I don't like the Micro USB plug, it was a bad idea IMO, it doesn't handle the necessary current), then a ferrite bead directly to TP1 (which bypasses F3).
The USB polyfuses were replaced with SMD ferrite beads and then 10uF ceramics to ground, on the power pins of each USB output.
These 10uF ceramics were removed from the analog output, because ceramics are some of the worst types for coupling caps (the other ones are tantalum).
The analog outputs have proper non-polarized 10uF electrolytics.
I also have heatinks on the CPU, the network chip, and RG2.
Why all this?
Well, in my book you need a stable, silent PSU, and good capacitance where needed.
The 7805 and the other onboard regs are not brilliant, far from that, but switching regulators are usually noisier - it very much depends on the implementation, and space is at a premium here.
Also, cool running improves stability and long time reliability.
I have my Pi running stable at 1100Mhz (CPU), 500Mhz (GPU) and 550Mhz (Ram).
This would not be possible with a lowsy mobile phone charger - those have copious amounts of ripple and the output voltage is not stable.

That is the crux of the problem with the RasPi. It relies upon the the power suply for way too much. The parts I quoted above have ripple somewhere between 240 to 350Hz. That is why I plan to mount them off board, so I can put on the capacitors to calm the ripple. I will mount a R-78C5.0- 1.0, 5v and a R-78C3.3- 1.0, 3.3v to the board and feed both 12V from a wall wart. The LM2596's seem to be pretty stable and would make a good alternative to taking a soldering iron to the RasPi itself for most people. I also am running heat sinks on RG2, SoC and the LAN9512. I want to run the a wireless router and the RasPi from the same power source(12v). My first attemp will be with the LM2596. I will experiment from there.

Currently I am Running RasPi #1 from a single Cable (power and data) from a modified USB hub.
http://www.raspberrypi.org/phpBB3/viewt ... 28#p143328
RasPi #3 is running on a Lapdock. The pictures show # 1 but it has moved on to the next project!
http://www.raspberrypi.org/phpBB3/viewt ... 93#p143293
RasPi #2 is waiting on an RMA for return. Luckily #3 was a ready backup at this point.
For all the fault that has been found about the Raspberry Pi, it is still an amazing little device. Except fo #2 they run really well and are easy to modify. Even #2 ran well on the Lapdock using an USB HDD. Its problem was with a fault in the SD card interface someplace.

[alexchamberlain]

People may be interested in these... http://www.dimensionengineering.com/products/de-sw033
or these... http://shop.tuxgraphics.org/electronic/ ... chreg.html

[chaoz2]

I have ordered a pair of adp3338 in 3.3v and 1.8v versions, i think, aside of switch psu, could be a valid solution, pin to pin compatible, vastly better than ncp1117.

http://www.analog.com/en/power-manageme ... oduct.html

am i wrong?

[Douglas W. Jones]

... The 500ma should be more than sufficient especially since RG1 would be replaced also.

If you replace both RG1 and RG2 with switching regulators, then you should also make RG1 take its input from 5V instead of 3.3V. There are two reasons for this:

a) This lowers the current through RG2, which currently supplies RG1. That design is sensible with linear regulators, but with switchers, it lets RG2 work closer to the "sweet spot" for the efficiency of 5V to 3.3V switching regulators, which is around 100mA for a regulator with maximum current capacity of 500mA.

b) RG1 will have about equal efficiency whether powered from 5V or from 3.3V. Power it from 5V, and you don't pay an efficiency penalty for dropping 5V to 3.3V first. Run RG1 at perhaps 85% efficiency, instead of running RG1 at 85% in series with RG2 at %85 (for a net efficiency closer to 70% on the output of RG1.

I suspect the net gain is marginal, but it makes sense to do it right.

Another thing to note about the bucking switching regulator ICs I've looked at. They all have overtemp and overcurrent protection built in -- that is, the regulator will turn off if it gets too hot, and it will turn off if the output current goes too high. If you put one of these regulators on the input to a system that's powered, for example, by 10 to 18 volts, does that eliminate the need for a fuse?

-- Doug Jones
-- jones@cs.uiowa.edu

[Tooms]

Hi

The guys of this thread has discovered that the RG1 1.8V LDO is not having any load and are not doing any thing and this is because that inside the LAN/USB chip LAN9512 there is also an 1.8V LDO and the two LDO's output is linked and the LAN9512 is having an bit higher output then the RG1, so this mean that the RG1 is not doing any thing and the LAN9512 is providing 1.8V to the board.
This is an design error because the 1.8v pins on the LAN9512 is design to the caps and not power external things...

