RG1 1.8v regulator

Just to expand on my original observations, the negative current I see on the output of RG1 seems to be due to it outputting 1.79v for a few seconds on initial powerup before stabilising at 1.804v. While the LAN9512 appears to power up at 1.805v and stay there.

My second order of Raspberry Pis arrived earlier in the week. A quick check shows one of them is running slightly warmer than the rest, not too much, I can still hold my finger on the LAN chip. I measure 1837mV at RG1 compared to 1793mV on one of the cooler ones (at idle, unmodified boards). Perhaps the LAN9512 core is running over voltage due to the external source and hence a bit hotter in this case. There may be multiple scenarios depending on components.

I don't expect it will reduce the life of the chip unless it gets really hot, so I'll leave the rest alone unless LAN chips start dying all over the place.

Troy Mackay wrote:

 

My second order of Raspberry Pis arrived earlier in the week. A quick check shows one of them is running slightly warmer than the rest, not too much, I can still hold my finger on the LAN chip. I measure 1837mV at RG1 compared to 1793mV on one of the cooler ones (at idle, unmodified boards). Perhaps the LAN9512 core is running over voltage due to the external source and hence a bit hotter in this case. There may be multiple scenarios depending on components.

I don't expect it will reduce the life of the chip unless it gets really hot, so I'll leave the rest alone unless LAN chips start dying all over the place.

That's consistent with my limited knowledge of voltage regulators.  My understanding is that if two positive linear regulator outputs are connected to the same node, whichever has the higher voltage reference will supply all the current.  The one with the lower reference will shut off, satisfied that it doesn't need to do any work.  RG1 probably has more precise regulation than IC3 (LAN), so it's more likely that +1V8 voltage above 1.800V is sourced by IC3 due to manufacturing variance.  Feel free to correct my understanding if I've got it worng.

My understanding is that if two positive linear regulator outputs are connected to the same node, whichever has the higher voltage reference will supply all the current.  The one with the lower reference will shut off, satisfied that it doesn't need to do any work. 

I'm not convinced that one will shut off. When you look at the equivalent circuit for these lm1117 style regulators in the datasheets there's nothing particularly clever looking in the design.

Simple ohms law on the other hand gives a good idea that when two are connected in parallel the one with the lowest resistance path will probably end up supplying the most current - at least until a higher voltage offsets the resistance difference.

When we're talking about something that has a large power/ground plane under it then it's likely going to be a toss-up on manufacturing tolerances - which regulator is a couple of mV higher or lower.. Unlikely anything will ever practically manage to keep at exactly 1.8000000000v across temperature and load variations, so things may even swap over as the board warms up, or when you stress the GPU playing some video etc.  Every board will be different and everyone will have different ambient temperature, peripherals, cpu usage, whatever..

John Beetem wrote:

 

RG1 probably has more precise regulation than IC3 (LAN), so it's more likely that +1V8 voltage above 1.800V is sourced by IC3 due to manufacturing variance.

Just confirming that does indeed appear to be the case. I popped RG1 off and still measure 1837mV. IC3 still feels about the same temperature wise.

edit: Scratch that last statement. Playing back video over network, IC3 is running considerably hotter, at the limit of my pain threshold, my finger is still sore. Guess RG1 was still providing some relief. Replaced RG1 and it's still a bit warmer than before, might need a touch more solder.

Hi Troy & all

Since this thread and Coder27's thread are actually all about the same issue, can we recombine or something.  Alternatively I'll limit my replies to the one or other so at least we can keep the flow going.

People coming to the forum will recognise a 'hot' SMSC if they have one, so that might be the best place?

Comments / Ideas?

Playing back video ramps the current requirement of the BCM considerably, again with RG1 o/c can you measure the surface temp of the SMSC?

Cheers

Pete

Nice to see you here again, Pete.

My finger is quite toasty now from all the finger tests on my very hot LAN9512.  Looking forward to your input on this!

