DC-DC converter from scratch.

[Zanderist]

I understand that you can use little modules to accomplish this.

For me through I'm quiet interested in building one from discrete components and a simple IC.

Preferably I want a build that was totally discrete components but I understand the size would not make it a practical handheld device.

SO has anyone built a DC-DC converter from scratch?

I'm thinking about eventually starting small with a 50 watt build.

[CraigHB]

Yes a number of them. 

You can construct a converter in various ways depending on voltage input, voltage output, and power output requirements.  DC-DC converter design varies from fairly simple to excruciatingly difficult depending on the topology and the controller you use. 

There's various levels of "discreteness" depending on the components you select.  You can go with a chip that has on-board switches and built-in feedback compensation (easy) to individual components including MOSFET drivers and PWM controllers (hardest).  There's literally hundreds of make and model chips to choose from.  If you narrow down what you're looking to do I can probably recommend something for you.

[Zanderist]

What I had in mind is what is seen in this video,


My idea was to have it power 50 watts at least with 3.6 volts.

I'd say that would be a boost converter

My problem is component size would defeat the purpose. Like a diode rated for 0 amps I imagine would be the size of mosfet. Ideally it should fit in an Altoids tin, my current housing preference.

What I'm really looking for is a controller chip that does not pass current, that is adjustable. I would not mind having to figure out components I need if the datasheet of the recommended chip provides formulas just as it does in the video link.

I'm not planning on a build of it just yet I am preparing for it. I should stated in the first boost that from what I've learned in college there was just a touch on the subject and no practical demonstration. I am looking to learn it on my own and also to set myself up learn about high voltage power supplies, which was not talked about at all.

[CraigHB]

This is going to be a long reply because you are getting into a highly complex area of circuit design.

You'll get the best efficiency with a synchronous voltage mode controller using external switches.  External switches are going to allow you to obtain the lowest on-state resistance and minimize resistive switch losses which maximizes output liimits.

The down side with voltage mode is you need to use a "compensation network".  This is an array of resistors and capacitors on the feedback line that compensates for magnitude and phase shift in the LCR circuit formed by the inductor, output capacitors, and load.  Specifically the type of compensation is called "Type III" which calls for three capacitors and three resistors. 

There's a whole science behind Type III feedback compensation and the math is ridiculously complex.  It involves the evaluation of phase and gain margins, Bode plots or Nyquist plots, and lots of calculations.  Fortunately most of that can be avoided by use of a simulation that shows time domain and frequency domain circuit response, but even so there's a whole learning curve in using tools like OrCAD Pspice or other circuit simulators with the required sophistication to handle these simulations, not to mention the cost of these tools.  Furthermore, a frequency analyzer can be used to find real world time and frequency domain response on an actual circuit, but that's a very expensive tool, cost prohibitive for a hobbyist.  It's typically what's used to tune converters using Type III feedback compensation on a professional level.

Most of the feedback compensation mess can be avoided by using a current mode controller.  There's still some compensation tuning involved, but it's only a single pole/zero with one resistor and one capacitor.  It's much easier to tune and can often be done using trial and error in a simulation or on the bench.  The down side of current mode is it requires a current sense resistor which results in an efficiency hit since all of the output current has to go through the resistor.  There are current mode controllers that use the drop across the low side Mosfet to measure current eliminating the current sense resistor, but there's not a big selection of them (if any at all) that can handle outputs of 50W.

The next hurdle to overcome is component selection in terms of the inductor, MOSFETS, and output capacitors.  Often you can use a couple high value ceramics, but there's lots of caveats with those over other types of caps.  DC bias is a biggie, though ceramics have the lowest ESR, highest ripple tolerance, and smallest footprint.  The lack of "bulk" capacitance (large value caps) can make tuning tricky.  Though most converters now do not utilize bulk capacitance, it's kind of a old school way of doing things.

Inductor selection becomes involved in that you have to balance size, DCR, inductance value, and current tolerance.  More inductance allows a lower switching frequency which reduces switching losses, but increases DCR and size which increases resistive losses.  There's also ripple currents that occur as a function of inductance value and switchcing frequency that have to be considered.  You'll find most small, high output converters use a frequency around 500kHz as it provides a good all-around trade-off.

