@gbulmer, large 1.5-2X desktop lens with shadowless backlight makes soldering of 0603 extremely convenient. That's why 0603 is my main part size for quite long time. It helps a lot with 0402 too, but soldering them remains tiring. Perhaps if I'll find desktop lens with better magnification (3-4Х) for reasonable price I switch to 0402.
I want RET6 Mini or, better yet an F4 Mini.
(32 posts) (6 voices)-
gbulmer, siy, Rob, those boards in the openstm32hw git repo look really good (aesthetically and technically) for only having 2 layers.
If you care about analog performance at all, I recommend a well-bypassed 4-layer PCB with dedicated solid ground and power planes. On these stamp boards, with 4 layers, it's probably best to have simply one power plane for both the analog and digital supplies to separate the ground return paths maximally. The return paths can get nasty if you have to share a plane between two supplies and they provide current to places that make digital signals run over the paths for the analog return currents, then analog signals will be coupled with some noise from the digital transitions.
I've measured the noise myself on all the Maple boards (they were 2-layer up to Maple r3) and a bunch of development (the Olimex STM32 2-layer boards) and other boards, and every two layer board loses several more bits of ADC resolution to noise compared to the 4-layer boards of Maple r5 and Mini, which have a noise floor that's right around one LSB at 12-bits. But, I guess it's not very important to most people because there are very few microcontroller development boards that have ADCs that have ADC noise levels as low as 1mV with sampling rates of hundreds of kHz. Not only are there hardly any PCB designs that are capable of it, but for awhile, the STM32 was the only microcontroller capable of that performance. The PSoC's that you mentioned advertise a 20-bit ADC, but I don't know what the sample-rate or noise is. They say that there's a 16-bit 192ksps mode, which is probably comparable to the STM32, but I haven looked into them. I agree that those PSoC's are on a different price level (~2x) than the STM32 line, including the F4.
I like the idea of having an Ethernet jack on a STM32F4 board.
Also, do you all get the sense that it's time to switch to microUSB from miniUSB? It seems like the number of cables of each laying around has flipped from a year or two ago to microUSB being more common.
I also think that my happiness falls off sharply between 0603 and 0402. 0402 appears to be the first size small enough for their behavior to be controlled more by solder and static electric rather than inertial forces, so they cling to tweezers and can't be deterministically (for me, at least) flipped and rolled around. Shouldn't there be a way to have tweezers with a button that can change the charge of them or something to control sticking? I also like how some 0603 resistors have values printed on them.
Posted 3 years ago # -
Requirements (Continued)
Six, Hand board assembly (soldering to, way, old school types)
I can't see anyway around a requirement for hand assembly.
Thanks to gbulmer, siy we are homering in on what that means in terms of board design:
1) Certainly 0805 and probably 0603
2) All SMD (which is probably all of the parts) on one side (maybe we could stretch this a little and put a chip or two on the other side).
3) BGA. I've got some links here that say its possible to hand assemble a BGA. This solves any MCU pin count problems as the 176 pin BGA is smaller than a 64 pin sop.
I haven't really looked at these links so I'm not sure I believe it yet.
http://openhardware.net/Misc_Stuff/ToasterSMD/
http://articulationllc.home.comcast.net/~articulationllc/sm0402.htm
http://www.seattlerobotics.org/encoder/200006/oven_art.htm
http://www.ustr.net/smt/oven.htmMaybe the next project should be a home brew pick and place and solder paste machine.
Actually I have or could recreate all the code from a real, commercial, pick and place machine (including machine vision) which I developed for Universal in the early 90s.
Universal is now defunct.Four layer board
I've got a bunch of Nays and only 2 (counting me) Yays for a four layer board.
I think of a four layer board as a two layer board with ground and power plans in the inner layers. We don't need to make the number of layers a requirement (we can drive off that bridge when we get to it) but its starting to look (to me) like we won't have a choice anyway (will have to go 4 layer to make layout work).
Besides I have these great connections for getting the boards done.Seven, Pin functions
Personally I just need all the analog I can get and a decent shot GPIO (besides on-board high speed interfaces). But, maybe there are some functions that the world just can't live without like serial, full speed USB, CAN, etc. Also the Lipo connector on the Maple is a nice touch to keep in mind. So, for starters (not in any order):
1) At least two channels for each of the ADCs
2) The DACs
3) decent shot of GPIO, PWM
4) A respectable Arduino interface.
5) digital, analog power and groundI'm assuming typical, on board power management (a couple of LDOs)
Eight, I forgot what eight was for.
Nine, 'cause I lost count.
Ten, is for everything, everything...Posted 3 years ago # -
Ethernet Digging.
I popped up some options for on board Ethernet (100Mb):BGA MCU with MAC, 44 pin PHY, little transformer, mini connector.
