Reason for 10 ohm resistor between AVCC and DVCC

Thank you for your reply.  That makes sense.

Into the already given and correct explanation for the presence of this 10 Ohms resistor, I might add some comments.

First, if you don't need to draw significant analog current (such as for a D/A), you can rise the resistor to 100Ohms. It gives a better low-pass effect and a cleaner AVcc.

In addition, the capacitors between AVcc and AVss should be a 10µF Tantalum/100nF ceramic combo, which combines low ESR with high capacitance and still low leakage current.
Also, to get best results, AVss should be routed separately from DVss to the regulator GND point. Since any current on a trace causes a voltage drop, digital GND has ripple voltage on it too. By routing the analog (reference) GND separately from teh digital GND, this is removed too and also improves analog performance. If you need a GND reference anywhere else for your analog circuitry (reference only, like OpAmp GND input reference, but not for analog high-power GND), use this AVss instead of DVss there too.

Jens-Michael Gross said:

First, if you don't need to draw significant analog current (such as for a D/A), you can rise the resistor to 100Ohms. It gives a better low-pass effect and a cleaner AVcc.

In addition, the capacitors between AVcc and AVss should be a 10µF Tantalum/100nF ceramic combo, which combines low ESR with high capacitance and still low leakage current.

I ran a simulation in LTSpice on the proposed RC filter where R = 100 Ohms and C = 10.1 uF. The image below shows that the voltage difference between DVCC (blue) and AVCC (green) exceeds 0.3V. The red is a plot of the difference between DVCC and AVCC. 

It has been recommended in previous posts and throughout app notes not to use different sources for AVCC and DVCC because they may have different start up times that could lead to a voltage difference greater than 0.3V that could cause latch up conditions and dissipate large amounts of current through the internal protection diodes.

It appears that the proposed RC filter may cause the same issues due to the voltage lag of the RC. Has anyone noticed any problems with an RC filter this big (R=100 and C = 10.1uF)?   

Zack Bomsta said:

they may have different start up times that could lead to a voltage difference greater than 0.3V

Yes. To avoid this you shall use low drop diode between DVCC and AVCC. Of course, the best idea is to measure AVCC ripple and analog parts performance of prototype to see - you need better AVCC filtering or not.

Troy Johnson said:

Can AVCC and DVCC just be connected together?  Especially if I have no intentions of using the analog functionality? 

In your case I would just connect them together.

Peter 

Zack Bomsta said:

I don't think that an external diode is the answer.

This is common and suggested by manufacturer, solution. - In case of separate digital/analog supplies or slow AVCC ramp-up times due to filter as in this case.

Zack Bomsta said:

The MCU already has internal diodes that do what an external diode would do

They don't. Especially if we consider internal protection diode max allowed current just 2mA. Even at seemingly "safe" 1mA internal diode will have higher voltage drop than external 100mA or 200mA diode.

Zack Bomsta said:

... that could lead to a voltage difference greater than 0.3V that could cause latch up conditions ...

Kudos on doing a good job of getting the specs and doing some solid engineering work making sure that your final design is correct. It is refreshing to see this compared to a lot of other questions here.

It is all about making sound tradeoffs and making sure you meet device specifications.

If you know that your board will never need to use the analog function, then the recommendation to just connect them is fine. If there is a slight chance you might want to use it in the future, then pick an appropriate RC value to give you some noise immunity.

Myself, for the cost of a resistor and capacitor, I would go that route, because I know how requirements change.

Ilmars said:

This is common and suggested by manufacturer, solution. - In case of separate digital/analog supplies or slow AVCC ramp-up times due to filter as in this case.

Interesting, in 5 years of MSP430 development, I have never seen this suggestion. Could you share the manufacturer's source that suggests this?

Zack Bomsta said:

Interesting, in 5 years of MSP430 development, I have never seen this suggestion. Could you share the manufacturer's source that suggests this?

I don't know that I've seen it from TI directly related to MSP430, but parts with multiple supply rails typically have sequencing requirements. I've worked with many processors, from DaVinci, to custom ARMs, FPGAs such as Xilinx and Altera, that have these exact requirments.

This is why there exists power supply sequencing control chips (like those from Lattice) and switch-mode controllers have enables and/or sequencing input/output control signals.

Brian, thanks for the reply. 

Brian Boorman said:

If you know that your board will never need to use the analog function, then the recommendation to just connect them is fine. 

Just to clarify, I am not the same user that said he would not be using the analog peripherals of the MCU (maybe I should have started my own post to avoid confusion, but this one seemed to touch on the topic well). My application relies on the analog peripherals very heavily. That is why I am trying to design an aggressive decoupling network between AVCC and DVCC of the MCU. I am afraid that if I get too aggressive with my design (one similar to the one that Jens-Michael suggests), I may end up with the same problems that separate analog and digital supplies would cause. In addition to the decoupling network, I have a dedicated voltage reference with high PSRR connected to the VeREF+ and am paying close attention to decouple the analog and digital grounds. 

Zack Bomsta said:

Could you share the manufacturer's source that suggests this?