Finally in switch selection there's the tradeoff of gate charge and on-state resistance.  Depending on switching frequency gate charge can be a big issue to deal with.  You really can't go much over 20nC at 500kHz or you start running into issues in switch speed and MOSFET driver capability.  Typically MOSFET drivers are "on-chip" for controllers, but they don't have to be.  More powerful external MOSFET drivers can be used to allow the use of higher gate charge but that also increases switching losses.  The lower the gate charge the higher the on-state resistance so there's a balance to strike there.  You can spend a lot of time on MOSFET selection.

Alternately you can use an MCU based switch mode power supply using a high speed MCU that provides the required resolution for the PWM signal.  That's a whole other science.  I've not actually built one like that, but I've looked into it.  It simplifies compensation a lot since you can do it in software instead of hardware.  It's always easier to tune things in software since there's no physical components to remove and replace.  It also eliminates the need for a dedicated controller since you simply control a couple MOSFET drivers off an MCU I/O pin.

I've not touched on PCB layout which can make or break your design.  Typically you'll find recommendations in the last section of a controller's data sheet.  Pay heed to those, PCB layout is critical for a DC-DC converter.  Buck controllers are highly sensitive to any parasitic impedance on the input side.  Boosters are sensitive on the output side.  You can not try to build a DC-DC converter on a breadboard, it simply will not work, well it might work sort of, but it will perform poorly and any determinations you make will not carry over to an actual PCB design.  You have to try your designs on a finalized PCB.

All in all converter design is quite complex.  You can find controllers that offer mechanisms to make design easier, but you'll find the field narrows quickly when getting into high outputs like 50W.  I would suggest you take a look at the Linear Technology current mode controllers.  Linear is not my favorite make since their prices are considerably higher than anyone else, but they provide good documentation and a free simulator specifically designed for simulating their converter controllers and other proprietary chips.

Well, that was long, it will either discourage you or help you.  It's actually pretty fun building switch mode regulators in the sense of accomplishment when you get something working well.  It can also be frustrating as hell.  Best of luck to you in your attempts. 

[Zanderist]

I'm not totally put off from after reading, it's just that I had the impression that it was simple to build based off of text book illustrations or as simple as this video:

[CraigHB]

If you check the efficiency on something like that, you'll find it's pretty dismal.  Also those quick and dirty converters are limited in output.  I highly doubt you'd be able to pull 50W from something like that without some overheating or stability issues. 

The whole point of using a switching regulator is to maximize efficiency.  Optimal designs can be as high as 95% efficient, for example input power could be 50W while output power is 47.5W, only 2.5W lost in the converter. 

To design a converter properly you need to consider the things I mentioned in the previous post.  You can slap one together as they do in the videos you posted, but it's not likely you would get very good performance out of it.  Those converters use a current mode controller that has an on-chip MOSFET with relatively high on-state resistance.  They're asynchronous meaning they use a diode rectifier which is very lossy.  All the things that make a converter easy to design, but kill efficiency and performance.  I don't know maybe it's good enough just to build something that functions on a minimal level, but then what's the point if you can simply buy something ready to go that works a lot better.

Personally I have pretty high standards for the converters I build myself.  If I don't get at least 95% efficiency at half maximal load or if I can't achieve projected maximal outputs without excessive heating or stability issues, I'll scrap the design.  It's not worth my trouble or my pride to settle for something that does have performance I can be pleased with.  There have been more that I scrapped due to failure to meet my own standards than ones I've actually put into a mod.

There really are no shortcuts.  Sure you can do something as shown in the videos if it's just a learning exercise and I suppose that's worthwhile.  Keep in mind any converter that requires a heat sink is not a very good converter.  Good ones don't need them because they don't waste enough power to heat up that much.

Oh, there's one thing I wanted to add on that previous post.  There's actually a couple sense resistor free types of current mode controllers. I did one recently that uses something called "Inductor DCR sensing".  These are great controllers since they use the voltage drop across the inductor to sense current.  Along with the other type that senses current across the low side MOSFET, these are probably the best controllers you can use.  You get all the advantages of voltage mode (high efficiency) with all the advantages of current mode (ease of design).

[nosmo ke]

This is going to be a long reply because you are getting into a highly complex area of circuit design. ...

That's a fantastic answer Craig. It pretty much sums up everything I found designing converters from scratch a few years ago!

