Switching Regulator design Questions

[jackinnj]

I am thinking of driving out to Dayton for the Hamvention -- haven't been in decades. I could load down my old MB with these boatanchors.

[N-Channel]

Sorry. no Can do. Got baby on the way in june/july. What's the call?

N8XO

[Claude Abraham]

The LM2679 is a great controller, featuring a built-in D-MOS FET device, rated at 5 amps output. I really like Natl Semi's "Simple Switcher" series, and have used them in the past. I may use it some time soon, depending on the application.
If you need more current, I would recommend a stand-alone (no internal FET) controller, with an external MOSFET. One such device would be the Texas Instruments (formerly Unitrode) UCC38C43. It is a current mode controller, as opposed to the LM2679, which is a voltage mode controller. When dealing with the large currents you mentioned, pulse by pulse current limiting is very desirable. Both the LM2679 and the UCC38C43 have it. The UCC38C43 current limit can be adjusted to a high degree of precision using a current shunt, and a high side current sense amplifier, such as the Tex Instr INA139, or an external resistor clamp network (see the TI application notes U-97, U-100A, etc.). The problem with the built in current limit of the LM2679, and other similar devices, is that it varies from 5.3 to 8.1 amps. I prefer tighter limits.
The external MOSFET, at these high current levels, should be N-channel, possibly several in parallel. To properly drive it, you'll need a high-side, or "bootstrap" driver such as International Rectifier's IR2117 or ST Micro's L6385, or a discrete circuit (TI recent application notes). I strongly recommend operating the inductor in the continuous conduction mode. Also, to keep losses low, I would use a low switching frequency (150 kHz, or lower). This will result in a physically large inductor.
I've designed several SMPS circuits in the last 5 years with these devices and have obtained good results.
I would not add another MOSFET to the LM2679 to obtain higher current. In order to drive the external FET (n-channel), you need a gate drive voltage higher than the input rail. The "switched output" pin is slightly less than the rail, and cannot drive the FET fully on. The LM2679's "boost capacitor" provides this elevated voltage for driving the internal n-channel FET.
As far as parallelling modules goes (stacking), if National Semi says that it is OK, I won't argue. To the best of my knowledge, stacking is best achieved with ***current-mode*** controllers, and the LM2679 is ***voltage-mode***. Stacking modules could result in one or more inductors saturating if the currents do not divide equally between modules. But seeing that the LM2679 has built in pulse-by-pulse current limiting, saturation should not take place as long as the inductors' saturation current rating is greater than the ***maximum*** value of the LM2679 current limit, namely 8.1 amps.
The downside to using the UCC38C43 current mode controller lies in the added complication. The error amplifier must be externally compensated, the current sensing requires filtering for noise immunity, and a leading edge current sense turn-off transistor is needed. Also, the inner control loop (current) must be slope-compensated for stability.
If you'd rather not deal with the added complication of the UCC38C43, externally compensating the error amplifier, the high-side MOSFET drive circuit, the current-sense noise filtering and leading-edge spike suppression transistor, then parallelling LM2679 modules may be the best route. Also, the UCC38C43 can only accept input voltages up to 18 V dc, so an additional bias supply would be needed to power it. The LM2679 can input up to 45 V dc. It looks like stacking several LM2679 modules is a very good option
There are other choices, as I've only suggested two I'm familiar with. Best regards.

[N-Channel]

Wow! Claude makes some awesome points, and he's right on the money with the IR2117 point. However, if you really a higher operating voltage for the 38C43, try using a plain old UC3843. It's the non-CMOS version of the 38C43, and it can operate from, I believe, 30 or 40V, but not too sure because it's been a long time since I've looked at the datasheets. The 3843 is available from Unitrode (now a part of TI), and ON semiconductor (formerly Motorola).

If you'd like to avoid the added complexity of a bootstrap drive for the N-Channel MOSFGET, then there is a possible solution. There is ONE P-channel that will do the job as a high-side switch: It's Harris' RFG60P05, housed in a TO-247 case, and capable of dissipating nearly 200W (Not that you'd need or want this much dissipation as a high-speed switch). Its BVdss is 50V, and its Id is 60 Amps, with a Rds(on) of 26milliohms. We have used MILLIONS of them as high-side switches in US Army military vehicle systems, with 0.00002% failure rate.

