First time designing with an IC, a few quick questions.

[engineheat]

I'm interested in designing a boost converter using the TPS61021A chip from TI.

Data sheet: http://www.ti.com/lit/ds/symlink/tps61021a.pdf

On page 9, there is an equation to calculate the required duty cycle, and you need to know the efficiency. It recommends using an efficiency of 90% for "most applications." On page 6 they even provide efficiency curves for a few different input/output voltage scenarios.

But mine applications requires an input voltage and output voltage that falls outside of the curves provided. How will I know what efficiency value to use then? Am I expected to calculate it in a lab setting or are there more detailed resources I can look into for this IC?

This is my first time designing using an IC, so just wondering what is normally done in situations like this.

As a side question, are all TI chips widely available in nations that TI does business? like China? I might make a batch of PCBs in China in the future.

thanks

[engineheat]

I also realize you can't built a prototype circuit on a breadboard for a high freq switching device like this. Can a solderable PCB prototyping board work decently? I want to build something that at least work (for learning purpose) before I do a PCB design with the help of others.

[basinstreetdesign]

This chip looks like a reasonably well-behaved chip for boost applications.  The big caveat for me is that it operates at a ("quasi") fixed 2 MHz.  This is a very high frequency for any regulator.  I do not recommend building this circuit, even if it is straight from the datasheet, on anything other than custom PCB.  Certainly a push-in protoboard is out.  I dont have much confidence in even a solderable PCB prototyping board.

I would do the layout of a custom board just as they describe in the datasheet and either get it made by one of the cheapo proto houses (OSHPark, SmartPCB, etc...) or make your own DIY board on a two-sided copper clad.  Use the opposite side as a ground plane.  Do NOT put any other project circuitry on the same board as you dont want to waste $$ or time if the switcher smokes and destroys another part of your project.

When choosing parts, follow the datasheet recommendations to the LETTER.  Buy the best parts you can find, money is no object.  Dont use chinese electrolytics instead of high-quality ceramics just because you can save $1.25.

Chances are the most important component in the design will be the pcb and its layout and it will be a major influencer in the operation of the circuit.

As for the efficiency use on page 9, the figure you choose doesn't need to be very precise.  90% is OK.  If it turns out that the circuit operates at 80% efficiency then just count yourself lucky to have survived to that point.  In fact count yourself lucky if you dont release the magic smoke at all.

Good Luck!

[engineheat]

This chip looks like a reasonably well-behaved chip for boost applications.  The big caveat for me is that it operates at a ("quasi") fixed 2 MHz.  This is a very high frequency for any regulator.  I do not recommend building this circuit, even if it is straight from the datasheet, on anything other than custom PCB.  Certainly a push-in protoboard is out.  I dont have much confidence in even a solderable PCB prototyping board.

So the higher the frequency, the more demanding everything has to be solid. Why is that though? This is the answer I found online:

"The problem with using a breadboard is that the current paths are restricted to (often) light-gauge pressed-metal conductors which aren't particularly good at handling fast edges incurred by the high switching speeds of a modern boost converter. Just be aware that the converter will almost certainly be electrically noisy (due to the less-than-ideal current paths)"

So is it due to noise caused by the above reason?

Chances are the most important component in the design will be the pcb and its layout and it will be a major influencer in the operation of the circuit.

So what's the worse that can happen? You mentioned smoke a lot. Is it the inductor or the IC causing the smoke usually? Worse case scenario, does the component just destroy itself and the circuit stops working or can it actually burst into flames and be a fire hazard? I just need to know for liability purposes.

It seems like designing a boost converter, even with an IC, is a lot harder than I expected. I guess even if I work with a professor and make a custom PCB, there will still be a lot of troubleshooting before the device can be "market ready" right?

Lots of questions, but I need to know since I might want to eventually commercialize an application that requires this. If it turns out to be too risky I might just use more batteries instead...

Thanks for your advice.

[basinstreetdesign]

One more thing - when you scope its waveforms make sure you use one at least 250 MHz bandwidth and probes to match; 500 MHz would be better.  And no long ground leads - 3" max; 1/2" spring lead would be better.  And dont forget to compensate the probes with the scopes internal calibrator!