Morgaine.

I came across a document "Standards for School Premises" at

  http://www.plymouth.gov.uk/standards_for_school_premises_dfes-3.pdf

that says:

   Hot Surfaces   

   83. Some pupils are more at risk than others from hot surfaces. In special

    schools, nursery schools and in teaching accommodation used by nursery

    classes, the surfaces of radiators and exposed pipes located where they

    might be touched by a pupil must not become hotter than 43 degrees C.

This probably isn't directly relevant to electronic devices, but it might suggest

that uncovered electronics with blisteringly hot chips may be considered a

safety hazard in schools.

In California, hot surfaces of 140 degrees F (60 degrees C) are considered

a safety hazard in the workplace.

http://www.dir.ca.gov/oshsb/hot_surfaces_isor.pdf

If you recall our "Finger Contact Survival Time" (:-) testing over in this thread , although initial contact temperature on my Pi was such that pure instinctive body reaction tore the finger away in a fraction of a second to avoid harm, subsequent fingers survived contact for progressively longer periods, indicating that the LAN9512 has low heat capacity and/or high thermal resistivity in its surface material.

Those tests tell me that it's not really a danger in the safety sense.  The temperature is below ignition point for all common materials, and there's not enough of a heat reservoir there to maintain its temperature if coupled to anything that is thermally conducting.  The main danger to humans is probably from the instinctive reaction:  a schoolkid's sudden violent withdrawal of their hand could knock out another kid's eye or break something.

The other thing that those tests tell me is that the main problem doesn't seem to be too much heat being generated but not enough heat being conducted away through the solder connections, because the surface of the device isn't able to conduct nor radiate much heat away at all.

For reference, link to post with thermal image showing the 65.1C peak temperature on LAN9512 .

PS. Because of the low heat capacity and/or high thermal resistivity of the surface material, a thermocouple attached to the surface is likely to indicate a lower temperature than a thermistor.  The thermocouple is entirely metallic and hence very likely to be a good conductor of heat away from the device.  A non-contact measurement like Remy's is best here.

Morgaine.

Yes, I recall the "Finger Contact Survival Time" discussion,

but I'm not sure how relevant it is. As you say, if you pull away

the first finger in a fraction of a second, you can use subsequent

fingers for longer.  But the safety concern is whether the first

finger will get burned if not pulled away in a fraction of a second,

because younger children, older adults, and the disabled are

assumed to have slower reflexes.

Pete,

Here's some additional data.

To hopefully make these reasonably repeatable, here's some detail of the setup I've been using

2012-07-15-wheezy-raspbian.img no firmware/kernel upgrades done.

overscan disabled

1600x1200 Dell 2007FP monitor

all 'idle' readings taken while at a text based login prompt, X has not been started

'playing video' readings taken after running startx and then omxplayer from a terminal to play a 720p h264 mkv file

temperature readings taken with a MCP9803 i2c temperature sensor on top of the lan9512

5v applied via the GPIO header with 16/0.2 cable and I'm arranging that TP1-TP2 reads 5.00v

Ambient, today:

outside 20C

inside 26C

50.3% RH

1005 hPa

at each step I'm giving the board 15mins to reach equlibrium

The blue column at the end is the unloaded voltage reading from RG1 while it's off the board.

Sorry, but I won't be pulling RG1 off that last board, hope you understand why

I have a couple of boards from FN120720xxx FN120722xxx, but as my results are mostly similar I've just picked either end of the range I have along with a couple in the middle.

Pete,

Here's some additional data.

To hopefully make these reasonably repeatable, here's some detail of the setup I've been using

2012-07-15-wheezy-raspbian.img no firmware/kernel upgrades done.

overscan disabled

1600x1200 Dell 2007FP monitor

all 'idle' readings taken while at a text based login prompt, X has not been started

'playing video' readings taken after running startx and then omxplayer from a terminal to play a 720p h264 mkv file

temperature readings taken with a MCP9803 i2c temperature sensor on top of the lan9512

5v applied via the GPIO header with 16/0.2 cable and I'm arranging that TP1-TP2 reads 5.00v

Ambient, today:

outside 20C

inside 26C

50.3% RH

1005 hPa

at each step I'm giving the board 15mins to reach equlibrium

The blue column at the end is the unloaded voltage reading from RG1 while it's off the board.

