# What determines SMPS Efficiency?

Discussion in 'The Projects Forum' started by AfdhalAtiffTan, Dec 27, 2012.

1. ### AfdhalAtiffTan Thread Starter Active Member

Nov 20, 2010
117
11
I've been tinkering with SMPS lately...
I searched over the Internet and various forums, now, I'm even more puzzled.

I just can't get over this, what makes specialised SMPS chip like the one from Linear and Maxim so efficient than my homemade one?

I tried to decode their operational block, still, I can't get it.

All of my design doesn't exceed 60%, despite topology.

Anyone here experienced in SMPS could explain what makes those expensive chips efficient?

So far, what I'm come out with is; it is directly related to switching frequency, switch saturation voltage, inductor's copper losses, diode's forward voltage, inductor's saturation point and topology.

I played with Mazzilli's ZVS, so far the best that I can get, but, it gives very little control and it requires high performance caps.

I'm eagerly curious need to know how to properly design a decent SMPS...

Any thought guys?

2. ### crutschow Expert

Mar 14, 2008
13,014
3,234
Your list pretty well covers the main factors in determining the SMPS efficiency except for switching speed. To properly design a high efficiency SPMS you need to calculate the losses from all those sources and determine which ones you can further reduce to improve efficiency. One big factor can be the switching time of the MOSFET. If you don't drive the MOSFET gate with a high current driver to rapidly charge and discharge the gate capacitance, the resulting slow switching time can dissipate significant power.

Depending upon the SMPS output current, you may be able to determine what parts are dissipating too much power by how warm they get.

The expensive chips are not necessarily more efficient, they just include all the control electronics you need in one package, so it takes up less space, and it likely cheaper then the cost of all the individual components to perform all the same functions in a discrete design.

If you post your design circuit diagram we can perhaps comment further.

3. ### bountyhunter Well-Known Member

Sep 7, 2009
2,498
507
You forgot a big one: switching losses. When the transistor or diode turns off, there is significant power losses associated with that transition. That's why they try to drive the FEt gates really hard so the transition through the turn OFF as quickly as possible. With diodes, they build them for ultra fast "recovery" to reduce losses.

Another one forgot: capacitor losses due to the switching current flowing in and out through the cap's ESR. Can be significant as frequency goes up.

4. ### AfdhalAtiffTan Thread Starter Active Member

Nov 20, 2010
117
11

Thanks both @crutschow and @bountyhunter; always give me delighting replies...

I just looked up online and planed to give a "MOSFET driver" a try. I've ordered TC4429, it has 25nS of rise and fall time!

My knowledge is limited, this is the best design that I have come out with in my entire month.
I played with TL494 and MC34063, might be LM2576 in future, it claim to have efficiency of 70% ish.

Almost nothing gets uncomfortably warm, this puzzles me because I can't found out where the wasted-heat go.

I used a Sziklai pair to aid saturation.

Above is the picture of the circuit in action, it regulates the current go through the LED at about 700mA, the current reference was from diode forward voltage, hence the familiar number.

I'm using small-power PNP, BD140, rated at 1500mA, drived to saturation, barely warm while in operation.

The NE555 output measured to have rise-time of approx 0.8uS.

Not shown in schematic, it does have some decoupling caps.

The inductor tested to saturate at about 1.8A, in operation, I could hear the inductor high-pitch sound.

Here's the circuit, green trace is powerIn, red is powerOut.

At the 5th pin of NE555, there's a diode as my voltage reference.
There's no timing capacitor, the frequency is determined by storage inductor, filter capacitor and load.

I've done a crude test, with the LED as the load, the result is close to 58% efficient, it varies with load of course.

It might be bad design, but in this purpose, I don't mind much about the ripple current.

Above picture shows current trace across 1R dummy load, it does gets hot obviously, so I run it for short time.

The ripple is from 0.5Iref to 1Iref, because the nature of the NE555 chip used.
That's very bad relatively to properly designed SMPS, but as stated before, it was designed for LED, so I guess it doesn't matter much.