[TheUnnamedNewbie]

So the higher the frequency, the more demanding everything has to be solid. Why is that though?

Our lumped-element models, with ideal components ^[1] are really only applicable to low frequencies (steady state DC). However, even at low frequencies they can give us accurate models of how circuits behave (how low depends on the application and sizes. Long distance power lines already have to factor in the transmission line effects, even though they operate at 50/60Hz. On-chip we can often ignore some of these effects up to a few GHz for small enough nodes). On a PCB, at 2MHz, this will already pose an issue. Switching converters are even more sensitive than most because they tend to have very high spikes of current trying to get places, big inductors and capacitors that have a good amount of parasitics in them and form a lot of losses, etc. This is one of the reasons why it will give issues on a protoboard (if it works at all!) - the protoboards give you a large amount of stray inductances and capacitance everywhere making most high-speed circuits act very strange (because suddenly they have a few pF of capacitance between pretty much any node on the circuit and tens of nH of inductance between every node).


^[1] Resistors have no inductance or capacitance, inductors have no series resistance or capacitance, capacitors have no leakage, series resistance, inductance, etc.

[basinstreetdesign]

So the higher the frequency, the more demanding everything has to be solid. Why is that though? This is the answer I found online:

Generally, yes, everything involved in the circuit becomes more significant.  Every conductor has series inductance, series resistance and parallel capacitance to every other conductor in the vicinity, including you.  Straight wire has about 10 nH inductance per cm length or 1nH/mm.  To put that into perspective, the inductors recommended for this chip are from 470 nH to 1000nH.  So a total of 1 inch extra distance that current has to travel around the ON or OFF current loops is already an extra 10% of the component inductance that does not appear on the schematic and affect the available duty cycle and consequently, max load that can be supported.  These loops through the inductor or the switches may carry several amps of current.  An extra 1 inch of thin copper path can have a voltage drop of 20-50 mV and be an appreciable fraction of the feedback voltage.  This can drive the stability of the regulator nuts for no apparent reason.

The push-in proto boards are notorious for making long paths in circuits built on them with relatively high contact impedances between wires and the sheet metal rails that everything connects to.  They are terrible for high-frequency work.

If you don't want to go to the trouble of a manufactured pcb then you might construct a proto over a ground plane by using copper-clad and using Manhattan-style or dead-bug construction.  This style can work up to 100 MHz.  If you dont know what this is just google "dead-bug construction".  Also, for the sake of low parasitic inductance/resistance construction, use only SMT parts, not TH with leads.  In fact, it might be the best option to first build a dead-bug style version over ground plane then go for the pcb once you know it can work

So is it due to noise caused by the above reason?

...

So what's the worse that can happen? You mentioned smoke a lot. Is it the inductor or the IC causing the smoke usually? Worse case scenario, does the component just destroy itself and the circuit stops working or can it actually burst into flames and be a fire hazard? I just need to know for liability purposes.

It seems like designing a boost converter, even with an IC, is a lot harder than I expected. I guess even if I work with a professor and make a custom PCB, there will still be a lot of troubleshooting before the device can be "market ready" right?

Lots of questions, but I need to know since I might want to eventually commercialize an application that requires this. If it turns out to be too risky I might just use more batteries instead...

Thanks for your advice.

The push in proto board itself doesn't cause noise, so much, as it introduces a lot of parasitic components which become so much more important at high frequencies and high currents.

As for smoke...
So many people, including engineers very experienced in other circuit areas have smoked SMPS regulators on their first (or 6th try).  I know, I did!  Follow the heavy current paths.  The ON current path is the input cap, inductor, lo-side switch, and back to input cap.  If the controller chip fails for any reason and leaves the switch on too long then both the switch and the inductor are toast.  The on-going current will be supplied by a source power supply that will be oh so happy to supply all the circuit wants through the inductor and the switch.  That wipes out both the inductor and the chip.  I hope you have extras.

Actual fire is not likely but a puff of actual smoke can occur.

If all of this seems alarmist, it's not.  But good results can happen if you don't get cocky and follow the chips datasheet recommendations as well as possible, using high-quality physically small parts (inductors with sufficient saturation current and low ESR, capacitors with low ESR, non-inductive resistors) and a small, tightly collected layout with optimally-short current paths.

