Fixing 1.8V power rail design error


65 posts   Page 1 of 3   1, 2, 3
by Douglas W. Jones » Tue Aug 14, 2012 3:10 am
With reference to the Switching Regulators for the Raspberry Pi thread, one conclusion reached was that the board has a design error in the connection of the output of RG1 (the 1.8V regulator) to the 1.8V pin on the LAN9512 chip. See:
-- http://www.raspberrypi.org/phpBB3/viewtopic.php?p=146154#p146154
The above thread references further discussion of this in:
-- http://www.element14.com/community/thread/19236

The consequence of this design error is that, on some Raspberry Pi boards, the LAN9512 chip runs hot because its regulator outputs slightly more than the set-point voltage of RG1, therefore turning off RG1 and supplying the entire board's demand for 1.8V. On those boards, the LAN chip runs very hot (too hot to touch, according to some reports). On other boards, RG1 outputs slightly more than the set-point of the built-in regulator on the LAN chip, therefore turning off that regulator and letting the LAN chip run cold. In fact, of course, these two 1.8V subsystems should be isolated from each other.

A brief aside: For practical purposes, it is unlikely to the point of impossibility that any two regulators will have exactly the same output, so when two regulators are put in parallel, unless there is enough resistance in the circuit to divide the load between them, one will always win over the other and do all of the work.

Response 14 by tmackay, Jul 27, 2012 11:28 PM, in this thread
-- http://www.element14.com/community/thread/19236
proposes a solution. There's a photo in the thread, but here's the text:
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)

I've reworked boards with surface-mount components before, and judging by tmackay's photo and my reading of the schematics, his rework is doable (for someone with a very steady hand). So, to my questions:

A: Has anyone else tried this to cure a Raspberry where the LAN chip runs hot?

B: Is there a better solution? This is a question that someone with access to the layout of the inner board layers must answer.

My examination of the board photos (my board has yet to arrive) shows vias from the 1.8V pins on the LAN chip to capacitors on the flip side, and to nowhere else. Obviously, the 1.8V power connection is somewhere in an inner layer of the board. That's why the most useful answers to question B will come from Raspberry Insiders. If only I knew the layout of the inner layers, I'd know if there was some spot where a hole could be drilled through the board, cutting the 1.8V power between RG1 and the LAN chip while not cutting anything else.

Another aside: Drilling a multilayer board is a fairly standard way to cut a trace in an inner layer -- so long as the trace is small enough to cut with a drill without cutting any other trace. If the hole goes through a ground or power plane, you should scrape the inside of the hole with a dental tool to prevent short circuits. Even with the scraping, it's practical to modify hundreds of boards per hour this way, once you set up your drill press with a board alignment jig.

-- Doug Jones
-- jones@cs.uiowa.edu
User avatar
Posts: 13
Joined: Thu Jul 26, 2012 4:06 pm
Location: Iowa City, Iowa, USA
by Burngate » Tue Aug 14, 2012 10:02 am
Unfortunately (!) my 9512s don't run hot.
But I'm in the process of researching the mechanics of this.
According to the schematics pdf, 1v8 is on layer 5, at least at one point on the board. However, at at least one point, it's on the top surface, according to the front-page post of 26 December 2011 by Eben.
It only goes to the SoC (13 points) and the 9512 (2 pins), with 8 100n caps and 2 larger ones around the SoC, 2 100n caps and a 4u7 around the 9512

Yes, a view of the gerbers would be good. The picture from 14 November 2011 isn't quite good enough to see what's going on. So answers from Delphi would be better than my own research.
Wyszkowski's Second Law: Anything can be made to work if you fiddle with it long enough.
Brain surgery is easier than psychoanalysis
User avatar
Posts: 2907
Joined: Thu Sep 29, 2011 4:34 pm
Location: Berkshire UK
by jamesh » Tue Aug 14, 2012 10:04 am
Cannot say much about this, except that the Foundation are aware of the problem and are working to rectify it.
Soon to be employed engineer - Hurrah! Volunteer at the Raspberry Pi Foundation, helper at PiAcademy September 2014.
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 11922
Joined: Sat Jul 30, 2011 7:41 pm
by Burngate » Tue Aug 14, 2012 10:26 am
I've finally found the video of the "red wire" fix - http://www.youtube.com/watch?feature=pl ... AcT5370Va4

And thank you, jamesh, I assumed they are, but confirmation is good.
I'm only doing this out of curiosity!
Wyszkowski's Second Law: Anything can be made to work if you fiddle with it long enough.
Brain surgery is easier than psychoanalysis
User avatar
Posts: 2907
Joined: Thu Sep 29, 2011 4:34 pm
Location: Berkshire UK
by redhawk » Tue Aug 14, 2012 10:34 am
Fortunately my LAN doesn't hot either but it would be nice to know how to tell which boards are affected.