Thanks again for the help...

5. ### crutschow Expert

Mar 14, 2008
13,014
3,234
What is the output voltage of the Q1 transistor emitter when ON? You are likely dropping significant voltage due to the high saturation voltage of the transistor pair and the fact that the output of a 555 does not go all the way to the positive rail.

You could improve the efficiency by using a P-MOSFET for the switch (and inverting the switching signal since 0V turns the P-MOSFET on).

Edit: Your 1 ohm sense resistor is also dissipating about 0.7W which is another source of inefficiency.

Last edited: Dec 30, 2012
6. ### Papabravo Expert

Feb 24, 2006
10,144
1,791
One more thing is the inductor. Did you purchase the core and wind it yourself or did you buy it as a finished part? If it is the wrong material, or the wire is too small or it has too many turns you could be sacrificing some efficiency by saturating the core or heating it up.

7. ### kubeek AAC Fanatic!

Sep 20, 2005
4,670
804
Alt click on a part to see a graph of power dissipation, then ctrl click on the trace label to see average power.

8. ### bountyhunter Well-Known Member

Sep 7, 2009
2,498
507
Looks like the frequency is about 1 kHz which is why you hear it whining.

9. ### AfdhalAtiffTan Thread Starter Active Member

Nov 20, 2010
117
11
@kubeek
Yup, I've done that, just realised I forget to take other component into account... Silly mistake.

This is the 555 rise time, I guess I measured it wrong last time (maybe it changes with PSU?).

Here's the fall time.... Significantly better than rising.

@crutschow
The upper blue trace is the supply voltage, yellow is the Vc of Q2, with respect to Gnd.
I'm surprised, there's about 1.4V of Vdrop relative to Vcc. Thanks for pointing that out.

Also, thanks for pointing out about the resistor, now I'm wondering how could I miss that?
Now I'm waiting for my mosfet driver to arrive.

This is the 555 output.

@Papabravo
This is the current measured across inductor via the 1R resistor, no sign of saturation.
The inductor is bought at my local electronics supply store.
It does gets a bit warm after run for long enough, maybe it's just ohmic losses.

@bountyhunter
It's running at acoustic-range frequency...
The frequency is affected by supply voltage too, it's about 2kHz at 7V and close to 8kHz at 13V.

If measuring current via a 1R is so wasteful, how could I done better?
Using 100mΩ gives voltage drop way too small to be measured.
Using opamp to amplify it would be too slow.

Posting this question to public somehow make me realise how careless I was.
This what makes electronics community accelerates my learning curve.

Thanks so much guys!

10. ### crutschow Expert

Mar 14, 2008
13,014
3,234
At the low frequency of your circuit, a high speed op amp (something with a GBW of 10MHz or more, depending upon the gain you want) should work. A more complex approach to measure current that doesn't use a shunt resistor is to use a current transformer or Hall Effect device.

11. ### takao21203 Distinguished Member

Apr 28, 2012
3,577
463
Nice circuit. I use a TL494 board to supply a 30W power LED.

Output voltage is 33.5V at 1 Amps.

Input is about 50V. But it is capable of 100V.

Efficiency is nearly 80%. It mostly depends on the right type of coil.
But I don't have oscilloscope, or any expensive equipment for measurement.
All I use for that is a DMM, and some analogue multimeters.

The transistor is a special PNP power transistor, which I had to order. IRF MOSFETs also work, but easily blow out when starting larger loads.

The cooler was not so easy to mount- there are 3 semiconductor devices underneath it, and they all have to sit in the right place.

File size:
203.9 KB
Views:
36
File size:
206.5 KB
Views:
37
File size:
101.5 KB
Views:
37
12. ### bountyhunter Well-Known Member

Sep 7, 2009
2,498
507
Looks like it's switching at 8 kHz. That noise would drive me nuts....

13. ### AfdhalAtiffTan Thread Starter Active Member

Nov 20, 2010
117
11

80%? That's impressive! What did you meant by special BJT? Could you explain more?

