What DC-DC topology should I look for?

[T3sl4co1l]

It's not a common topology, so you'll get a lot of oddball, special case things.  Or LT, they have everything... if you're willing to pay for it.

3579: I don't mind the hot plug feature, it's one of those quirky additions that LT has a lot of (again, oddball special case things, are one of their specialties), and I can't say I've needed that particular feature, but if you do, sure, it's handy having it all in one chip (plus support components).

And the 1370, fine I guess.  It loses points for me because of a peculiar set of errors:
1. "Switch Current Limit vs Duty Cycle", why is this truncated?  If it's because of slope compensation just added in, evidently they added a clamped sawtooth wave (see the ramp waveform and "Σ" (summer) in the 3579 block diagram -- this representation is a common sight in block diagrams of peak current mode controllers), so that it doesn't compensate anything at low duty.
2. The block diagram is very basic, and seems to have current sense go directly into the comparator, no slope compensation at all.  Is this in fact only good for DCM (discontinuous conduction mode)?
3. The paragraph on page 8,

Caution should be used when synchronizing above 700kHz
because at higher sync frequencies the amplitude of the
internal slope compensation used to prevent subharmonic
switching is reduced. This type of subharmonic
switching only occurs when the duty cycle of the switch
is above 50%. Higher inductor values will tend to
eliminate this problem.

seems to confirm the truncated-ramp hypothesis.  To be clear: it's an oft-repeated claim that subharmonic oscillation occurs over 50% duty, but this is only true when the voltage ratio is the same.  That is, a boost to 2*Vin, or SEPIC to Vout = Vin, etc.

What's really meant is, the necessary consequence of operation at that point: for the voltage ratio to be exactly 1, AND duty to be over 50%, the inductor current must be continuous (CCM).

And this is the fact that actually matters, current being left in the inductor from cycle to cycle.  That is, when the switch turns on the next time, the current will have decayed an unknown amount, and so the on-time (until the comparator ends that cycle) is also unknown.  It turns out, at first the duty cycle alternates every cycle (half subharmonic), then at a little higher load current, a cycle of 4, and so on; the number of states diverges rapidly (period doubling) until full chaos occurs.  In fact, if you plot the states (plotting a point for every pulse width seen, at a given load current), you get what's called a bifurcation diagram (as the number of states doubles every time it splits up like that).  The figure is identical to that of the logistic map -- it's actually an analog implementation of it.  (Cool, huh?)

Under chaotic operation, inductor current is always bounded, as is duty cycle, so it's not unsafe to operate, as such; but since exact operation is impossible to predict, there is higher output ripple, and usually audible hiss from the inductor.  Switching loss may also be increased (which if it causes overheating, could lead to failure).

And, ironically -- they suggest higher inductance, but this acts to reduce the ripple fraction, pushing the system closer to CCM, and subharmonic oscillation or chaos.  And, I assume the comment about "reduced slope compensation" at high frequencies, is just due to the truncated ramp being also delayed by internal circuitry, a delay which is relatively significant by that frequency.  That's fine.

So, that explains that situation.

I don't know what exactly they did in the design of this chip, but it's just... weird.  Not weird enough to be like "eww no", but weird enough I would keep looking for something better.

They also discuss tantalums and ceramics, which haven't been an issue for YEARS; ceramic and polymer won, and most chips support them by design.

Just how old is this datasheet, anyway?  1370 must be a pretty early part, if they're just numbering sequentially... ah there it is, very corner of the last page, 1998 it seems.

Which I guess agrees with the price: the D2PAK version is $18 in singles at Digi-Key, clearly they don't want you designing it in.  If you want to keep buying it to support a legacy product, sure, it's up to you how much you want to spend doing that...


What would I use?

Probably something like a TI Simple Switcher of comparable ratings, or a controller of whatever sort (probably something nicer than ye olde UC3843, I'd have to go shopping to see what), and just tack on the op-amp to invert the feedback (which is exactly what LT1370 does, they simply set V(FB) = -0.5 V(NFB) with an op-amp).  Indeed, I've done exactly that before.

Or even more random, probably the flashlight board I made, which indeed I've used for a Cuk converter before.  But that's not exactly helpful here.

Tim

[LoveLaika]

Thanks for your input. If only TI provided good simulation models for their products. Partly, that's why I was drawn to the LT ICs, as LTspice already has most of the models built in that actually work.


Since you mentioned the SIMPLE SWITCHER family from TI, I think it gave me some ideas. Initially, I was looking at a converter with an integrated switch, like the LM2588 even though it's kind of old. Other converters I found were those rectangular integrated modules that had strange footprints which I wanted to avoid. But then, looking at controllers (without the integrated switch), those have the advantage of allowing higher currents via the external switch. Thinking about it like that really helped me with search results.