Is it a Farnell issue or are boards shipped by RS also affected??

Richard S.
Last edited by redhawk on Tue Aug 14, 2012 11:01 am, edited 1 time in total.
User avatar
Posts: 3519
Joined: Sun Mar 04, 2012 2:13 pm
Location: ::1
by bredman » Tue Aug 14, 2012 10:38 am
Can I ask why you are trying to fix this?

I understand that it is unfortunate that on some boards the LAN chip takes responsibility for providing the 1.8v rail, but

1. Is there any evidence that the LAN chip is overheating beyond its design limits?

2. Is there any evidence that the LAN chip cannot provide enough power on the 1.8v rail? Note that RG1 is available if the LAN chip regulator is stressed.
Posts: 1413
Joined: Tue Jan 17, 2012 2:38 pm
by jamesh » Tue Aug 14, 2012 10:47 am
To be clear, I severely doubt there is any necessity for fixing already delivered Raspi's
Soon to be employed engineer - Hurrah! Volunteer at the Raspberry Pi Foundation, helper at PiAcademy September 2014.
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 11922
Joined: Sat Jul 30, 2011 7:41 pm
by Burngate » Tue Aug 14, 2012 11:22 am
In most cases it's not necessary.

I assume (with zero evidence) that a fix will consist of changing the board. So all Pis made after a certain date will be ok.
Boards made before that date mostly will work. Some few which don't could be dealt with by RMA. The ones that will fall in the cracks are the ones which have been delivered and fail due to reduced life-time of the 9512. Already some few people have been complaining of very hot 9512s - can they now get it replaced, or are they stuck?

My interest is in the Pis exhibiting no symptoms - like mine. I'd just like to find a way to cure the disease, not treat the symptoms, which it isn't showing.
Wyszkowski's Second Law: Anything can be made to work if you fiddle with it long enough.
Brain surgery is easier than psychoanalysis
User avatar
Posts: 2907
Joined: Thu Sep 29, 2011 4:34 pm
Location: Berkshire UK
by TarjeiB » Tue Aug 14, 2012 11:42 am
Would be great if a manual fix is provided though, when it is found - having something run hot means there's power wastage, and my Pi will be set up to run unattended on solar power with little to spare :) And my 9512 chip is definately running hot, hotter than the CPU.
Posts: 130
Joined: Thu Jul 12, 2012 3:33 pm
by carlosfm » Tue Aug 14, 2012 3:00 pm
Yes, my lan chip has always run hotter than the CPU, even with some hard core overclocking...
That's why I heatsinked both chips.
Most find it unnecessary but in my case the lan chip gets really hot.
Do you Pi?
Posts: 132
Joined: Fri Oct 21, 2011 3:23 pm
Location: Lisbon, Portugal
by hojnikb » Tue Aug 14, 2012 4:22 pm
Just out of curiosity, what is 1v8 rail used for anyway ?

My Pi's lan/usb chip runs hotter than the SoC, but i can easily touch it as its not that hot.
Also there are no hangs when i copy stuff from lan to usb stick. But i do get interrputs on my keyboard sometimes..

So bottomline, should i be worried in any way becouse of this ??
I dont want a fried Pi on my desk :cry:
+°´°+,¸¸,+°´°~ Everyone should have a taste of UK Raspberry Pie =D ~°´°+,¸¸,+°´°+
Rasberry Pi, SoC @ 1225Mhz :o, 256MB Ram @ 550Mhz, 16GB SD-Card, Raspbian
User avatar
Posts: 112
Joined: Mon Jun 04, 2012 3:59 pm
Location: @Home
by Burngate » Tue Aug 14, 2012 5:26 pm
If you can hold your finger on it, it's probably ok (lots of posts about the finger test, though!)
Most Pis appear to not have any symptoms, and will be ok.

It does lots of stuff inside the SoC. In fact the other power rails seem to do only peripheral stuff.
It doesn't go anywhere else - bar the 9512, which is our pet problem

Regarding the occasional keyboard stutters, that's possibly a separate problem - search "elephant"
Wyszkowski's Second Law: Anything can be made to work if you fiddle with it long enough.
Brain surgery is easier than psychoanalysis
User avatar
Posts: 2907
Joined: Thu Sep 29, 2011 4:34 pm
Location: Berkshire UK
by mahjongg » Tue Aug 14, 2012 5:41 pm
This is all very unfortunate, and especially because I have told people countless times that cooling of the LAN9512 chip should not be necessary, :oops: I suppose I have to eat my own words now as obviously I wasn't counting on the chip acting as the 1V8 regulator! :cry:

Its also that "being hot" is such a subjective thing, and not many people will have two PI's, let alone one that runs hot and the other one that doesn't. And if they do how big is the chance that they will notice, and make a remark about it in the forum?

In a classroom filled with PI's this would have become obvious in no time.

