Getting a High Speed Synchronous N-Channel MOSFET Driver working right

[TonyStewart]

It is very difficult to even copy a good design because you don't know all the physical properties of conductors and insulators. When you combine that with a lack of understanding what to expect and lack of attention to ESR, DCR, Q, SRF Z(X(f) ZC(f), and FET Vgs vs Vt to control RdsOn and stored energy vs load energy with minor effects from Ciss, Coss for efficiency and thermodynamic properties makes your chance of success even smaller but not zero. My point is you have a lot to learn befoe you can successfully design SMPS to pass step load ringing , ripple and efficiency tests without self-destruction or NOGO. I suggest an eval kit and scope after a good SMPS simulator. There are a lot of things to learn.

The wrong inductor and pulse f and no soft start even with the right values can cause trouble.

 

[Baluncore]

... makes your chance of success even smaller but not zero.

The paralysis of fear.

In any venture, all is lost if you do not take the next step.
I know that after a big step, I can always regain my balance, or jump back.

By definition, a well-designed circuit will work every time.

 

[TonyStewart]

I see things differently. If you aren't afraid to fail, go ahead . But learn the prerequisite theory of operation then the detailed acceptance criteria. You only have to meet these specs to have a perfect design. But if you don't understand the basics , how will one understand the complexities of control systems.

One of the world's best teachers now is my friend Christophe Basso who has many books on SMPS design and online help info. Formerly he was with ON Semi in France.

http://www.how2power.com/newsletters/2108/H2PToday2108_bookreview_DennisFeucht.pdf?NOREDIR=1
http://powersimtof.com/Spice.htm

There are also many books by others in [Mentor Note: links to illegal download websites deleted]

 

[Rive]

By definition, a well-designed circuit will work every time.

If it's on the limits and there are too many things depending on component and PCB/layout variables then it's not so sure, but prototypes and design iterations are just for that.
So (within a realistic approach) there is no need to be paralysed. If it does not work, then it'll give clues to learn from.

 

[TonyStewart]

IMHO, Some inaccurate comments have been posted so far.

If you have an IRFxxx FET with Vt=2 to 4V , they can work well at 5V if your samples are near the 2V threshold, but this is a bad recommendation to use Vgs=5V for IRF series.
When parts are near the 4 V threshold you need 10V to compare with the spec for maxRdsOn. So the rule of Thumb is 2.5 times the max threshold voltage for minimum gate drive to be reliable. (10V/4V=2.5x) However on the sub-threshold types or "logic level FETs", this Rule of Thumb relaxes to Vgs= 2x threshold for minimum gate drive.

Also the IRFZ series is an excellent choice for logic level and the IRFZ44N is a popular low RdsOn logic level FET that "can be an excellent choice". But keep in mind RdsON * Coss = Tau tends to be constant in the same family or technology and voltage rating ,so when you choose really low Ron then it becomes a critical tradeoff with self lower resonant frequencies and losses increasing with switching rate.

Half-Bridge Dual-N Theory of Operation
1. The main requirement to work for the half-bridge with dual Nch FETs is to get the low-side PWM working to create the boost voltage needed above Vbus to greater the gate voltage for the high-side driver. check if Vgs = >=2.5 times Vt (250 uA threshold)
2. The damping losses for every part affect operation and efficiency. This means RdsOn, DCR of choke, and ESR of all caps. Too much total loop R, it falls overdamping. Too little it fails from high Q resonance, overshoot, and instability. So these are critical choices.
3. All control loops prefer 1st order feedback. This means current feedback for current error correction and voltage feedback for voltage error correction. However, the LTC4449 does not have current feedback inputs and all examples use a smart chip ahead of the drivers to take the current feedback signals to control the PWM. Without this, your dynamic performance will suffer. like step load response. Although 2nd order feedback can work with compensation to take a partial derivative of voltage feedback, it is not as good without current feedback.
4. Dead time is critical to check, so as not to cause shoot-thru currents and overheating. Too much dead-time then the flyback power duration increases. This is affected by junction temp and reactive loads. Consider 1<t< 10% of switch cycle.

these are my highlights. Good luck.

My best recommendation is follow a TI Webench design and or clone the kit layout and BOM parts. SMPS design is not trivial.

The Basso design book online is OK to use says the author to me. He welcomes the user feedback more than the publisher's revenue. This is possibly the best theory and practical book on SMPS design to date.

Also a schematic is not a "layout" of physical parts like a PCB, a schematic is a logic diagram and is missing all the parasitic analog parameters like 0.2 to 1.2 nH/mm or 0.05 to 0.5 pF stray capacitance or mutual inductance or mohms/cm resistance or ESR or DCR , decoupling caps, spectral impedance, dynamic load impedance etc .

 

[techn0]

Update. With the new PCBs running the NCP3420 I was getting a better high gate wave form but things were still very off. Then after Checking to see if I had damaged some mosfets I was testing around with my ohm meter when I saw something strange.
New PCB

The problem that was present on both PCBs

Trace is not connected on output of low side mosfet. I was unable to see in software and raw pcb because of the silk screen. This directly caused the boost circuit to not function making the high side gate driver useless.

Turns out it works like I expected all along with both revisions now when the trace is connected.

I am using dual 5v regulator in rev 2 which I think works better. I was using a 9v regulator for the gates for a bit but kept burning the 3420 out because it was exceeding its specs.

 

[DaveE]

Turns out it works like I expected all along with both revisions now when the trace is connected.

At the high risk of sounding like a jerk...
Let's review post #3:

My best guess is that something isn't what you think it is. A wrong or shorted trace, a backwards diode, a 220pF cap instead of 220nF, etc. I would pause and go back and systematically verify everything, especially the stuff you think is correct. This circuit should just work.

My point is this, an essential step in troubleshooting, when you think "this doesn't make sense" is to always go back a few steps and verify your assumptions, carefully this time so you don't make the same assumptions again. The more odd something is, the more likely you are thinking about it incorrectly, then it's time to put down the scope probe and think systematically.

I've lost count long ago of how often I've seen the equivalent of your story, usually in my stuff. On several occasions I've asked other EEs to come help, just as a sanity check because they think differently than me. Often troubleshooting is quick, easy, and kind of haphazard. This is usually the best first attempt. But as soon as that doesn't work, you need to switch approaches to the most slow and logically correct process you can.

On the plus side, I'm convinced the only way to learn this is to waste a lot of time and effort troubleshooting a simple, but elusive, problem. Next time you get that feeling of "what the heck, this doesn't make sense at all" you'll remember to start over with much more care. This is "the school of hard knocks", and that's a very effective school.

 

[techn0]

I couldn't agree more.. I felt like I was checking everything but much like looking for something you lost. It is always found the last place you look. Trust me I was mad at myself more then anything.

 

[Windadct]

Your application is not the same as the example in the DS. The bootstrap needs get a voltage above the Upper switch

 

[techn0]

Im with DaveE on that comment. The design is pretty much a carbon copy of the DS.