TPS54A20EVM evaluation - part 3

This is part three of evaluating a TPS54A20 on an evaluation board. The TPS54A20 is a buck converter with an

interesting topology - it has two phases merged with a switched-capacitor. This isn't a formal road test (I

was given the board by Jan Cumps, nice person that he is) and I'm just doing what interests me and blogging about

it. Disclaimer: some of what I do here is quite sloppy and, to be fair to TI, it shouldn't be taken as a

proper review.

This time it's transient output testing - what happens to the output voltage when the load changes suddenly.

In case it's not clear why anything happens, imagine a situation when a load of 10A is suddenly demanded from

the output. Initially, the only components that can supply the required current are the capacitors on the

output side. They are ceramic with low ESR and low inductance and will supply this kind of current, but as

soon as they do they start discharging and the voltage falls rapidly. The controller will then sense the fall

and act to bring the voltage back up, but that all takes a bit of time. This test isn't so much about what

happens in a real system, instead we're looking at how the control system responds - how stable is it and what

the damping is like, and questions like that.

My original idea for this was to switch my resistor load using a MOSFET, but then I realised that the board

includes a circuit to do this test. So, I've decided it would be interesting to do both and compare the

results.

This is the circuit down in the corner of the board:

and here's the circuit diagram [copyright TI, I hope it constitutes fair use]:

A major advantage of the inbuilt circuit is that (presumably) it was what they used when they were writing the

datasheet, so I'll have a point of comparison - my results should be roughly similar to theirs, except we

would hope that they have considerably better test equipment than I have. A second difference compared with my resistor

load, is that it should have lower inductance - the loops are much tighter and the planes naturally have low

inductance.

Curiously, the documentation doesn't tell you how to use the circuit, but it's fairly obvious I suppose. The

51R resistor at the gate is intended to take the signal from a waveform generator and the two 20mOhm resistors

in parallel are for current sensing - the combined 10mOhm giving 10mV per amp. The rest of the load resistance

comes from the MOSFET [it's just a matter of getting the Vgs right - the datasheet suggested 2.7V for 10A, but

I was seeing something a bit higher, around 3V]. I was slightly nervous that my ultra-cheap waveform generator

wouldn't give a precise enough output, but it was ok - rather than watch the generator waveform, I looked at

the current waveform and soon got used to bringing the current up to the level I wanted. The only other thing

to say is to make sure you have a reasonably long off period - I ran it 10:1 [a limitation of my poor signal

generator] and it seemed quite happy - if you don't, you'll exceed the dissipation limits of the MOSFET and

the resistors.

Here's what happens with a step to 10A. The blue trace is the current (100mV=10A, bandwidth limited to 20MHz),

the yellow trace is the output voltage, ac coupled, at the full bandwidth of the scope (200MHz). The step up

to 10A looks fine, the step down less well controlled.

For comparison, here's the trace that TI have in their datasheet:

Curiously, my board seems better on the step up, but slightly worse coming down.

Here, in more detail, is the step up, first the accumulated trace and then a single trace. [The accumulated

trace shows the overall form nicely and the spread as the waveforms vary, the single one lets us see the

switching frequency that makes up the bulk of the noise.]

Here's the step down, which is less assured [again, accumulated first].

It takes quite a long period before it settles at the reference level, something that doesn't really show in

the trace on the datasheet. And there's quite a lot of variation at the start [where the accumulated trace broadens out].

Now for my transient test circuit. Here's the circuit diagram [yes, I know about these new-fangled CAD systems, but it's still quicker to draw it with a pen]:

The choice of MOSFET was just what I had ready to hand. It would have been better to use one with a low Vgs threshold voltage - my generator

can only just get to the voltage this one needs to switch driving into 47ohms - but then I'd have had to search around for one. It needs to have a

low rDS on value, obviously, otherwise it will add too much to the resistor values. [Confession - I say obviously, but I had an IRF510 on a piece

of board, tried it, and then puzzled why it didn't work and would only go up to 2A or so. When I looked at the datasheet I realised it had an rDS-on of 0.56 ohms. How stupid is that?]

Here it is built on a piece of pcb board underneath the resistors (that's a good photo, some of the bits are in focus!). I've kept all the

wiring as short as possible to minimise inductance, though it will be a lot worse than TI's circuit.

Because the MOSFET has an 'on' resistance of a bit under 20mohm, the high current will be about 9A. The low current will be about 2A. Here are the waveforms.

The yellow trace is the voltage output. The blue is the generator waveform that's switching the MOSFET - it isn't showing us the current.

If you compare this to the measurement from the on-board circuit, it's not too much different in the broad view. There's a bit more difference in the

detailed views closer to the edges where the increased inductance will be having more of an effect. There's a bit more spread, but the shape still

gives us a good idea of how the control loop behaves itself.

I think it shows you can do a reasonable amount of testing like this with very simple equipment if you are sensible (and careful about how you interpret the results).