KiCad prototype to Production Experience on 100W power supply

[eepete]

I’ve done around 4 complex boards and 8 or so “Let me test this chip” or “connector break-out” boards in KiCad. I recently (summer of 2020) made a power supply to run a server in a mobile/vehicle environment. It has two 50W boost-buck supplies so that cranking the car doesn’t affect the output voltage. An on-board micro allows configuration of output voltages, monitoring of current and temperature sensors and also does all the mimicking of the front panel button to bring the server up and take it down.

The board is a 4-layer, use OSHpark to make it and also did a stencil for assembly. I have a cheap (too cheap, I need to get a better one) hot air reflow that I got used to on the 4 complex boards. The power supply board has a lot of 0402 and some parts with pins on the bottom of the package. The main power supply chip is the TI TPS55288.
The proto was easy to assemble. I made the placement so that I could cut the stencil into pieces to populate the micro processor, the power supply and control circuitry, and each of the 50W supplies separately so I could bring the board up in stages. This was a huge win, since there wasn’t the “Turn everything on at once and hope” moment.
I over-cooked on of the main power supplies chip, damaging a FET. I was able to test it in buck mode, which let me know the wiring was OK but in boost mode the main external FETS just got nice and hot. Fortuatnly, the other supply worked 100% so I could test it.
It was much cheaper to be able to proto the PCB my self than to have a stuff house make 2 or 3 boards. The extra effort to allow for incremental power-up was not too bad and really did not waist any board space or make the board larger. The board had to fit in the extra space in a chassis designed for an ATX-FLEX when there was only an ATX-MINI motherboard present. The chassis was a SuperMicro SCE300 and the motherboard is a SuperMicro A2SDI-8C-HLN4F.

I got 25 boards back from the assembly house (“stuff house”) and so far all 3 i’ve tested worked. I made some BOM errors, and need to write some software to cross check a .csv version of the BOM with the Pick and Place to find errors in counts and parts omitted fro the BOM. Somehow, I dropped 5 parts. All but 1 were 0603 or 0805, the one was an 0402. Never blew a BOM in 40 years, just must be getting old. Software will let me make the minor BOM changes from a proto without having to start from scratch and should help the transition from prototype to fab. I should also be able to re-format both the BOM & P&P to whatever format the stuff house wants.
I had the house work with the PCB group of their choice, and just pointed to the OSH park stack-up for the 4-layer board. I also left the panalization up to them. A little extra $$s but the stuff house can get things in a form that works best for them. The only change I made from the OSHpark stack-up was the top and bottom layers went to 2 oz copper due to the high currents. The production boards tested better than the proto due to this change. I also went from just using 10 mill hole vias in the high power traces to also using 38 mil vias along the path. This resulted in lower via resistance and inductance. Never thought to do this until now. When both supplies are running 12V at 4 Amps, and the input goes down to 6 volts, the DC power draw off the battery is 20 amps. Normally, with Unix running and the vehicle idling, not cranking, the maximum possible from the battery current current draw is around 7 amps. The idle current from the server PCB (and the 4 fans in the chassis) when Unix is running is about 1.7 amps at +12, about 1.5 amps for that supply from the vehicle battery. A “fun” project with high currents, uP control, lots of EMI to worry about, and tight integration with the server motherboard and vehicle battery/ignition/tempurature.

Just wanted to share this “proto to production” story with everyone. Enclosed is gratuitous pix of the PCB.

Pete

[skajam66]

Interesting. Can I ask how you cut the stencil and what margin did you leave between parts? Cutting a stencil is always something I’ve shied away from but I can see how it can really help bring a proto board up.

ac

[paulvdh]

The way I read it, the stencil was also ordered from OSHpark.

If you’re interested in cutting your own stencils, according to several youtube videos some vinyl cutter apparently works reasonably well down to 0604 size. I particularly like the video from MikesElectricStuff, but there are others too.

[eepete]

