External power: one or both ground pins?

Status
Not open for further replies.
Epyon

Epyon

Well-known member
A quick question that I always wondered about (and that only Paul will be able to answer presumably). When making a PCB that will accommodate a T3.X and provide external 5V input to VIN, is it best to ground both pins or does only one suffice? I'm asking this because on my current design the 'bottom' GND pin (next to Program) is easily connected to my PSUs ground, but the topleft GND pin (next to RX1) will need a trace that will add a bit more length (and thus impedance) to it. It's best practice to ground devices on only one point (star topology), but I have no idea how the ground plane on the T3.X is poured.

I don't plan to drive anything high power with it, but it may get a Wiz820io stacked on top of it.
 
A wizNet 820 can pull a bit of juice though. I resort to comical levels of routing in order to achieve a unbroken ground plane on one side.

I believe Paul has mentioned the use of a 4 layer board in his design. I presume that one inner layer would be dedicated to GND (and perhaps AGND) while the other inner layer is for 3.3V. Given how wide the GND layer potentially is, I doubt that it cannot handle the juice of a few hundred mA running through it safely.

Signal integrity may be a different matter, however. That is, you may have high-speed pulses associated with the power supply impacting analog readings, even if AGND is separated with the use of a ferrite.

Bottom line, only testing will tell. If it was me, I'd add the trace.
 
Grounding both pins is best. But for most applications just 1 is ok. Either will do.

Indeed the entire layer 2 inside the PCB is dedicated to ground. But the same is not true for power. Layers 1, 3 & 4 are for signals and power routing. For 3.3V power, if you're going to draw 250 mA (for example, an ESP8266 module) the 3.3V pin between AGND and pin 23 is best, since it's closer to the voltage regulator chip.

On the upcoming Teensy 3.5 & 3.6 boards, layer 2 is dedicated to GND and layer 5 is dedicated to 3.3V.
 
Thank you for answering. I had the Wiz820io drawing power from this pin. I have the footprint for a MCP1825S on my board, but it saves me three parts using the Teensy LDO.

@Constantin: I just have some low frequency digital pulses. For analog I use an off-chip ADS1115 ADC. In previous designs I saw no big impact on using one or both ground pins, but I was wondering about the 'best practice' when designing this board :) .
 
PaulStoffregen said:
On the upcoming Teensy 3.5 & 3.6 boards, layer 2 is dedicated to GND and layer 5 is dedicated to 3.3V.
Click to expand...

6 layers! Pure LUXURY! ... when I was a kid, we were lucky to have 1/2 of a layer! The other side was usually coated with shoe polish that we made semi-conductive with coal dust that we ground in a mortar and pestle! ;)

... but seriously, I can only imagine how hard you had to work to make those BGAs work as expected! Any buried Vias in there or was it a through-hole design?

Epyon said:
For analog I use an off-chip ADS1115 ADC. In previous designs I saw no big impact on using one or both ground pins, but I was wondering about the 'best practice' when designing this board :) .
Click to expand...

Off-chip is frequently the way to go for ADCs. Better control of the ADC performance, custom-tailored solutions, and so on. That's not to say that the Teensy doesn't have a capable ADC - it does! ...and Pauls work proves that you can design a very capable system that works very well in a generalist sense and which offered in excess of four times the resolution/accuracy of the AVRs it is replacing. For many applications, it is perfect. But for specific applications like measuring AC voltages (requiring a complicated front end that always presents the unipolar ADC inside the Teensy with a positive voltage signal, etc.) then resorting to an external ADC made for that specific application may be a better bet.

Similarly, if you need something above 12 ENOB in single-ended or 13 ENOB using differential inputs, an external ADC will be your better choice, no matter how much the Freescale marketing speak / datasheet for the MK20 attempts to imply otherwise. When it comes to claims like the 16 ENOB for A10/A11 in differential mode, I'd really like the chip makers to have to demonstrate how they got that performance.... i.e. publish what they did to achieve it, from the hardware front end required to the software configuration... I'm not convinced that Freescale actually consistently achieved 16ENOB without resorting to heroic measures!
 
Last edited:
Status
Not open for further replies.
Back
Top