Here you go: "Sequencing Power Supplies in Multiple Voltage Rail Environments"

Excerpt:

llmars,

This really only applies to digital-only rails (IMHO).

The topic of this thread relates to isolating noise between digital and analog supply rails. A diode of that nature would short the digital noise right to the analog AVcc pin.

Ilmars said:

Here you go: "Sequencing Power Supplies in Multiple Voltage Rail Environments", www.ti.com/lit/ml/slup228/slup228.pdf

BTW, your link has three extraneous characters at the end, causing it to not work. Here is a corrected hyperlink: Sequencing Power Supplies in Multiple Voltage Rail Environments

It looks like a good document to read for those new to supply sequencing.

Brian Boorman said:

A diode of that nature would short the digital noise right to the analog AVcc pin.

Not at all. We are not talking about ideal diodes here but schottky. So if voltage rail difference due to ripple does not exceed diode forward voltage - it will not short anything and will not conduct digital noise to analog supply rail. If you are going to have ripple which exceeds schottky forward voltage then most probably you will exceed max allowed DVCC/AVCC difference either. The key here is: DVCC/AVCC allowed difference is more or less close to diode forward drop voltages.

Ilmars said:

So if voltage rail difference due to ripple does not exceed diode forward voltage - it will not short anything and will not conduct digital noise to analog supply rail.

Are you forgetting about the capacitance of the junction?

A BAT41 device, for example, has 4.5 pF of capacitance at 0V Vr. A capacitor is an AC connection element, and AC current can flow even if the DC voltages are the same. Noise on a supply rail is an AC signal.

Now, is the current flow enough to cause problems? Well that depends on the situation... how much switching noise is there, how sensitive is your analog circuitry, etc.

Brian Boorman said:

Are you forgetting about the capacitance of the junction?

Yes. I miss to consider it. Agree to our considerations, thank you for pointing out. However it would be nice to know capacitance between DVCC and AVCC pins of msp430. Anyone having RLC meter and wish to measure?

Zack Bomsta said:

I ran a simulation in LTSpice on the proposed RC filter where R = 100 Ohms and C = 10.1 uF. The image below shows that the voltage difference between DVCC (blue) and AVCC (green) exceeds 0.3V.

LTSpice can only simulate within the parameters of its simulation.

The datasheet limit of 0.3V is there because the internal clamp diodes of the MSP have a forward voltage of >0.3V at a current of 2mA. So as long as a voltage difference of <0.3V is applied, the resulting rated current will be <2mA. (peak current can be by magnitudes larger without killing the diode - for a few ms or less.)

However, the current through the clamp diode will provide additional charge to the AVCC capacitor (a small current even before 0.3V are reached), so the simulation is not correct.

Zack Bomsta said:

not to use different sources for AVCC and DVCC because they may have different start up times that could lead to a voltage difference greater than 0.3V

Right. Mainly because the voltage difference may be extremely large for an extremely long time. In fact, the faster rising source might source the slower rising source (and all load on it, even externally) though the MSP's clamp diodes, which might blow them. So this is a completely different (and way more dangerous) situation. Since with the RC combo from same source, the only load is the filter cap, and charging it reduces the voltage difference, the clamp diodes partly bypass the filter at startup, relaxing the situation. Since the MSP doesn't draw any current from AVCC at startup, the power dissipated at the clamp diode while bypassing the filter is minimal. (< 3µWs)

I used this filter combo in several projects and it didn't cause any problems.

BTW, a diode (Schottky!) between the two VCCs (from the faster rising to the slower, or even two in both directions) will take the majority of the current flowing between the two VCCs during ramp-up. So the internal clamp diodes won't blow. So if having separate supplies, this is indeed a good idea.

However, the mentioned junction capacitance partly voids the filter. On the internal diodes, the junction capacitance is negligible because of the small diode size. Else you couldn't filter AVCC at all and having a separate AVCC would be futile. :)

Jens-Michael,

Thanks for the definitive response. Everything you discussed makes sense. In light of what you shared, I think the right question to ask is does 0.3V of potential between DVCC and AVCC pose a threat due to excessive current through the diodes for extended periods of time, or is it a threat because it causes latch-up almost instantaneously? If it is just a matter of limiting the power dissipation through the internal clamp diode, then I agree that the limited potential difference caused by the filter is negligible; however, if quick transients cause latch-up, then even the filter could still pose a very real threat.

Based on what you said, you have used this filter extensively in the past and not experienced problems. That leads me to think that latch-up is not the main potential for problems in this scenario.  

Zack Bomsta said:

, or is it a threat because it causes latch-up almost instantaneously?

I don't think so. During development, I have more than once exposed the MSP to conditions far beyond this and almost all devices survived without any permanent damage. I wasn't this lucky with the ATMega128 we use in our access point. :(

Of course it is still possible (and may have happened in the few cases where things went wrong), that transients faster than the diodes may cause permanent damage (degradation or destruction) of other parts like port transistors. On the other side I had MSPs which were almost reaching boiling point due to defective voltage regulators or shorted pins, which recovered fine. (Well, I only tested the required functionality in these cases)