Oh well... I have been assured that this problem will be fixed, and meanwhile the effects of it seem not to be incredibly dramatic for the functioning of the PI. Although they might have some effect on USB error rates if the LAN9512 gets far too hot (like "what do I smell" kind of hot), I do not get the impression that is the case in most situations.

Battle of the regulators, who would have thought! :mrgreen:
User avatar
Forum Moderator
Forum Moderator
Posts: 5696
Joined: Sun Mar 11, 2012 12:19 am
by Montala » Tue Aug 14, 2012 5:49 pm
Frantically starts searching for purpose made LAN chip cooling heatsinks! :D
User avatar
Posts: 638
Joined: Mon Mar 05, 2012 11:54 pm
Location: Herefordshire (U.K.)
by jojopi » Tue Aug 14, 2012 7:14 pm
Douglas W. Jones wrote:A: Has anyone else tried this to cure a Raspberry where the LAN chip runs hot?
B: Is there a better solution? This is a question that someone with access to the layout of the inner board layers must answer.
I am curious what metric you are using for "better" here. Even if there did exist a single point that could be drilled to separate 1V8 into two nets (which seems unlikely, since this is power we are talking about, so it should be widely tracked and/or planed), I would consider that worse (in the sense: more risky) than simply cutting the three surface tracks leading to the relevant vias, as per tmackay's test.

I am not surprised that the results so far suggest only modest temperature difference from the mod. The 9512 is supposed to dissipate over 0.75W in a tiny area, so it already runs pretty warm. The rest of the board must draw much less than 300mA at 1V8, since that has been quoted as the total current for a Model A. (tmackay measured up to 80mA after the mod.) And the VDD18CORE regulator is presumably not less than 55% efficient, or it would have been linear. So the LAN might run 15% warmer than it should...
User avatar
Posts: 2122
Joined: Tue Oct 11, 2011 8:38 pm
by mahjongg » Tue Aug 14, 2012 9:44 pm
not to put a too fine point on it, but you must mean that it just (at maximum I presume) dissipates 15% more energy. If that linearly translates to a 15% higher temperature is an interesting question.
User avatar
Forum Moderator
Forum Moderator
Posts: 5696
Joined: Sun Mar 11, 2012 12:19 am
by mahjongg » Wed Aug 15, 2012 12:23 am
Montala wrote:Frantically starts searching for purpose made LAN chip cooling heatsinks! :D
Use "VGA-RAM" heatsinks, like this
Image
They seem to be perfectly suitable for this task. Most come self adhesive, so you can just stick them to the LAN9512 chip.

This is one of the many sources fo them. http://www.zerotherm.net/eng/product/ZH100.asp

Here is a picture of a PI equipped with VGA-RAM heatsinks of the "cooltek" brand.
Note that both the LAN and the Broadcom chip have heatsinks fitted.
Which makes sense as a set of eight costs just eight euro.
Image
User avatar
Forum Moderator
Forum Moderator
Posts: 5696
Joined: Sun Mar 11, 2012 12:19 am
by Douglas W. Jones » Wed Aug 15, 2012 1:18 am
jojopi wrote:I am curious what metric you are using for "better" here. Even if there did exist a single point that could be drilled to separate 1V8 into two nets (which seems unlikely, since this is power we are talking about, so it should be widely tracked and/or planed),

"Battle of the regulators." I like that.

What is my metric? tmackay's solution, response 14 in this thread:
-- http://www.element14.com/community/thread/19236
involves cutting 3 traces, adding 2 wires, and abandoning 2 small capacitors to ground (C36, C43) on the 1.8V supply. A solution that cuts fewer traces, adds fewer wires, or preserves more of the proper capacitors to ground would be better.

If the 1.8V supply was used by more chips, yes, I'd expect it to be a plane, but it appears that there are only two major customers for power from RG1, the SoC itself and the LAN chip. With only two customers, there's a possibility that it's not a full plane, but rather, a local plane in the vicinity of the SoC and another in the vicinity of the LAN chip. Since I haven't taken sandpaper to a board to expose the inner planes (that's how you reverse engineer a multilayer board), I can only speculate.