The thread is here
http://www.element14.com/community/thread/19236

To test this i have add some extra switchmode LM2596 modules in paralle with the RG1 and RG2 and then set the LM2596 output volt an small bit higher then the LDO's on the board is having and by this make sure the LM2596 was doing the work and not the LDO's.
That saved around 0.5 watt and the LAN9512 is around 8c cooler because it is no longer powering the boards 1.8v rail.
see my post 40 in this thread
http://www.element14.com/community/message/57800#57800

so just replacing the RG1 with another LDO/Switchmode at 1.8V is not working, you have to raise the volt level an bit over the 1.8v to take the load off the LAN9512 or have to cut the 1.8V lines from the LAN9512 off.
When taking the load of the LAN9512 buildin 1.8v LDO and let an external switchmode module do the load then i have seen an saving of around 80mA power.


Tooms

[Douglas W. Jones]

The guys of this thread has discovered that the RG1 1.8V LDO is not having any load ...
This is an design error because the 1.8v pins on the LAN9512 is design to the caps and not power external things.

-- http://www.element14.com/community/thread/19236

I read that entire thread, and the second thread it referenced. In summary, there is a genuine and acknowledged design error in the Pi. Some boards have the LAN9512 output voltage just epsilon higher than RG2, so the LAN chip runs very hot, supplying the 1.8V power to the entire Pi. Other boards have RG2 providing epsilon more volts, so it powers the Pi and the LAN chip runs cool.

Officially, the LAN chip is not supposed to supply or sink any power, and its 1.8V pin is just there for filter capacitors. In fact, I've found datasheets for voltage regulators that suggest putting regulators in parallel with exactly this kind of setup, except that the epsilon voltage difference is carefully controlled. Do this, for example, to power your circuit from battery when external power is not available, and from external power while the battery is recharging. Looking at regulator circuit diagrams (both linear and switching), most will tolerate this kind of cross connection without damage.

But, it's clearly not good to power the entire Pi board from the LAN chip's onboard regulator. One user posted a board mod that requires a very steady hand on the soldering iron to fix the design error:

-- http://www.element14.com/community/thread/19236

See response 14 ty tmackay, Jul 27, 2012 11:28 PM. I quote:

Carefully, I cut the circuit traces from pins 15 and 38 on LAN9512 and the 1.8V side of C29 to the vias. I then very carefully soldered some wire between these three points. This more closely resembles how the circuit should be (minus the 100nF caps). (The 9512 reference design says 4u7 should be by pin 38 btw)

He included a photo and he said that his Pi worked no worse after making this change than before, along with measurements that make it clear that he successfully isolated the two 1.8v supplies from each other.

-- Doug Jones
-- jones@cs.uiowa.edu

[Lob0426]

... The 500ma should be more than sufficient especially since RG1 would be replaced also.

If you replace both RG1 and RG2 with switching regulators, then you should also make RG1 take its input from 5V instead of 3.3V. There are two reasons for this:

a) This lowers the current through RG2, which currently supplies RG1. That design is sensible with linear regulators, but with switchers, it lets RG2 work closer to the "sweet spot" for the efficiency of 5V to 3.3V switching regulators, which is around 100mA for a regulator with maximum current capacity of 500mA.

b) RG1 will have about equal efficiency whether powered from 5V or from 3.3V. Power it from 5V, and you don't pay an efficiency penalty for dropping 5V to 3.3V first. Run RG1 at perhaps 85% efficiency, instead of running RG1 at 85% in series with RG2 at %85 (for a net efficiency closer to 70% on the output of RG1.

I suspect the net gain is marginal, but it makes sense to do it right.

Another thing to note about the bucking switching regulator ICs I've looked at. They all have overtemp and overcurrent protection built in -- that is, the regulator will turn off if it gets too hot, and it will turn off if the output current goes too high. If you put one of these regulators on the input to a system that's powered, for example, by 10 to 18 volts, does that eliminate the need for a fuse?

-- Doug Jones
-- jones@cs.uiowa.edu

Thanks for you reply:
That makes sense. But on RasPi that means you are going to have to "wire up" a power rail. So you would have to leave Vin not connected to RasPi just to your power rail. So you would be running wires, which is what I will have to do also. I would really like to see a straight, pin compatible, component replacement like is in the original post of this thread!
In my particular case:
The main problem with the R-78c's is that their needed input voltage is high. The R-78c3.3 needs at least 7v input and the R-78c5.0 needs at least 8. So I cannot put the 3.3 behind the 5v. They will both have to be powered by the 12v.
There are other options appearing in here now. The R78c's might end up on other electronics projects and something else might be used instead. Or the R78c5.0 might end upstream of something with a low drop out. Keeping an eye here!
The R78c's are over volt and temp protected, but I think that is for input on overvolt not Vout. I do not know about the others. I will have to check the Datasheet for mine. So I think I will leave F3 in place just in case, for now. Really the best option for me is to use the adjustable DC/DC converters I also have, at least until an "on board mod" solution is found!