[engineheat]

Wow. Thanks for the very helpful answer.

I've been carefully studying the instruction manual multiple times. I will definitely follow all the recommendations.

[engineheat]

If you don't want to go to the trouble of a manufactured pcb then you might construct a proto over a ground plane by using copper-clad and using Manhattan-style or dead-bug construction.  This style can work up to 100 MHz.  If you dont know what this is just google "dead-bug construction".  Also, for the sake of low parasitic inductance/resistance construction, use only SMT parts, not TH with leads.  In fact, it might be the best option to first build a dead-bug style version over ground plane then go for the pcb once you know it can work

Is a copper-clad a thin layer of copper laid on top of a insulating layer or can a solid sheet of copper also work? I have some spare sheet copper.

I understand with a "dead bug" construction, you solder points together without dealing with breadboards for better connection and more closely packed components.

So I use the copper clad as ground. All grounded points will be soldered directly to the copper. For connection points not on the ground, do I just build it in "mid air" using some kind of insulator on top of the copper? I believe that's what's done in the following pic:

 

The TI data sheet also makes a differentiation between power ground and signal ground.  What is the significance of this difference and do I need to treat them separately in my construction, or do they all get soldered to the copper plane?

thanks

[engineheat]

Okay, after some research, I realize now that a "copper clad" is merely two layers of copper with fr4 in between. It's pretty much raw PCB material.

I decided to go with the Manhattan approach, just building small insulated "islands" on top of the copper clad.

You mentioned "Use the opposite (bottom) side as a ground plane", but I think you only need to do that if you etch or carve the top side as your circuit.

If I'm building Manhattan style on the top side, then can't I just use the top side as ground? The pic above seems to just use the top layer copper as ground plane, without using the opposite side (otherwise there should be holes drilled).

I also read that "Ground planes are sometimes split and then connected by a thin trace. This allows the separation of analog and digital sections of a board or the inputs and outputs of amplifiers. The thin trace has low enough impedance to keep the two sides very close to the same potential while keeping the ground currents of one side from coupling into the other side, causing ground loop." So I probably should do that for my power and signal grounds too?

Thanks. If I understand it correctly I'm just going to order everything from Digikey and get started...

[basinstreetdesign]

Sorry to take so long responding.   
Yes, "copper clad" is exactly what you say and if you use it, then single-sided board is enough.  The copper is always used as a universally-available ground sheet which will have extremely low impedance from any point on it to any other point on it.  The little insulating islands you see in your picture are, IMHO, unnecessary, just connect the leads (or ends of SMT parts) directly to each other.  The 2nd shot below shows what I did for a 10 MHz WWV receiver.  You can see the reason this style is called "dead-bug" as the chips and transistors are upside down.

The 1st shot shows a boost converter I built on DIY single-sided pcb for a tube-based project.  That one converted 12Vdc to 125 Vdc and worked quite well. but then it ran at only 100KHz.  Your circuit would run 20 X faster so that is why my emphasis on watching your p's and q's about the design and parts choice as well as circuit layout.

One thing I just realized is that the controller chip has a thermal pad on its bottom used for DGND.  This pad is also used to get heat out of the chip.  If you put the chip upside-down on the copper with the pad on the top then the DGND is readily available but you may still need to wick off some heat from there too.  Maybe a chunk of large-gauge copper wire soldered to it may do that, something like a 1-2" piece of 14 AWG house wire soldered to the pad at its middle and just sticks up in the air on both ends.  Normally the wide ground tracks of the PCB would do that.

Anyway, I am glad you decided to go ahead with this build.  I am very interested in anybody who is eager enough to stick their neck out in unfamiliar territory even a little bit in order to learn something.

[engineheat]

The 1st shot shows a boost converter I built on DIY single-sided pcb for a tube-based project.  That one converted 12Vdc to 125 Vdc and worked quite well. but then it ran at only 100KHz.  Your circuit would run 20 X faster so that is why my emphasis on watching your p's and q's about the design and parts choice as well as circuit layout.

