TonyMac32 Posted April 3, 2018 Posted April 3, 2018 I've done some load testing of my first prototype power "mezzanine" board (can't say "hat") The voltage drop of powering the RPi touchscreen off of the GPIO while the tinker is on micro-USB is insane, I was reading 3.9 Volts on the USB ports when I ran minerd --benchmark. I may have finally worn out the Micro-USB connector... This is a project for learning the ins and outs of kicad and get an idea of the capabilities of the PCB houses I have reasonably fast access to (Chinese ones unfortunately always get stuck in customs for a week or more on top of the normal shipping and fab times, so the cost savings is lost in time) I have some more interesting things in mind, but I thought I'd start out (very) small. The input is protected by the recommended ideal diode circuit from the RPi guys, stress testing didn't even make it warm. If a better solution is needed, it can obviously be designed, but this is extremely cheap all things considered and that was part of the G&O for this project. By comparison, using this GPIO board to power everything resulted in a minimum 4.8 volts while also powering a USB 2.0 spinning disk HDD. (lower than I'd hoped, I'm going to make some adjustments to see if I can squeeze a bit more out once I get the redesigned version 2 back, I forgot about the temperature rise factor... ) One of the issues goes back to our friends at RPi, the 5V pins are on the outside edge, so you have to go around the inner row to get to them. This complicates moving big conductors around somewhat... The chip in the back is a SPI NOR for booting the Tinker (experimental, I threw it in there as an experiment and never actually performed the experiment...) I'll probably wipe it from the board, perhaps leave an 8-pin SMT-adapter-sized pattern to the SPI for the adventurous. Power supply used here is the Odroid XU4 supply. This initial run has no overvoltage or overcurrent protection. And I need finer tipped soldering iron... I'll GitHub this one after some cleaning and citation of sources, normal disclaimers for those who want to build one. I will have moved one of the USB ports though since @chwe happily pointed out that Y-cables don't reach that far to get from the real ports to the power-only ones... (oops) Boards successfully powered by this configuration (I don't / can't own one of everything... ): Tinker Board Le Potato NanoPi K2 (Already has a barrel jack, by the way, but you never know, maybe you want the bigger one) 2
chwe Posted April 3, 2018 Posted April 3, 2018 6 hours ago, TonyMac32 said: By comparison, using this GPIO board to power everything resulted in a minimum 4.8 volts while also powering a USB 2.0 spinning disk HDD. (lower than I'd hoped, did you use a reliable pinheader? just cut one in half to see how good they are.. A real 'shithole' to put a pin inside... Did you check the voltage drop once with a benchtop PSU? by the way, you constantly refuse my INA219 circuit on the board for a more advanced power-consumption monitoring on the board.. no space for another 8pin smt, RN and a shunt wired to i2c and GND VCC of the barrel jack?
arox Posted April 3, 2018 Posted April 3, 2018 Are you powering the board through Dupont connectors ?
chrisf Posted April 3, 2018 Posted April 3, 2018 10 hours ago, chwe said: by the way, you constantly refuse my INA219 circuit on the board for a more advanced power-consumption monitoring on the board.. no space for another 8pin smt, RN and a shunt wired to i2c and GND VCC of the barrel jack? At 2amps a 0.1R shunt will drop 0.2V and dissipate 400mW A 10mR shunt would be more appropriate, dropping 20mV and 40mW
TonyMac32 Posted April 4, 2018 Author Posted April 4, 2018 17 hours ago, chwe said: Did you check the voltage drop once with a benchtop PSU I monitored at the barrel jack, at the output of the FET, at the GPIO pins, and at the USB out on the board in the various load conditions. 8 hours ago, arox said: Are you powering the board through Dupont connectors ? No, the Dupont connectors go to the RPi touchscreen. 17 hours ago, chwe said: did you use a reliable pinheader? just cut one in half to see how good they are.. Yes, I was able to verify the drop was on the trace. The terrifying reality is, even with a crappy header, you have about 4x the contact area as a microUSB. Now, were this the 1950's I'd just say to mercury wet the pins for luck. It's a shame all the best functioning things are also toxic... As far as the spartan nature of the board, the goal was to be simple and effective. What always happens (at least to me), is if I don't draw a line in the sand and call it done, then it will never be done. An example, I could use a hall-based sensor to have a much better SNR and an effectively non-existent loss. So, as far as features go, this is the first version, like I said I've moved a few things around and am "beautifying" it, labels/messages/warnings in the silkscreen, etc.