Sorry, but I won't be pulling RG1 off that last board, hope you understand why

I have a couple of boards from FN120720xxx FN120722xxx, but as my results are mostly similar I've just picked either end of the range I have along with a couple in the middle.

Selsinork

Thank you so much for this great piece of work, really helpful.  Cannot wait to try some FN1204xxxx,FN1205xxxxx boards with the same configuration.

I'll let you know how I get on.

(I wonder if this post will dissolve into the Ether as soon as I post? )

Cheers

Pete

PeteL wrote:

 

(I wonder if this post will dissolve into the Ether as soon as I post? )

You're in luck Pete!  The SHA-512 of the characters in the subject line multiplied by the current distance to Jupiter in furlongs added to the day of the week is not prime!

Or maybe Loki was so stunned at selsinork's awesome chart that he forgot his usual duties.

Morgaine.

Pete, if your examination of Pi design around the LAN9512 leads to board changes in that area, you might wish to consider the problem of self-powered USB hubs attempting to power the Pi upstream through their A-B lead, which typically fails and sends the Pi into a rapid continuous reboot cycle and a consequent rapid power cycling of attached HDMI equipment.

I reported it in a thread here -- http://www.element14.com/community/thread/19316?tstart=0

Unfortunately this affects an official Farnell "Pi Accessory" hub, and I'm sure it's not the only one.

Morgaine.

My IR thermometer arrived yesterday and I was able to take some more measurments. The laser sight is mounted off to the side, so it is necessary to scan around to find the hot spot. Measurements taken after 20min of video playback over network.

FN120508290 The first board the mod was applied to, still in a configuration where I can switch between the 1.8V power sources with a jumper.

  Powered from RG1 exclusively

    RG1 ____mV/IC3 ____mV idle (I forgot to measure this, will edit later as it might give an indication of each regulators ouput as a function of load)

    RG1 ____mV/IC3 ____mV playback

    45.7C

  Powered from IC3 exclusively

    RG1 ____mV/IC3 ____mV idle

    RG1 ____mV/IC3 ____mV playback

    49.1C - hot spot seems to be larger

FN120744907 The second board which seemed to be running a bit warmer from the beginning

  Unmodified (RG1 removed and replaced in previous test. Heat stress may have influenced result)

    1837mV idle

    49.9C playback

  RG1 removed

    1837mV idle

    50.7C playback

  Mod applied

    RG1 1795mV/IC3 1845mV idle

    RG1 ____mV/IC3 ____mV playback

    46.0C - 49.5C (initially 46.0C after 20mins video playback, but it continued to rise for another 5 minutes or so)

I thought I was getting consistent results until that last measurement. Not sure what happenned, maybe it needs more time to equilibriate or there was an action scene. I measured again after leaving it overnight and still read 47.2C and climbing after at least 20 mins _at idle_ (lan plugged in, but no traffic, no video playback), then after another 10 minutes back down to 46.0C. Maybe this is normal, at least we seem to have elliminated the dependance on video playback.

The results seem to suggest that powering the 1.8V rail from IC3 might account for up to 4 degrees higher temperature on IC3. Looks like there might be quite a bit of variation in the LAN9512 chips based on the observed variation of VDD18CORE voltage so some may be affected more so than others.

This doesn't really confirm if this bug is the source of the _really_ hot LAN chips, but my chips don't seem to be in that category anyway (50C seems to be the norm from what I see). Perhaps some chips handle this extra current drain more gracefully than others though.