[miskol]

Yes a number of them. 

if i'm not mistaken, you are probably the only one successful? at least in this forum that i know of 

Zanderist, i've tried to build my own from scratch but my steps are messy. i'm into building it first before i even vape and have no full understanding about what i was doing and even how to fully test em. i've built three designs:
i) DAC (because i thought that's how you control voltage output)
ii) buck-converter (but the operating voltage is for a single cell, so kinda useless)
iii) boost-converter (failed PCB layout, it was only the 2nd time i ever designed a 4-layer PCB, some routes are too thin)

i've probably learned and experienced a lot of things that could help me if i ever want to continue do it in future but these past experiences have cost me much. so i did some other things to recover the expenses and one of them is taking the easier approach of designing the control system for market-ready buck converters out there (OKR, OKL, Raptors etc).

[CraigHB]

if i'm not mistaken, you are probably the only one successful? at least in this forum that i know of

Yeah they're tricky bastards, probably scrapped more designs than I've actually used, though I've got a number of them to work well from micro-power buck converters to relatively high power boosters.  Done a couple three buck-boost converters that worked well, but not any that put out enough power for an e-cig.  I do various electronics projects, only a portion have been e-cigs.  The most interesting one I did was a SEPIC converter, but I scrapped it due to poor efficiency, it did work well other than that.

[miskol]

interesting news that you have done few buck-boost converters that is working although not for  e-cig application. are you gonna design any new e-cig project next?

i have my eye set on a few TI's buck-boost module, one of them is TPS43000. a lot of my IC search for buck-boost converter found that these ICs requires at least 4 MOSFETs, only a few of their given example shows the use of only 2 MOSFETs, which is my preference, smaller number of components the better. however this TPS43000 seems to only be used for either buck or boost but not buck-boost.

[CraigHB]

Haven't been doing much with the e-cig stuff lately.  I have a design works well for me and have been happy with it.  There's other projects I work on so I don't want to spend all my time on the e-cig stuff.  I was working on a high power buck-boost converter, but that's on the back burner for now.

The field narrows for controllers when looking at outputs upwards of 40W.  The best controller to use IMO is a current mode controller that utilizes inductor DCR current sensing.  Also one that uses a high side NMOS driver.  There's not a huge selection of controllers like that.

I've been using the TPS43000 and it's a pretty good controller but has its limits.  It's an older chip designed when power MOSFETs tended to have higher on-state resistance.  I found through tests on the bench it doesn't work well with newer transistors.  The reason is the TPS43000 senses voltage drop across the MOSFETs to detect cross conduction.  With the lower on-state resistance of newer parts the controller fails to detect it properly and regulation destabilizes.

I bench tested a Linear Technology current mode boost controller that uses inductor DCR sensing which provides similar efficiency to a voltage mode controller.  It worked well and I was thinking of doing an e-cig board based on that, but I haven't got around to it.  That one uses a high side NMOS which is better than the high side PMOS used by the TPS43000.  PMOS can be a problem for low voltage stuff since you may not have a large magnitude in gate-source voltage available.  Controllers that use a high side NMOS incorporate a simple boost circuit to drive the high side switch which works a lot better.  They can output lower voltages as the gate-source drive is independent.

When looking at high power buck-boost really the only viable option there is to use a 4 switch controller and for high outputs you can count what's available on one hand.  I actually spent a good amount of time looking at high power buck boost and I still have not come up with anything that does not involve a lot of engineering.  It would be possible to design a digital 4 switch converter using a high speed MCU, but the problem is high power consumption for MCUs with enough speed.  MCU based power supplies are not particularly suitable to battery powered devices.  Another option is to link a separate buck converter and boost converter, but then it quickly gets out of hand in terms of complexity and part count.

A high power buck-boost converter design suitable for an e-cig has to be one of the most complex and difficult projects I've looked into doing.

[Mayor]

In regards to buck and boost converters, what do they use on commercial boards like Yihi sx350j?

[miskol]

in general, they might be using a buck/boost controller with external mosfets.

to try and get more glimpse of exactly what they are using, i think you'll need to purchase one and scour the components on the board for part numbers. good luck though, china companies prefer their own components if i'm not mistaken.

[Mayor]

Yes, I'm probably screwed on that. I just figured someone here may know somehow. I'm learning Arduino to make a mod controller and am just trying to figure some hardware issues out.

[CraigHB]

In regards to buck and boost converters, what do they use on commercial boards like Yihi sx350j?