Tido Posted April 4, 2018 Posted April 4, 2018 7 hours ago, TonyMac32 said: warnings Like this: Objects in mirror are closer than they appear (not written in Europe by the way)
TonyMac32 Posted April 4, 2018 Author Posted April 4, 2018 I was thinking "Maximum input 5.3 VDC" and a center positive image, and probably "Do not exceed 4.5 Amps"
TonyMac32 Posted April 4, 2018 Author Posted April 4, 2018 1 hour ago, Tido said: no warranty use on your own risk ...maybe with a stick man on fire?
JMCC Posted April 4, 2018 Posted April 4, 2018 What do you think about making a hole in the center for a fan (and four small ones for the screws? I think that would be a great plus, specially for the Tinker that shows a significant performance improvement when active cooled. [EDIT: Maybe not a full hole, but more of a grilled circle]
JMCC Posted April 4, 2018 Posted April 4, 2018 And also, beware of the Tinker S, it has that under-current protection and may not turn on when powered only through GPIO. Maybe a solution in that case could be to plug a short cable from one of the power USB's to the board's microUSB in.
TonyMac32 Posted April 4, 2018 Author Posted April 4, 2018 1 hour ago, JMCC said: may not turn on when powered only through GPIO. My understanding is that this has been rectified, I found that to be true on my early sample and reached out, the concern obvious. So, production boards should be fine. 1 hour ago, JMCC said: [EDIT: Maybe not a full hole, but more of a grilled circle] I thought of this, I could put that in here. As far as a grilled circle goes, well, the slats will make noise depending on the fan orientation and proximity, but perhaps that's a secondary concern in this case. Probably a slotted rectangle to account for the maximum number of cpu positions, and increase airflow in any event.
JMCC Posted April 5, 2018 Posted April 5, 2018 Okay, I knew your board was missing something, but I couldn't point what it was. I think I finally got it: That makes it perfect! Don't you think so? 3
TonyMac32 Posted April 5, 2018 Author Posted April 5, 2018 1 hour ago, JMCC said: That makes it perfect! Don't you think so?
chwe Posted April 6, 2018 Posted April 6, 2018 problem is.. chuck can handle everything... It must be DAU approved ('dümmster anzunehmender User' - dumbest possible user) to be a good board..
TonyMac32 Posted April 6, 2018 Author Posted April 6, 2018 Just because you hooked 12 V into Chuck and your credit card computer didn't survive doesn't make it Chuck's fault. Chuck is a member of the Church of Darwin.
TonyMac32 Posted April 6, 2018 Author Posted April 6, 2018 Given the simplicity of it, a ton of holes aren't really a problem. This rearrangement of the "Lite" version puts a USB power where needed in case you have any "Y" adapter situations, leaves one in the original position to power either the touchscreen, provide power to the sbc the traditional way (probably not the greatest idea), or something completely unrelated (I powered a Tinker Board off of the USB on the power board while powering "Le Potato"). Thicker traces for power and an improved area around the header pins to reduce the drop there as well. @JMCC's slots for cooling, we'll see if the fab gets mad about that. ;-) 1
chwe Posted April 6, 2018 Posted April 6, 2018 1 hour ago, TonyMac32 said: Chuck is a member of the Church of Darwin. Chuck says to the PSU that it should step down to 5.2V 2 minutes ago, TonyMac32 said: @JMCC's slots for cooling, we'll see if the fab gets mad about that. ;-) make 2.54mm holes and they can't judge you.. people can drill out the hole they need.. 4 minutes ago, TonyMac32 said: Thicker traces for power and an improved area around the header pins to reduce the drop there as well Do you feed all possible VCC pins? symbol and written (center positive)... Does it come from the american 'don't dry your dog in the microwave' warning' (an evergreen joke when we were young... )
TonyMac32 Posted April 6, 2018 Author Posted April 6, 2018 48 minutes ago, chwe said: Does it come from the american 'don't dry your dog in the microwave' warning' Why yes, yes it does. And words on silkscreen are free. ;-) TBH, unless the reverse voltage were high enough to cause breakdown in the FET it doesn't matter anyway, the wrong polarity simply won't work. 49 minutes ago, chwe said: Chuck says to the PSU that it should step down to 5.2V there will be some losses across the FET and various contacts, so there should be no trouble, I doubt you'll ever see more than 5.25 at the PMIC/LDO/etc. 54 minutes ago, chwe said: Do you feed all possible VCC pins? Yes, there was some playing with pad shapes to get through there "safely", but both 5V pins are solidly attached. I improved that area a bit further, but honestly that isn't where the drop was occurring anyway. Would you like me to prepare a boundary diagram and DFMEA?