Here is the second attempt at the mod

My IR thermometer arrived yesterday

I was looking at picking one up, but noticed that even rather expensive ones seem to have a minimum measurement 'spot' of something around 20mm which is rather larger than the lan9512. I was wondering just how that affects the readings..  Presumably they average the total emissions over whatever is in the sensors view somehow.

Like you I don't have a Pi that runs as hot as some reports, and without one it's difficult to form a theory on what the cause is.

selsinork wrote:

 

My IR thermometer arrived yesterday

I was looking at picking one up, but noticed that even rather expensive ones seem to have a minimum measurement 'spot' of something around 20mm which is rather larger than the lan9512. I was wondering just how that affects the readings..  Presumably they average the total emissions over whatever is in the sensors view somehow.

 

Like you I don't have a Pi that runs as hot as some reports, and without one it's difficult to form a theory on what the cause is.

yes the IR meter have there issues and you have to know it to get good readings and i have even an very good Fluke 568 there is on the high side of them and i dont use muchs because they crude in eletronics work and it is better to use an IR cam

The Fluke 568 IR meter i am having is having and IR spot of (the number is is IR spot size and the distance from the surface)

Spot 19mm @ 300mm distance

Spot 18mm @ 900mm distance

Spot 42mm @ 1500mm distance

you also has to hold it at an right angle

next time i will do some testing i will take both some IR images and use the IR meter and let you know what it is showing, but in the mean time try look at the images i have posted there is an graf showing the temparatura across the surface of the chip and you can clearly see the edges of the chip and that there is an very tinny hotsport at the center of the chip.

Tooms

it is better to use an IR cam

I'd love one.. I looked at prices. No way can I afford one

The Fluke 568 IR meter i am having

Yeah, that's the one I was looking at, it was the only one I could find that gave a good explanation of the D:S relationship and that was what decided me not to bother getting one - my dodgy sensor stuck to the top center of the IC is likely more accurate

Has anyone found the emissivity coefficients that we should use for the encapsulant on the chip (0.92 has been suggested)?

I've asked a few chip suppliers FAEs and been met with a blank stare!  I guess that's why I still use thermocouples, but they are messy as I always gunk them up with heatsink compound.

Cheers

Pete

Hi

I am using the 0.95e on my Fluke 568 as this is the default one and it is for black plastic

the 568 build in tabel has:

0.93 paint

0.95 plastic

0.93 water

an book i have about IR is having

0.85 Brick, common

0.94 Brick, refractory rough

0.95 electrical tape black plastic

0.92 Glass

0.93 lacquer, bakelite

0.94 paint average oil based

0.94 paper black

0.93 rubber

0.92 tar paper

(I have only listed the values there seems usagbol to this as the list is alot longer..)

So based on this i will say around 0.93-0.95 so a good point will be 0.94e but as other users here is having IR meters and they are very likly fixed at 0.95e so i think it is best to use 0.95 so it is easyer to compire the results there is posted.

Thomas

Seems like a good suggestion - as you say at least we can compare results with some confidence, I think the issue on resolution, spot size is also a complicating factor.  From what I understand you will get the average of the spot size so any real hot spots will be averaged away.

I've been trying to find a IR lens that will give 10mm x 15mm as the total field of view.  Interestingly it gets even more complicated when you look at effective touch temperature and we need a gizmo called a "thermesthesiometer probe" but these are not commercially available as far as I have found - if you find one let me know.

NASA (slightly old but the most recent I could find) have suggested that 45C is tolerable (indefinitely) for polished metal with an effective infinite heat capacity, and obviously a very high thermal conductivity, but this does not play well to our circumstances unless we measure less than that.

So I think there are three issue in this thread:

1) Touch temperature concern.

2) Die temperature and effect on performance.

3) Effect on functionality of USB? (Circumstances yet to be determined)**.

I have not seen any reports that making the suggested mod fixes anything, I'm sure you will let me know if I've missed it.

As always - comments welcome.

Cheers

Pete