ADR452x based reference board - What do I need to improve?

[TheUnnamedNewbie]

So, for a project in my uni I built this reference board based off of the Analog Devices ADR4520 (or any other version of this chip would work really, provided you increase the supply as well). Since the thing quotes 0.02% initial accuracy, I figured it could be a good start for me into the world of volt-nuttery and wanted to do a good job, even though it's first application in the uni project was a 12bit ADC. Ofcourse, it's not anywhere near the 6.5/7.5/8.5 digit level of accuracy and drift, but it's a start for me, allowing me to at least verify the performance of some of my cheap-oh meters.

I woud like to know what needs improvement to make this an even better supply. One thing that comes to mind is that perhaps, using a buffer on the output would improve the performance. I don't know how essential things like cutouts between the pins for preventing stress due to thermal stuff is needed. I added a good amount of capacitance on the output but don't know if there is any reason why I would want to not do this? (note that there is a error in the schematic and PCB: the REG113 enable pin should be shorted to the VIN pin. I did it with solder on the final PCBs, but never got to update the actual schematic and layout).

PCB has a full ground plane on the bottom.

Schematic:



PCB:

[Kleinstein]

Cut outs help to prevent mechanical stress from the board (e.g. mounting) to influence the ref chip. So it might be a good idea to have some kind of cuts at 3 sides of the reference chip. Usually no need to go between the pins. For a single hand soldered piece one could even bend the pins at the SO8 case to get less board stress to the reference. However this can be a 2 sided idea too. At the layout shown the mounting hole in the lower left corner might be a problem. Such a small board should be more like 1 or 2 mounting points. 4 Holes is definitely to much.

I would consider the two 47 µF caps at the output more than needed. One electrolytic of 10-47 µF might be a good idea - it's ESR (maybe an extra 5 Ohms in series) might help damping the peak in the output impedance. This does not matter if an extra buffer is used.

If more current is drawn from the reference (e.g. more than about 1 mA) an external buffer might be a good idea. The load regulation is not that great and there can be an additional thermal effect due to heating the reference chip. However the buffer will add some offset, so access to the non-buffered reference should also be there.

The voltage regulator might be a little close the the reference. More distance could reduce heating effects. Though this would be more of a problem at high currents only. Heating on the board is also a source of mechanical stress. The LED should be OK with much lower current - a modern one should not need more than 1 mA. Some heating on the board can also be an advantage, as there would usually be a slow increase in temperature during turn on. A slightly higher than environment temperature also helps to reduce humidity effects.

The worst part of the layout is using the GND plane for the GND connections. For a reference one should use point to point connections with the GND pin of the reference as a kind of star ground. So at least the ouput GND terminal should be a line to the reference pin, without any other connection except the output capacitors from there.

[TheUnnamedNewbie]

The worst part of the layout is using the GND plane for the GND connections. For a reference one should use point to point connections with the GND pin of the reference as a kind of star ground. So at least the ouput GND terminal should be a line to the reference pin, without any other connection except the output capacitors from there.

I feel like such an idiot now that you point it out! Of course! I'll try and fix that. I'll also take a look at the other points.

I put 4 holes in for standoffs - the boards were part of a prototype setup and had to rest on little pcb standoff feet on the desk. They aren't screwed down.