DC Supply for digital ICs

engr_joni_ee said:

Quantitatively how much we can expect Vpp ripples in Linear Regulators and DCDC converters ?

Click to expand...

Counter question: If I buy a car, what color can I expect it to be?

Simple answer: You get what you buy. This is true for both power supplies and cars.
And if you want numbers: Read the datasheet, since every converter has different specifications.
I do this all the time - and it works. This is why I recommend it.

My recommendtion:
As always: FIRST decide what you need, then buy/design according this.

If you want a blue car: buy a blue car.
If you want a power supply with less than 10mVpp ripple, then buy a power supply with less than 10mVpp ripple specified.

For sure you may buy a yellow car and paint it blue afterwards.
You may buy a power supply with higher ripple ... then filter/reduce the ripple afterwards.
If it´s worth the effort: only you will know.

You may say it´s obvious for cars .... but for me it´s the same obvious for power supplies.

Klaus

A common approach, is usually a 3V3 SMPS for the main voltage (very little logic uses 5V these days), with local LDO regulator's close to the relevant devices, for core voltages of 1V8, 1V2 etc. This is s simple example, I have done designs with upwards of 20 power domains, usually 12+ layer designs, and I have done designs with just 1 (heaven, but rare).
It all depends on the complexity of the design, the thing to remember, if your power delivery system is not correctly engineered, then EMC and signal integrity issues will be more prevalent. The whole PDS requires engineering, layout and decoupling capacitor placement is as important as the correct supply devices.

Choosing :



Regards, Dana.

But it "general better" to have power supply and an extra linear requlator
It add a extra "filter" especially these days where all psu are smps and when they "fall" the add significant ripple and some times higher than rated voltage

it also depends on power. if you’re only drawing a few tens of milliamperes then a switcher makes no sense. If you’re drawing amps, then a linear will waste a lot of power (in heat)

Hi,

plouf said:

But it "general better" to have power supply and an extra linear requlator
It add a extra "filter" especially these days where all psu are smps and when they "fall" the add significant ripple and some times higher than rated voltage

Click to expand...

I know what you mean.
But I i don´t think "generally better" is true.

"better" in which regard? If you focus on "low ripple" .. then it maybe better, but if you "focus" .. then is it still "generally"?

***
I have done a lot of designs, where i just just well desgined SMPS. No fancy filters, just well designed, well chosen parts, optimized PCB layout.
I have seen no drawback with these solutions. Remember: the OP talks about digital circuits only.

When doing a proper design - as said above - I´m not surprised to get a better ripple performance than a badly designed PCB layout with linear regulators.
Often in the forum we see members complaining about circuit malfunction (maybe when a relay switches) - even if they use linear regulators.

When talking about analog or mixed signal circuits .. I have done many of them with SMPS where the (microcontroller internal) ADC reading is stable down to 12 bit.

***
When I need high precision, like 16 or more bits ... then I tend to use linear (post) regulators.

Klaus

barry said:

it also depends on power. if you’re only drawing a few tens of milliamperes then a switcher makes no sense. If you’re drawing amps, then a linear will waste a lot of power (in heat)

Click to expand...

power is a good point.
For sure there is a benefit of the SMPS in power dissipation .. especially with high power supplies.

To be honest - I like to use SMPS. Even for less than 50mA.
Here the problem often is that in those "low power modes" the SMPS go into burst mode operation. This causes a lot of ripple voltage and even worse with rather low frequencies. Really difficult to filter. Not even a bigger capacitor helps. Proper part selection is essential. And testing.

Klaus

Yet again my point is accurate
"Generaly " comes to vast majority of cases

And also the majority of cases of PSU in market fail output capacitors after 2-4 years.
So you must consider that this is a well proven case

plouf said:

And also the majority of cases of PSU in market fail output capacitors after 2-4 years.
So you must consider that this is a well proven case

Click to expand...

Proven by whom?
I by far did not experience the failure rate you are talking about. Maybe some crappy ACDC PSUs die within 10 years.
In post#5 you talk about "power supply AND linear regulator". If you have both ... still the PSU may fail ... independent of the additionally used linear regulator.
Adding a linear regulator will not reduce overall failure rate.

Indeed this does not mater, because the OP does neither talk about ready to buy PSUs nor about the failure rate.
The discussion is rather on (self designed) low power on board DCDC converters .. and their voltage ripple at the output.

plouf said:

Yet again my point is accurate
"Generaly " comes to vast majority of cases

Click to expand...

Good to know what "generally" means for you.
I thought about the OP´s topic "voltage ripple".