I see an IC in your 1st shot. Is it a dedicated IC for boost converter (like the one I'm using) or is it just a generic microprocessor? Most commercial ICs I can find for boost converting (like those from TI or Linear Technologies) switch at Mhz level frequencies. I doubt my application requires that.

How did you get it to run only at 100Khz? Did you program the switching logic yourself?

[basinstreetdesign]

Yes that IC is a dedicated SMPS controller, not a uC/P.  It is a LM3478 and can be programmed with a resistor to run anywhere from 100KHz to 1MHz.  I chose that one because it had no internal switching FET and I needed to be able to use a switch of my own choice which could stand up to Vds(max)=200V.  I ran it at 100 KHz precisely to reduce inefficiencies caused by parasitic ringing, or switching times which were not super fast.  It worked out OK (I got efficiency about 90%) but the chip does have an irritating (and undocumented) idiosyncrasy.  If the input voltage drops below about 7 volts it goes "high impedance" and turns the switch on PERMANENTLY.  This conveniently smokes the inductor and the switch.  Theoretically, it will operate with input voltage from 3V to 40V but I wouldn't guarantee its behaviour so I don't recommend using it.

[engineheat]

Thanks for the help, I'll build it and keep you updated on my progress.

[engineheat]

Yes that IC is a dedicated SMPS controller, not a uC/P.  It is a LM3478 and can be programmed with a resistor to run anywhere from 100KHz to 1MHz.  I chose that one because it had no internal switching FET and I needed to be able to use a switch of my own choice which could stand up to Vds(max)=200V.  I ran it at 100 KHz precisely to reduce inefficiencies caused by parasitic ringing, or switching times which were not super fast.  It worked out OK (I got efficiency about 90%) but the chip does have an irritating (and undocumented) idiosyncrasy.  If the input voltage drops below about 7 volts it goes "high impedance" and turns the switch on PERMANENTLY.  This conveniently smokes the inductor and the switch.  Theoretically, it will operate with input voltage from 3V to 40V but I wouldn't guarantee its behaviour so I don't recommend using it.

Hi, do you know if something like I'm building requires FCC certification if one day I want to commercialize the device? I believe for frequency above 9 or 10khz, you need FCC approval. I've done some investigating but still a bit confused which class my device falls. If I do commercialize it, it will be targeting consumers. The boost converter is the only high frequency component of the device. I wonder how much the process will cost roughly.

thanks

[colorado.rob]

Hi, do you know if something like I'm building requires FCC certification if one day I want to commercialize the device? I believe for frequency above 9 or 10khz, you need FCC approval. I've done some investigating but still a bit confused which class my device falls. If I do commercialize it, it will be targeting consumers. The boost converter is the only high frequency component of the device. I wonder how much the process will cost roughly.

You can hire an EE to do the board layout when the time comes.  I've hired a number of professionals, such as mechanical engineers, for contract work on ODesk.  If you already have the schematic, they can do the board layout, and provide advise on enclosures and shielding, work with your mechanical engineer on the enclosure design, etc.  Find a local compliance lab and talk to them.  They will walk you through the process and provide you budgetary guidelines.

[engineheat]

Guys, the TI TPS61021A chip is only available in the WSON package. Its size is only 2mm by 2mm. Is it even possible to manually solder this for my Manhattan build? If so, some good tips would be appreciated. I'm thinking masking tape...

[engineheat]

One thing I just realized is that the controller chip has a thermal pad on its bottom used for DGND.  This pad is also used to get heat out of the chip.  If you put the chip upside-down on the copper with the pad on the top then the DGND is readily available but you may still need to wick off some heat from there too.  Maybe a chunk of large-gauge copper wire soldered to it may do that, something like a 1-2" piece of 14 AWG house wire soldered to the pad at its middle and just sticks up in the air on both ends.  Normally the wide ground tracks of the PCB would do that.

Anyway, I am glad you decided to go ahead with this build.  I am very interested in anybody who is eager enough to stick their neck out in unfamiliar territory even a little bit in order to learn something.

I just realized how small this chip is. It's 2mm by 2mm. That would make it impossible to hand solder this point to point wouldn't it? I'm not even sure if I have wire leads thin enough.

If there are no good ways to solder this by hand, I might just go straight to PCB prototyping.