For 64 pin MCU chips we need a PHY and a MAC
Chips with both come with SPI, SDIO, and High Speed USB (MCU host) interfaces suitable for STM32F415RG ish MCU. The STM32F415RG SPI only runs at 37.5 MHz so we'd only get 1/3 throughput. The SDIO looks promising but, I haven't found actual chips (need the chip set from a Cardbus NIC. USB NICs can go on or off the board but tie up our only High speed USB port. maybe we could make the OTG port look like a USB hub and so still have a least a high speed USB host along with the NIC.There might be an option for a brain damaged MAC that only does 100BaseTX full duplex (so it wouldn't talk to a Ethernet hub, only switches and other devices).
This would be in the form of another micro on board).Posted 3 years ago # -
gbulmer
Moderatorokie - I think we are exploring a different 'volume' of the requirements and design space from LeafLabs.
siy and I are interested in making it easy and cheap to prototype using DIY boards.
We've abandoned Arduino-compatibility, so we can reduce signal interaction compared to Maple trying to mimic Arduino headers.siy has been able to route a single sided PCB.
So following his approach, a two sided board could have a carefully designed ground plane, which might have the potential to reach good noise levels.
siy can get several 'turns'/day using DIY PCBs. Trying many prototypes may get there.A PCB designer, trained on high-end ECAD, has suggested to me that if I had access to state-of-the-art ECAD, it is likely good noise levels can be reached because that is what modern, high-end ECAD can help with.
I've also spoken to a guy who 'quietens' boards. He believes he can identify small changes to a board to reduce noise because that is what he does for a living at an electronics company.
I am no expert, but based on those conversation, I'm hopeful there may be several ways to get good results.
One of my interest is developing boards that inexperienced folks can build. Ideally, I would like to enable them to make a DIY PCB, then assemble the board. I'm also aiming for a layout I can embed into larger designs, so I constrain my routing options more than siy. Hence I am willing to compromise on noise to reach that part of my solution space.
Edit:
Also, do you all get the sense that it's time to switch to microUSB from miniUSB?
Some folks have USB mini cables, and they are cheap and easy to get.
Mini feel more robust a connection than micro, but that might be purely tactile and not real mechanically.
I have not found any through-hole USB micro sockets, but I have found through-hole USB mini sockets.
If you remember, several people pulled the USB socket off a Maple, so a through-hole socket seems to be a handy fall-back.I've had real problems finding a USB Micro OTG plug to USB A *socket* cable (plug USB Micro OTG plug into STM32F4-Discovery socket, and plug a USB flash stick into the A socket)
Posted 3 years ago # -
gbulmer
Moderatorsiy -
... large 1.5-2X desktop lens with shadowless backlight makes soldering of 0603 extremely convenient. ...
It isn't just my eyesight, it is the steadiness of my hand!-) No strong coffee in the morning for me :-(
Unfortunately, schools in my local area don't seem to have a classroom-full of good desktop lenses. I know we don't have enough at the hackspace.
Worse, teenage children can be a bit clumsy and uncoordinated, and older folks are less steady than they once were. So some will be able to do 0603, but their seems no purpose in designing boards which are too challenging. My current view is I'd rather make them bigger, and use easier to handle components, so that more folks can succeed.
Posted 3 years ago # -
gbulmer
ModeratorMediumKahuna
1) Certainly 0805 and probably 0603
In my experience it is easier to start with a 0805 design, substitute 0603, and compress it than vice versa. I'll stick with 0805 or bigger to make it easy to hand assemble.
2) All SMD (which is probably all of the parts) on one side (maybe we could stretch this a little and put a chip or two on the other side).
As soon as you put SMD parts on both sides, the simple 'toaster oven' or 'electric skillet' techniques become harder, which is what I want to avoid. My technique is to put through hole parts on 'the other side' to the SMD. This can free up space, and seems to work okay for switches, plugs & sockets, LEDs etc. Components on both sides will always be a bit more challenging.
3) BGA.
I am willing to believe folks can do it, but IMHO it is hard to debug and fix. So it isn't a good fit for SMD newbies.I'd love a home brew pick & place machine. If there were a viable machine, *starting* at sub $1000 (but maybe going up to quite a realistic price) I think quite a lot of colleges might get one to try.
Four Layer
I have no fundamental problem with using them for 'production', but I like to prototype with two sided because we can DIY them.
I quite like using Eagle (I never thought I'd ever say that). I either need to buy Eagle, get comfortable with DesignSparks, KiCAD or some other Open Source ECAD package before I'll do 4-layer.I also think that by abandoning Arduino header compatibility, signal interaction, and hence analogue noise, is reduced.