Indeed .. I think one step further.
What is the true benefit of a reduced voltage ripple in a purely digital circuit?
Cost reduction? surely not
Reliability? I don´t see a reason. Decades a go I replaced linear regularos on industrial control units by on board DCDC converters.
Before the temperature of the linear regulator was below 60°C .. but surely above 40°C even with heatsink. I had some fails of the linear regulators.
I can´t remember a single fail of the DCDC version (of otherwise identical circuits). The temperature rise of the DCDC converter was barely recognizable.
So I blamed the fails on the temperature. Since a rise of 10°C is said to reduce overall lifespan to half.
I can only tell about your experience ...
If your experience is a fail in 2-4 years .. it may be true. I won´t disagree. What do you think: why is the failure rate so high? Is it a general SMPS problem, or a design problem?

From many posts here in this forum I can tell there are lot of badly designed SMPS circuits around.

Klaus

the "proven by whom" is by statistics
dont know if your area, use only "high quality" psu

but here, in all electronics , from routers, cctv, led stripes, access etc use cheap

the OP said "if the main PC board is 5V" from this sentence, i dont understand it as "custom design" but outsourced devices

from the above i still believe that using an "uknown" psu (it can be good/bad/shared with other device/cheap" it is still better to have the psu and ON TOP of that a linear regulator on your PCB,
this method reduces rippling and add a "security" layer over a bad or broken psu about rippling.

also temperature as, correctly, you point, varies in diffirent areas
here at summer we have 40+ under shadow , and in closets,where electronics left, can go 60-70.
obviusly we have a higher fall rate that north countries based on that and only criteria..

Hi,

plouf said:

the "proven by whom" is by statistics

Click to expand...

It should be easy for you to give a link...

engr_joni_ee said:

The digital IC usually need 3.3 V or 1.8 V as supply voltage. If the main supply to the PC board is 5V

Click to expand...

so it´s 5V DC --> 3.3V DC (or 1.8V DC).

plouf said:

i dont understand it as "custom design" but outsourced devices

Click to expand...

I put "self designed" in brackets .. to show this is optional. I neither will force him to design them on his own nor to buy a ready built one.

plouf said:

from the above i still believe that using an "uknown" psu (it can be good/bad/shared with other device/cheap" it is still better to have the psu and ON TOP of that a linear regulator on your PCB,
this method reduces rippling and add a "security" layer over a bad or broken psu about rippling.

Click to expand...

Again: The 5V PSU is not the topic. (The OP may correct me on this)
It´s the DCDC that converts 5VDC down .. to the digital supply voltage.

plouf said:

also temperature as, correctly, you point, varies in diffirent areas
here at summer we have 40+ under shadow , and in closets,where electronics left, can go 60-70.
obviusly we have a higher fall rate that north countries based on that and only criteria..

Click to expand...

Makes sense.
Doesn´t this call for reduced power dissipation (generated heat) where the SMPS converters are beneficial?

Klaus

Added:
I just did an internet search about DCDC fail (reliabilty) statistics. Nut much numbers to find. But one source talked about an MTBF of 1.6 million hours.
4 years are about 35,000 h. This is a discrepancy of about 1:40.

I mean power design on custom PCB.

We have 5V DC supply as input power to the board. It's actually a Microcontroller based stepper motor control PCB with LEDs and switches etc and serial link.

I guess if the current on 3.3 V suppling power to digital ICs is within 500 mA then it should be ok to use linear regulator. The same for 1.8 V. I don't think we will have more then 500 mA on 1.8 V.

I guess if I chose SMPS for 3.3 V and 1.8 V, it will be more noisy with vout ripples. Is that true ? Linear regulators would be much better in this application.

yes
LD1117 is the common used option rated at 800mah
i would use dpak version in order to use PCB for cooling

I just notice that the thermal resistance junction-ambient for DPAK package is 100 degC/Watt. This seems bit higher.

I've seen single chip multi-output POL DC-DCs that
might be interesting to you.

Look at min-load current and figure DC-DC efficiency
there from datasheet curves.

Look at Iout*(Vin-Vout) loss for linears.

Match them up. There will be a load current range where
a Class A LDO is still better than a Class D DC-DC,
because LDOs have no switching losses.

Also look to right-sizing a DC-DC because bigger FETs
for bigger current / conduction losses cost you in switching
losses.

plouf said:

And also the majority of cases of PSU in market fail output capacitors after 2-4 years.
So you must consider that this is a well proven case

Click to expand...