YiHi does a pretty job of keeping that stuff unidentified.  A lot of Chinese makers use Chinese chips that don't even show up at big electronics shops like Mouser and Digikey.  Unless you can read Chinese, you wouldn't be able to look them over on a company web site either.  Companies with financial horsepower can even contract a chip maker to design and manufacture a custom chip.  It could be done with either on-chip switches or external switches, more likely external switches since it would be a lot cheaper.

In any case, there's a number of ways to build a buck-boost converter in terms of what's commercially available in the mainstream.  They can be done digitally using an MCU, but that requires an expensive high speed MCU like the 2000 series from TI and they're pretty power hungry.  A typical MCU does not have the PWM resolution to do it.  You need PWM resolution into the nano-seconds which requires a 32 bit high speed MCU.  Can't imagine YiHi is going that route with the low cost on their boards. 

In terms of what's available to a hobbyist, TI just released the LM5175 which is a wide voltage buck-boost controller utilizing external switches.  It's ideal for use in building a high output, wide voltage range buck-boost mod.  I would do one myself based on that chip, but I'm doing other projects right now.

[Mayor]

CraigHB, thank you so much for your time and in-depth answer. To give you some more info: I am wanting to build a 100+ watt device, and I see the LM5175 chip has a max output of 1.8A. I'm not sure how I'd achieve my goals in power without huge lipo cells with it. I am wanting to use 18650 batteries. I am a newbie to electronics though, maybe I am missing something?

[CraigHB]

Where are you seeing current limited to 1.8A?  If you're looking at the gate drivers, they're limited to 2A which would be a reasonable maximum.  A 2A driver limits gate charge on the MOSFETs to 20nC or so which is plenty high enough for modern high performance MOSFETs running a typical 500kHz switching frequency.

The only limitation in output power comes from the input current sense resistor with a threshold of 76mV for buck and 170mV for boost.  Using a 2 mOhm current sense resistor would get you an input current limit of 85A for boost and 38A for buck.  With a 3.7V input, that results in an input power limit of 300W for boost and 140W for buck.  Output power would be input power less losses.  The output current sensor is optional and I would opt for not using it myself.

The one ding against that controller is it does not offer the inductor DCR current sense feature or the ability to adjust current sense threshold.  Running with a current sense resistor as low as 2 mOhms may be an issue for the control loop.  It may result in stability issues.  You'd have to build one on the bench and test it to be sure.  I've been able to get current mode controllers to work fine with current sense resistance that low in the past.  There's no guarantee though.

It's possible to get around the 3.5V input voltage limit by using a simple voltage doubler to power the chip.  I've done that before successfully with controllers that use high side NMOS drivers.  That's where the limitation comes from.  Driving the chip with a voltage doubler gets around that limitation and allows the use of a single cell down to 3V typical or possibly even down to 2.7V.

Oh, I should add there's a big dependence on what you use to power your converter.  They are sensitive to input supply impedance meaning a single high drain 18650 would not cut it for outputs over about 60W.  You would have to use a set of parallel high drain 18650s or better, use a set in series.  Series cells would also remove the requirement to use a doubler to power the chip.  It would be possible to use a large single LiPo with a high drain limit, at least 40A for an output limit of 100W.  I actually have some 2200mAh 20C LiPos laying around that would be perfect for that.

[Mayor]

Awesome info. Forgive me for the stupid newbie question, but would you mind telling me how you calculated that the gate charge in a MOSFET would be limited to 20nC? I can't figure it out so far.

As far as the 1.8 amp output thing, in the description on the Mouser page for it it says output current is "1.8A", but I thought I remembered it saying "max output current". I didn't get it from the data sheet or anything though. But lthe MOSFET would amplify the current anyway.

Next time I make a mouser or digikey order I may buy this LM5175 chip and try it out though. Other than the potential stability issues you describe it sounds pretty good. Worth a shot at least. I will try to find some more good options for other chips to try too. So cool. I've been lurking here and on reddit for a month or so but I'm glad I decided to join and ask some questions

[CraigHB]

There's a formula for calculating current demand based on gate charge and switching frequency.  I don't have it off-hand, but you see it in data sheets every now and then. 

The gate connection of a MOSFET can be modeled with a series capacitor and resistor.  So understanding the step response of an RC circuit will allow you to understand the way a gate drive behaves on a MOSFET.  If you look at a simulation (which you can easily with Linear's free LTSpice), you can see that current flows in and out of the gate in an exponential manner, kind of a spikey looking waveform.  It's exactly the same as charging and discharging a capacitor. 