@crutschow
Thanks for pointing out about the alternative method of current measurement.

I just played with LM2576.

Modified it to work with external transistor, having trouble to fully saturate it.

14. ### takao21203 Distinguished Member

Apr 28, 2012
3,577
463
It is simply a very large power PNP.
Ordered from Farnell. It is kind of a 150 Volts, 30 Amps device.

I have 3 lamps running now from the regulator.

Current is about 3.5 Amps at 32 volts.

One of the coil turns more hot than the other, they both get some airflow from the cooler...

Today some hours ago I added a new Schottky rectifier. The 6 Amps diodes turn very hot already. Same as the LED lamps, I glued 4 DPAK diodes to a heatsink. Works very well, only heats up a little, and 1V more DC!

Then I tried the 4th lamp directly from toroid DC (which is 47 volts now).
0.7 Amps! All together I want to run 6 lamps (each 30W, one 40W).

Ordered some LM2576 HV for the additional lamps!

The effectivity is not always 80%, depends on the output voltage + current.

Right now, input is 47V, output is 32V, and more than 3.5 Amps current.

The VGA cooler turns warm slightly but that's it. All the current runs through one large power PNP.

So this is the first permanent longtime test for this circuit.

It is not so easy to build a SMPS with more than 100W.
You need large toroid coils with the right properties, or they heat up a lot and efficiency will be bad.
More than 200W is almost only done with special flyback transformers.

The base resistors also turn very hot. I use 3x 1W in parallel, on the cooler surface, but they have blackened already.

• ###### P1050088.jpg
File size:
211.4 KB
Views:
32
Last edited: Jan 5, 2013
15. ### AfdhalAtiffTan Thread Starter Active Member

Nov 20, 2010
117
11
@takao21203

Just today, I believe I gathered enough information on why all my design failed miserably: My switching element doesn't saturate as it should be!

I modified the LM2576 to run with external NMOS, with a NE555 as my high-side driver. Almost all components are cold to touch, even at full designed current, without heatsink.

Apparently, BJTs are pretty hard to work with (to me), especially the ones with high-rated current; the gain is too low to be practical, for instance, TIP2955 was like about 20.

Saturating BJTs requires high Ib, which's another power waste, hence, making MOSFET a better choice.

MOSFETs however, pretty sensitive voltage-controlled device. Running it at low gate voltage wont saturate it enough, hence the need of a 'MOSFET driver'. [pointed by crutschow and bountyhunter]

NMOS being more popular among hobbyist (at least me), assigning it to high-side requires a high-side driver, which's in most cases are bootstrapped ones.

Ordering one would takes time, and thus killing my enthusiasm. Constructing from discrete BJTs wont satisfy the need of fast rise/fall time of the MOSFET gate, this hesitation leads me to design one which runs with NE555.

Being impressed by takao21203 efficiency motivates me to spend for like 4 days cramming my brain on MOSFET's driver design, it was new to me, so it hurts me a bit...

After satisfied for the MOSFET being cool to touch, next thing to target would be the Schottky diode used in the design.

I'm thinking to play with synchronous Buck SMPS, but it just beyond my capability right now, maybe in future.

I guess my past problem now solved.
I just don't know how to thank you guys enough...

Thanks again everybody!

16. ### takao21203 Distinguished Member

Apr 28, 2012
3,577
463
It's pretty normal for large BJT to require a smaller driving transistor.

If the load is low, the resistor divider does not draw so much current.

Only when the base is pulled low, large currents flow.

The circuit I built equally works for MOSFET, or BJT, interestingly.
One smaller BJT is used to drive the base.

I think 200 Ohms on the high side, and about 800 Ohms on the low side, switched with a smaller BJT.

At only 12v or 20V input not so much of a problem, but if you have 60V or more on the high side, you need suitable components.

The TL494 is old, but still a good design. More flexible than the lm2576. Yes it is possible to use external MOSFET or BJT, but the lm2576 has internal current limit. It is not so easy to get rid of it, since it is internal only.