My experience is much different. Caps fail when used with insufficient margin for ripple current and ambient temp. Arhennius Laws with MTBF doubling every 10 deg. C reduction from rated values.
All my MOBOs use programmable SMPS regulators with high-quality Caps that last > 10 yrs. You can see all the small coils on the MOBO for these purposes to RAM, CPU etc. It would be best if you looked there for reliable designs.


Linear regulators have very low ripple from continuous forward current and negative FB amplifying the low output impedance. Efficiency depends on the choice of low voltage drop for the required load current. The LDO noise can be as low as audio sources with ripple dependent on source/load impedance (f) ratio.

SMPS cannot achieve as low a ripple as LDO's and not suitable for sensitive analog circuits but good designs are suitable for logic design.
SMPS ripple of 50mV is normally a good result, but more and less is possible. Noise is necessary to regulate each pulse of power.

Note that most LDOs have inadequate GBWP to knock down incoming switcher ripple. LT has some that make a point of HF PSRR. If your 5V is filthy a LDO regulator may not clean it up any, just drop the average voltage. And LDO GBWP tends to fall off at both extremes of load impedance.

So pay attention to your source and your ability to filter it within cost / mass / volume envelope at least as much as your dreams of designing a custom power tree from piece parts.

Hi,

dick_freebird said:

that most LDOs have inadequate GBWP to knock down incoming switcher ripple.

Click to expand...

Good point.
How to solve this?
Using good electrolytics capacitors, made for high frequency switching with LOW ESR is the first what I do.
It´s not much effort to add let´s say a 1uF ceramics in parallel to the electrolytics.
It´s very LOW ESR helps to reduce ripple, especially at very high frequencies .. where the electrolytics loses performance.
It also takes away some ripple current from the electrolytics, so it reduces temperature rise at the electrolytics to extend it´s lifetime.

For sure PCB layout needs to be adequate.
Swithing node --> diode --> electrolytics (very solid GND connection --> ceramics (very solid GND connection) --> load
The capacitors need to be placed in the most direct line between switching node and load.
Using the capacitors with a
* thin trace
* remote branch
* bad GND connnction
makes all the effort useless.

Proper decision of the current consumtion, proper part selection (coil) and proper PCB board layout keeps the SMPS in in clean uniform switching mode.
Wrong decisions like dimensioning the SMPS for 500mA .. while the real circuit only needs 25mA is .. one of the worst things for an SMPS. It probably makes the SMPS to go into burst mode resulting in very high ripple voltage .... and low ripple frequency. Adding capacitors won´t help then.

Let´s say everything done right so far.
Still ripple is 50mVpp. But now you ahev an analog section that nees a more clean supply.
"critical analog sections" usually need rather low power. (Opamps, filters, microcontroller, MUXes, ADC...) Let´s say 25mA.
Now you simply could add a series 1R resistor for this 25mA power supply section. It drops 25mV DC ... usually not the problem at all. (3.300V vs 3.275V)
Adding a 10uF capacitor will reduce the ripple voltage down to 3mVpp (from 50mVpp).

For sure one may optimize on series resistance and capacitance ... to further reduce the ripple voltage ... for the critical node. It shouldn´t be a problem to reduce it way below 1mVpp. And it´s about no effort. A simple resistor, a simple ceramics ... at the right place.... with cheap parts you should have in your stock..
There are many other ways .. like using an inductor instead of the resistor ...

And this is all simple physics, simple math. No "magic", no "I guess", no "I hope" ...

Klaus

Thanks for replies.

There are two terms mentioned for linear regulators, LDO GBWP and HF PSRR.

LDO GBWP: I know LDO is the low drop out linear regulator and GBWP is gain bandwidth product. Usually gain bandwidth is used in operational amplifiers. I do not know how it work for linear regulators ? Internally in the linear regulator there is one operational amplifier which all the time compare the voltage divider output with the band gap voltage reference and then switch on/off a transistor to control the output level of the linear regulator. Is GBWP refer to that operational amplifier ?

HF PSRR: This is high frequency power supply rejection ratio. i guess it's the ability of linear regulator of maintaining the output power stable even though there is some fluctuation in the input supply to the linear regulator. I just check for LD1117. It is defined as SRV "Supply voltage rejection" with a typical value 75dB. How we express power supply rejection ratio in formula. I guess higher the PSRR is better, right ?

Power supply rejection is had mostly by error amplifier
(inside LDO but including pass transistor and FB pin)
AVOL. As that fades with frequency so must PSRR
(meanwhile in-coupling from VIN to error amp guts
is increasing, with freq).