-- Doug Jones
-- jones@cs.uiowa.edu
User avatar
Posts: 13
Joined: Thu Jul 26, 2012 4:06 pm
Location: Iowa City, Iowa, USA
by Lob0426 » Wed Aug 15, 2012 2:21 am
mahjongg wrote:This is all very unfortunate, and especially because I have told people countless times that cooling of the LAN9512 chip should not be necessary, :oops: I suppose I have to eat my own words now as obviously I wasn't counting on the chip acting as the 1V8 regulator! :cry:

You better start printing @mahjongg, cause I am holding you to that, and your not eating air. :lol: ;)

Oh my the world has strange ways of getting back at people! :D

Most people are still not going to need "cooling". Out of the three I have checked none of them were much different than the others. So either all of them are bad or they are all good. The replacement will get here later this week and I do not expect problems with it either. Just think none of this will be a problem on the A model! Of the few that do end up, with real hot RasPii, most will not notice it as they do not "fondle" their hardware. Some will return them and the rest might heat sink them. It will all work out in the end I am sure. A fix will probably be done in the future.
512MB version 2.0 as WordPress Server
Motorola Lapdock with 512MB
Modded Rev 1.0 with pin headers at USB

http://rich1.dyndns.tv/
(RS)Allied ships old stock to reward its Customers for long wait!
User avatar
Posts: 1940
Joined: Fri Aug 05, 2011 4:30 pm
Location: Susanville CA.
by Tooms » Wed Aug 15, 2012 6:38 am
Hi

For your talk about how hot it is getting under different loads then try see this post there is showing some tests and the IR images of it.

http://www.element14.com/community/message/57800#57800
Posts: 12
Joined: Wed Aug 08, 2012 5:06 pm
by hippy » Wed Aug 15, 2012 1:04 pm
Lob0426 wrote:Most people are still not going to need "cooling". Out of the three I have checked none of them were much different than the others. So either all of them are bad or they are all good.

They are all bad; but whether that affects anything in a practical way remains to be seen.

The 1V8 pins on the LAN chip are neither specified to provide power to external circuits nor to be powered from external circuits. Period - According to the LAN chip manufacturer.

While some people do not seem to be encountering problems with either excessive heat or operation the affect of running out of spec for a prolonged period is unknown and the situation may change over time. While those suffering excessive heat may not see any problems now ( other than heat ) that could also change in future.

The thing is no one really knows what will happen. The boards out there now might not be technically "good" but may well be "good enough". Notionally it shouldn't really matter if the issue turns out to be a problem as every R-Pi owner so far should be well aware they have a "Development Board" and there are implied risks in buying such a thing which they have accepted. Whether buyers actually understood exactly what it was they were buying is a different matter.

Ironically the prototype boards for EMC testing had an opposite flaw; the 1V8 rail not connected to supply the SoC. I don't know if the 1V8 regulator was already connected to the LAN or came about as a consequence of fixing that. Either way it doesn't really matter as we have to deal with what we have.

The pragmatic way forward IMO is to fix for future boards, make sure those can be identified from the current. Trust to luck that the ones in the field keep working, adding heatsink if necessary, and where there are problems note it as the nature of the beast. Document the fix for those who want to apply it. I'm sure some enterprising people could even make a business out of delivering that for others.

I'm not opposed to licensee manufacturers replacing boards ( and would encourage that as good PR ), but, so long as they made it clear customers were getting "Development Boards" with all which comes with that, they probably have no legal obligation to replace.
Posts: 820
Joined: Fri Sep 09, 2011 10:34 pm
by Badger101 » Wed Aug 15, 2012 1:41 pm
Can you not just cut the track from the VDD18CORE output of the LAN952 I.C. and let RG1 power everything?

Badger.
Posts: 16
Joined: Tue Jun 12, 2012 1:26 pm
by Burngate » Wed Aug 15, 2012 2:56 pm
Badger101 wrote:Can you not just cut the track from the VDD18CORE output of the LAN952 I.C. and let RG1 power everything?
Badger.

Yes, if you know which of the two pins of the 9512 is the output. The pdf only says "Refer to the LAN9512/LAN9512i reference
schematics for connection information." But I can't find those! :roll:
Wyszkowski's Second Law: Anything can be made to work if you fiddle with it long enough.
Brain surgery is easier than psychoanalysis
User avatar
Posts: 2907
Joined: Thu Sep 29, 2011 4:34 pm
Location: Berkshire UK
by harpingon » Wed Aug 15, 2012 5:06 pm
My board has a LAN chip that runs mad hot. I don't know whether to add heatsinks (might that void any warranty?) or just to not worry. In case of board failure, we have warranty/SOGA so no worries there - except for the heatsink question.

Mine's in a clear perspex case (Yoctopuce) so I think I'm inclined to leave it until it either fails, or doesn't.
Posts: 2
Joined: Wed Aug 15, 2012 4:53 pm
by hippy » Wed Aug 15, 2012 5:46 pm
Burngate wrote:
Badger101 wrote:Can you not just cut the track from the VDD18CORE output of the LAN952 I.C. and let RG1 power everything?
Badger.

Yes, if you know which of the two pins of the 9512 is the output. The pdf only says "Refer to the LAN9512/LAN9512i reference schematics for connection information." But I can't find those! :roll:

Looking at Figure 2.2 in the SMSC datasheet the two VDD18CORE pins are shown as connected internally, so there's no separate input and output.
Posts: 820
Joined: Fri Sep 09, 2011 10:34 pm