JMCC Posted April 6, 2018 Posted April 6, 2018 Also, the slot for the camera ribbon-cable is a good idea
TonyMac32 Posted April 6, 2018 Author Posted April 6, 2018 Yeah, it's there on the current ones I have on hand, it's just hidden by the barrel jack. I started with https://github.com/xesscorp/RPi_Hat_Template to get the board shape, Now that I (think) I've learned how to import layers into Kicad (like board outlines) I'll be making my own, it's easier to draw in an external application if you want precise dimensions/etc.
Tido Posted April 7, 2018 Posted April 7, 2018 16 hours ago, TonyMac32 said: a ton of holes aren't really a problem actually, yes they are. Everything in the airflow will slow it down. When you buy DELL, HP or Lenovo Server they have this lovely cover on the Rack-Server. Studies have shown that this kind of advertisment does slow down airflow by 2-digits %. See pictures of FB Datacenter no shield in front of the server - hardware from the Open Compute Project.
TonyMac32 Posted April 7, 2018 Author Posted April 7, 2018 9 minutes ago, Tido said: actually, yes they are. From a fab perspective, as in the context of the statement, they aren't. There isn't much on the board, so adding holes is not a technical concern. Now, to discuss fluid dynamics in conversational terms, a ton of small holes are most certainly inferior to a few properly sized ones for the rate of flow desired. For a given volume of air to travel through a reduced cross section it must gain velocity. Without a certain amount of pressure that cannot happen, so it simply maintains it's velocity and thereby less air flows.
chwe Posted April 7, 2018 Posted April 7, 2018 Fluid dynamics is like diffusion... A lot of calculations with results which are never correct.. Like Fick's law... 2.54mm pitch holes would be useful for a mini breadboard for prototyping... It's obvious that the user has to cut out bigger holes when he needs it as for better ventilation.. or a solid Cu surface and building a 'sandwich' with the Cu-foams from the RPi thread and spread the heat through the whole mezzanine board...
Tido Posted April 7, 2018 Posted April 7, 2018 inspired by @TonyMac32 I have sliced an old CPU heat sink - attached with 'kafuter k-5204k' by the way 80g is huge as you need just a tiny_little_bit to attach a heat sink. I have to admit that watching the movie @tkaiser posted before, I had to overcome a temptation of doing that as well. But maybe the surface cannot cope with aluminum (insulation)? I haven't done any testing so far, but idle 6-10° cooler than without any heat sink So Tony's "mezzanine" (also known as hat) will never fit
TonyMac32 Posted April 7, 2018 Author Posted April 7, 2018 Of course it will, you just need some clippers to cut out the slots. ;-). I guess it depends on how close to the header that is...
TonyMac32 Posted May 11, 2018 Author Posted May 11, 2018 Since I'm playing more and more with hardware on my own time, I'm going to start the process of uploading my work to github for general consumption. At the moment I only have DXF board outlines for a few things, I'll follow those with proper KiCAD templates so no one has to wonder where things go if they decide to build. https://github.com/Tonymac32/Board-Geometries 1
TonyMac32 Posted May 15, 2018 Author Posted May 15, 2018 Some feedback from @chwe and some improvements on the protection circuitry: And for the record, electrical cleaner can/will dissolve your electrical connectors... Also, lead-free solder is an angry and terrible thing. ;-) I'm charging my Pixel off of one of the power ports while I play music over bluetooth on the Tinker... 4
sfx2000 Posted January 16, 2019 Posted January 16, 2019 On 5/14/2018 at 8:09 PM, TonyMac32 said: Some feedback from @chwe and some improvements on the protection circuitry: Nice work! Where did you get your PCB's made - from the purple mask, I'm reminded of OSHPark... https://oshpark.com/ I've used them in the past - nice for very small batches during proto...
TonyMac32 Posted January 16, 2019 Author Posted January 16, 2019 Yes, Oshpark. I've not been pleased with the circuit itself, it's a bit more sensitive than I like with the FET's and reliance on passives. I have a new one out at fab now using a buck converter, will allow up to 24 V in and output up to 3 amps. I'll be setting it loose in the wild via GitHub and an openhardware.io project in hopes people want to buy me a beer. ;-) that platform uses pcbway and seeed, so you wind up with 10 bare PCB's, unless you provide an adequate BOM for them to build kits. All are good options quality wise, oshpark is a bit spendy if you are doing iterations, but it's also not in China so for me it saves any customs delays (waited on some pcbway boards for 4 weeks after the proposed delivery date one time)Sent from my Pixel using Tapatalk
Recommended Posts