Further (I think) by changing a few signals, which will require new board files for the IDE, and new #defines for bootloaders, signal interaction can be reduced even more. So analogue noise might be better than folks expect.Think of it this way. By using a single layer signal layout (no signals crossing at all), and a carefully designed ground plane, you might get acceptable noise figures. Then by using a simple adapter to get to Arduino-header compatibility, you get the other two copper planes, but in a cheap, optional way :-)
Seven, Pin functions
Using my 56-pin layout, all signals for a 64pin part, except USB and main crystal oscillator, are available on a giant-DIL 1.2" pitch format using PCB design rules that we can DIY. I have done a quick (experimental) routing using commercial tolerances on 0.9" pitch. That gives 50 signal pins, 5V (USB), digital 3.3V & ground, analogue 3.3V & ground, and Vin.Rod and siy have shown that it can be shrunk even further.
So, if you want 2x28 pins, I am confident you could have a 64pin STM32F4 on 2.8" x 1.0" PCB.
Using Rod and siy's approaches, even smaller.
We all have on-board power regulators.
Siy has a beefy digital supply. I can't remember Rod's design. Inspired by siy, I might change my power regulator approach.Have a look at the data sheet, and see if that is enough for your non-Ethernet, non-external memory needs.
Roughly, some pin-constrained combination of:
16 ADC channels, with any channel to any of the three ADC's
2 DAC
15 or more PWM (IIRC it was 15 PWM on STM32F103 medium density)
3 SPI
3 I2C
4 USART
2 CAN
USB HS
JTAG/SWD
any pin can be digital I/OFor me, the Arduino-headers are a second adapter board. That is almost the other two layers of a 4 layer board.
Schools in the UK don't use LiPo's because they are quite delicate and dangerous (rarely explode, but do burn well, I'm told).
Micromouse and small robot makers don't seem to put LiPo chargers on board because the robot is either not connected to anything that could charge the batteries, or they use several sets of batteries to keep the robot continuously active, so they have 'proper' LiPo battery chargers. Also, many use 7.2V or even 11.1V LiPo setups, making it quite complex.
So I am not trying to get LiPo charging on anything.Posted 3 years ago # -
STM32F417IGH6 in a 176 pin BGA, $12.00
DP83848I Ethernet PHY 8.00
TG110-S050N2RL Isolation Module 2.00
Miniture connector (still learning)
They all fit on a 2" by .8" or even Maple Mini compartible boardBetter go, power has been out a while and I'm torturing my lead acids.
Posted 3 years ago # -
gbulmer
ModeratorMediumKahuna
3) BGA. I've got some links here that say its possible to hand assemble a BGA. This solves any MCU pin count problems as the 176 pin BGA is smaller than a 64 pin sop.
I haven't really looked at these links so I'm not sure I believe it yet.
http://openhardware.net/Misc_Stuff/ToasterSMD/
http://articulationllc.home.comcast.net/~articulationllc/sm0402.htm
http://www.seattlerobotics.org/encoder/200006/oven_art.htm
http://www.ustr.net/smt/oven.htmI looked at those four links, and the browser Search didn't find 'BGA' in any of them.
Did you not post one?Posted 3 years ago # -
siy
Member@okie, few early prototypes of Mini64 were using microUSB socket, but they I decided to return to miniUSB for very simple reason: everyday use of microUSB did show that it's very easy to break it from PCB. This never did happen for me with miniUSB despite much heavier and much less careful use.
Posted 3 years ago # -
siy
Member@gbulmer, I've never used BGA in my boards (although I'm going to try), but I think it is doable as long as PCB has solder mask. From the other hand, my experience with MLF packages (they are closest to BGA among ones which I ever used) suggests that there might be problems even for PCB's with solder masks. While smaller packages (<32 pins) are quite simple to solder, larger packages often caused problems for me. And while MLF can be easily desoldered and soldered again with hot air gun, BGA will require reballing after desoldering which is quite complicated procedure (although still doable at home). I guess that coffee is absolute taboo before doing it :)
Posted 3 years ago # -
siy
Member@gbulmer: quite impressive video http://www.youtube.com/watch?v=JB1InDsWCjQ
Posted 3 years ago # -
gbulmer
Moderatorsiy - Thanks for the video. Was it intended to persuade me that I'll never be able to do BGA ?-)
I should admit, the fact that a 176pin BGA is *smaller* than a 64pin LQFP is exciting.
If Chris has perfected making DIY solder masks, then it might be feasible to DIY BGA, but fixing a board is still beyond my skills. I'll show Chris the video, and see what he thinks. He is more skilled, and with a much, much better touch than me.Posted 3 years ago # -
siy
MemberThe only intent of posting this video is to show that working BGA without special equipment is quite complicated and not very well suited for DIY-ers. At least for most of them. Of course, times are changing and while few years ago most DIY designs were based on TH components, today most of them using SMD components. Same may happen with BGA.
P.S. The main obstacle for me in regard to BGA is need to have plated through holes. Most BGA's interesting for me can't be brought out in single layer even with 0.1mm spaces/traces (and so far 0.1mm is still above my process capabilities, although I'm very close to these limits).
P.P.S. If you need any help with solder mask - don't hesitate to contact me, I'm ready to share any knowledge/experience I have obtained while working with them.
Posted 3 years ago #
Reply »
You must log in to post.