Indeed I ordered the stencil from OSHpark, I get the 3mil thickness polyimide, . Then all it takes is a good pair of scissors to cut it. You only need about .1 inches of gap to make it work for the small packages (1206 and smaller) near the edge, a bit more if you have a big chip.
It gets rough if you have a small board size constraint as there is a bit of waste in space. But that’s just a new twist on an old trade-off: As board density goes up, the time to place and route goes up exponentially, not linearly. Some “hack” boards (size not a factor, just needs something that works) look like golf courses when they are done, but the place and route in no time at all. As such, organizing for an incremental population and testing is easier.
I’m currently working on a 3 board set. 2 of the 3 are pretty easy, for the fixed board size (Raspberry Pi sized PCB) they don’t have too much on them. But the 3rd board is packed and mostly 0402. The best I can do is make one side the power supplies, have a .1" zig-zag gap, and the other side is the circuitry. At least I can verify the power supplies before building everything, and hopefully I can get enough of the “0402 land” circuitry working to verify things are wired right. This beats the daylights out of contracting out for 5 boards at a $4K cost in both time and money.
I do need to get a hot air station with better temperature control, and a stereo microscope-ish device. I’m luck to be near sighted, but, that has it’s limits even with a magnifying glass.
One other “trick” I use is to cut a small PCB to test the foot prints of new chips I’m using. They are typically < $10, so even with stencil and shipping for < $50 you can be sure all your footprints are good both in pad size and solder paste size. Part of the “logistics of the design process” is to do this early on so the time to get the boards and stencil is not an issue.
Many trades offs here. Is this a production board, so size cost money or is this a “make a few that work”. Are you verifying a design with the intent to re-lay out later (and perhaps use the golf course layout so that software development can go in parallel with the hardware re-layout) ? Are you aiming for the middle of these two approaches? For this power supply, I had a fixed area and I used it all, and with the wide traces for 20 amp currents, it made it pretty easy to break things up. A rectangular board is easier to do this than a square board.

Long answer, hopefully it puts context to a lot of the factors in play and helps.

Pete

[Russ]

Where did you get your hot air reflow station ?

Do you plan to check your board’s output voltage and current when connected to a load AND while starting the engine ?

[eepete]

Hot air was from Amazon, X-Tronic Model #5040-STS-XR3. It was $300, my thinking ws try something so I can understand what’s really needed. Regrettably, my thinking was correct. The issue with this one was the air temperature goes up and down by 20 degrees C. So if you set it for 220C, the peak temp for some reflow profiles, you’ll get a range of 200 to 240° C. I spent too long trying to lift the main power supply chip, and on of it’s FETs was damaged. It would buck, but when I tried boot things shorted out.
I’ve since used it with a set temperature of 105° C and done OK, haven’t popped anything yet. The pre-heat is important with the 4 layer board where the 2nd and 4th layer are ground. If you don’t pre-heat, and you’re soldering a chip with a pad or pads on the bottom, you’ll be there to long and by the time the chip “twitches” so you know it’s on liquid solder, you’ve overheated the cihp. I need to investigate other options, probably in the $1K range and definitely not on amazon. Time to get a stereo device to inspect too. Nothing gives you focus on what you want more than experience where you learn the hard way.

The first step in this project was to take the MSO scope and put it on vehicle batteries. I lost the images (ARGhhh…) so I need to re-do this. On large fire engines and tankers, where you have 2 big truck batteries, the minimum voltage on cranking was about 9 volts. On an F350, 8.5 volts. On an F-150, 7.5 volts. The worst is a small passenger car, where it dipped to 7 volts somtimes. Most of the time, the dip lasted for about 150 to 200 mS until the starter motor developed some back EMF. Then things went up by about a volt. Gotta get new pix, that was valuable information. The voltage traces showed a clear pattern that provides much insight.

On the power supply, I took current readings for one supply where the output was 12V, load was 4A. I went from 6 to 18 volts and measured the current. Bottom line, is this thing can run forever on 7.5 volts with both supplies at 4A (so total power out is about 100W). The battery current will be about 14.5 amps. Regulation was maintained down to 6 volts, where a single supply drew 10 amps. In software on the micro, I enforce a “dump” at 6.3 since that’s the worst case input voltage where the on board regulator for the power supply chip is happy. It was critical that the server can stay up when the vehicle is being cranked. Efficiency at 12V battery in and 12V, 4A out is about 96%.

Enclosed is the .pdf of the manual I did for this board, it has a graph in it showing input battery votls to current for one of the supplies.

Conducted EMI on the battery input looks good too. Good enough to “risk” taking it to get certified once it’s all connected up to the server. The big honking 680uH 15A common mode choke does wonders in the 0 to 50 MHz range. The measurement was taken by simply connecting the battery input directly into the spectrum analyzer.

The TPS55288 really needs a micro to work, but for me that’s a feature not a bug. A lot of configuration can be soft. The uP also deals with booting and taking down the server, and logs events so you can get some (but not all, I’ve only got 32K of MRAM for the program) of the extremes and faults seen. You can see the functionality available from the diagnostic serial port in the manual.

Here’s the supply in the chassis with the speaker and serial port to motherboard connected, the yellow and black wires will go to the two outputs. Lots of fans, but, this has to work at ambient temps in the 120° F range.

Hope this and the manual helps answer your question.

Pete

serverSupplyManual.pdf (832 KB)