When they specify the limit on a gate driver, it's in terms of peak current.  It can get pretty high when gate charge is high.  I just know from looking at these things often enough that a 2A driver will allow you to go up to 20nC or so without issue.  More powerful drivers are not uncommon with limits upwards of 6A for MOSFETs as high as 60nC.

The gate charge of a MOSFET switching at high speed can result in considerable losses.  Again there's a formula for power loss which I don't recall off-hand.  For battery powered stuff it's a consideration so even if a gate driver can handle a higher gate charge it's still something to avoid in terms of power loss and decreased efficiency.  Not a problem though, TI makes some really nice low on-state resistance MOSFETs with low gate charge.  Vishay also makes some nice high speed MOSFETs.

Selecting components for a DC-DC converter can be a rather invovled process.  Select your MOSFETs for lowest on-state resistance and gate charge.  Select your inductor for lowest DCR and highest saturation current.  I usually run a switching frequency around 500kHz, it's a good all around trade-off.  With that frequency I usually use a 680nH inductor.  You could use a 1uH inductor to reduce inductor ripple current, but you get lower DCR with a lower inductance value.  I don't get too concerned with inductor ripple current since the converter is only powering an atomizer, no concern with electrical noise on the output. 

Pin the chip for forced CCM.  It's always better to avoid DCM if you can.  Allowing DCM mode is only a benefit for power sensitive applications and can result in stability issues running with no load.

As the LM5175 is a current mode controller there should be little issue in tuning output capacitance.  Current mode controllers are less sensitive to things that can result in stability issues.  I would suggest two of the TDK C3225X5R1C226K250AA in parallel on both input and output.  If you're shooting for output currents over 20A, you should run 4 of them in parallel on both input and output. 

The data sheet may recommend a particular capacitance arrangement, but MLCC arrays are what I like to use regardless of what the data sheet says.  A good option to the MLCC type of capacitor is the polymer electrolytic type also known as a POSCAP.  Those are really nice capacitors and I use them when space consumption is less of an issue.  Though pay attention to max ripple current with those.

[Mayor]

Thanks again, Craig; truly educational. I've only been studying electronics casually for a month or so and this seems a bit over my head for now, but I understand a lot better now.

I'd have to ask a bunch more stupid questions and would need a lot of help along the way to pull this off at this point though so I'm probably going to wait until I'm better educated to attempt something like this.

[CraigHB]

Well there's a lot AC of theory involved in DC-DC converters and the math can get out there pretty fast.  Though a lot of times the best way to learn is to build a circuit then deal with only the education you need for that particular circuit.   

Unless you already have a science degree of some kind, understanding the math behind all this stuff is probably not worth the trouble.  Just use a simulator and let it do the math for you.  The DC stuff is no problem, just simple math, but the AC stuff is not really something you can punch out on a scientific calculator.  Some of it yes, but most of it no.  At one time I could do it all with pencil and paper, but I haven't bothered in a long time, always rely on a simulator.  I would have to re-learn it all to do it on paper again.

If you want to understand DC-DC converters go to the Linear web site and download LTSpice;

http://www.linear.com/designtools/software/#LTspice

Then pick a controller of interest and download the demo circuit for it, example;

http://www.linear.com/product/LTC3786

Run the LTSpice simulation and observe currents and voltages throughout the circuit to understand its operation.  Then you can tweak things to your own specifications and check performance.  Once you get your circuit working well in the sim, design a PCB for it and build it.

LTSpice is not a general simulator, rather it's proprietary.  However it has a lot of the tools you need for general simulation, but you won't be able to drop in models from other makers.  You certainly can build a converter with a Linear controller, but that can be limited.

TI has a free spice simulator called Tina-TI for its controllers, but unfortunately they don't always post Tina-TI models.  They always supply Orcad PSpice models but not always Tina-TI models.  There's some differences in Spice programs even for the same variant that don't always allow models to be interchanged.

Circuit simulators actually use a high level programming language under the graphic interface that is a little strange, but once you understand how it works it's pretty powerful and easy to use.   There's different ones like Berkeley Spice, HSpice, PSpice, and several other variants.  LTSpice and Tina-TI use the PSpice syntax which is pretty much the industry standard now.