[jackinnj]

Originally posted by Claude Abraham
As far as parallelling modules goes (stacking), if National Semi says that it is OK, I won't argue. To the best of my knowledge, stacking is best achieved with ***current-mode*** controllers, and the LM2679 is ***voltage-mode***. Stacking modules could result in one or more inductors saturating if the currents do not divide equally between modules. But seeing that the LM2679 has built in pulse-by-pulse current limiting, saturation should not take place as long as the inductors' saturation current rating is greater than the ***maximum*** value of the LM2679 current limit, namely 8.1 amps.

I want to be clear that <em>stacking</em> was something National said you "could do", but they weren't endorsing it. It has been done. I think they would be happier avoiding potential problems with synch and running a high speed BFO as N-Channel pointed out.

[Claude Abraham]

Yes, "N-channel", the non-CMOS UC3843 works very well. I just like new parts. The 38C43 has other added features. The UC3843 is rated at 30 V max input. I still recommend an additional bias supply, since the IR2117 is rated at 25 V.
As far as P-FETs go, the latest parts from IR, ST Micro, Vishay, Infineon, etc. are quite impressive, and I've been using them (IRF6216, IRF5800). In this case P channel does not simplify things. Since the voltage exceeds 20 V, clamping circuitry is needed to protect the gate to source from punch through. To switch the P-FET quickly, large gate drive current (several amps) is needed. To minimize wasted power in the gate to source clamp. Zener or otherwise, low current is needed. Also, the UC3843, and similar controllers have a MOSFET gate drive of the n channel polarity. To use a P part would require an additional inverting driver (IR2118). The p part requires as much or more additional circuitry than the n part. What is the point in using the P-FET??? At voltage ratings of 20 V, p channel FETs make sense up to 7 amps or so. At 150 V rated parts, I recently calculated 3 amps as the useful limit for P-FETs. This is a 40 amp design. No debate at all. Use N channel, with its inherent 2.7 times better physics.
I agree with "jackinnj" that stacking, although viable, is not the best choice. At this power level, a push-pull or full bridge topology is the way to go. Best regards to all.

[fr0st]

Thanks for the offer jackinnj but shipping a 70 pound PSU from USA to australia sounds isn't my favourite of tasks
Otherwise I would have loved to take one off you
A very nice offer to say the least

Although everyones solutions are good most of the IC's mentioned are unavailable to me. I had to order from the other side of australia for a SG3525 controller and I work for an electronics store
In respect to my original idea... What does it take to fully turn on a mosfet? I've searched everyone for and can't find out
Say if I was really keen and made a seperate low current PSU at a higher voltage than the one I need @ 40amps so I can drive the fets at a higher voltage and then use the added Fets regulate the large PSU?
Say If I had my 40 amp rail @ 22v and a regulated rai lto run the LM2677 off @ 25v, would that give me enought head room when the 40 amp rail drops under load?
Thanks all

[fernando_g]

Originally posted by N-Channel
If you're going the National SimpleSwitcher route and want more current, try the LM2677. It puts out up to 5 Amps, and is synchronizable. Simple put a 300kHz clock (like a 555) on its sync pin, and you can papallel as many serctions as you want. I recently did a 24-12V DC-DC Converter paralleling 3 chips and clocking them from a 555, and I get uot nearly 12 Amps. Works great. I am using the surface-mount 12V version, LM2677S-12. Samples are available from National.

Best of Luck.

'73,

de N8XO

Dear N-channel (nice handle, BTW)

I understand the part bout the synchronization, but how do you match the current sharing between the devices? The thing is, there will be always a device with a slightly higher voltage than the rest, and that one will be essentially maxed out, while the others in parallel will not carry their fair share of current.

I rmember seeing a current share circuit trick for linear adjustable regulators (like the LM317). Did you try something similar?

[N-Channel]

Thanks for the compliment. It took me a long time to come up with a good one. I can't believe I fat-fingered that many typos in that last post! Anyway, I forgot to mention that I trimmed each section to within 1mV of each other before I tied the outputs together. Normally, you would have to put a 0.1- to 0.05W power resistor in series with the output to force current-sharing, but I was after efficiency, and more resistance did not seem appealing to me.

[fernando_g]

Originally posted by N-Channel
Anyway, I forgot to mention that I trimmed each section to within 1mV of each other before I tied the outputs together. Normally, you would have to put a 0.1- to 0.05W power resistor in series with the output to force current-sharing, but I was after efficiency, and more resistance did not seem appealing to me.

Godd point about the trimming!
And the PWB's intrinsic DC resistance will balance out any remaining offset.