An amp that might not need biasing?

OK, not really, but when a friend of mine at ON send me this it was relatively exciting:

http://www.onsemi.com/pub/Collateral/AND8196-D.PDF

Basically, they put diodes on the die with the BJTs and brought them out to seperate leads, so that they can be used for realtime temperature tracking. The app note shows an amp that more or less self-biases.

This seems really cool, but I have my reservations. It seems to me that any high fidelity design with this scheme will still require some bias tweaking. They imply that in the app note in the last paragraph as well.

The thing that I thought of first was active biasing. Think of it as anti-lock brakes for your amp. The diode will give a realtime sense output for die temp, which is inexorably related to dissipation and Ids. It could allow realtime foldback on a per-device basis, or realtime current sharing adjustments. That leads me to think of killer apps in 5-9s reliability places like touring amplifiers (as well as off-topic apps like motion control).

So the topic for the drawing board:
You've now got realtime information from the die of your big beefy audio BJT. What do you do with it?

-dave
edited for typos and punctuation

Maybe not an audio application, or maybe. A big beefy linear regulator can avoid thermal runaway by sensing the temperature of the pass transistor die and shut down before secondary breakdown kills the device. I would have loved this about ten years ago when I was designing linear regs that beefy.

Many newer servo drives are switchmode so the transistors don't get hot, and are usually power MOSFETs or IGBTs, but some of the linear servo drives might find it useful.

> put diodes on the die with the FETs

BJTs. Not FETs. How quaint.

This is an idea whose time came 20 years ago. I have no idea why this did not become a common part a LONG time ago. It is so obvious.

The datasheet and app-note give no clue of the relationship between diode current and transistor current. It is simply impossible to design from the data given. Yes, try-and-see, but over the wide range of loads and temps a power amp will see, that does not give a heap of confidence.

The reference implementation is strange. Several odd (to me) details in the pre-driver, but nevermind that. They take 6 reference diodes as bias, but wrap them around three stages, two of which do not work at output-device heat. (Maybe this is why they don't give relative diode-transistor current data: in real life you have other transistors in the pile.) If somehow the bias were perfect at idle (again, can't predict from the data), then when the heatsink gets hot under load it will be underbiased. Or if one of those driver devices has an unusually low Vbe, the idle current will be too high AND there is no way to trim it back.

> It could allow realtime foldback on a per-device basis

Yes, if you want to add all those foldback-sensors. And note that audio amps classically die of second-breakdown or hypersonics, not really melt-down.

> You've now got realtime information from the die of your big beefy audio FET.

BJT.

> What do you do with it?

Any honorable audio-amp designer stays with what works: LOTS of beefy output devices and non-minimal emitter resistors. Especially when they are under $3 a pop, and a lot more SOA than the trannies I used to buy at $14/each.

The bias must track the drivers as well as the outputs. Even if they are on the same sink, they don't track perfectly. The on-die diode is handy; and in a big amp it would be affordable to use a pair of these 200W transistors as the next-to-last stage, so it could diode-track.

As for current-sharing: I like the dumb inefficient idea of big emitter resistors so it MUST balance, and lots of devices so slight unbalance does not overstress anything. Inefficient? Not compared to an amp that has died.

> avoid thermal runaway by sensing the temperature of the pass transistor die and shut down before secondary breakdown kills the device.

Second breakdown is a localized problem that won't be sensed by a die-average sensor.

> avoid thermal runaway by sensing the temperature of the pass transistor die and shut down before secondary breakdown kills the device.

Second breakdown is a localized problem that won't be sensed by a die-average sensor.

Click to expand...

Related stuff: Nationals SPiKe protection scheme. The original Widlar IEEE-article is an interesting read, but I don't know of an available online version.
So this one:
Audio Amplifiers Utilizing: SPiKe® Protection (pdf 1.2MB)
http://www.national.com/an/AN/AN-898.pdf
(enjoy fig 6 & 7 ! :wink: it sounds real funny but keeps your amp alive).

The NSC LM3875 uses this and there are Philips ICs as well that use related (non-infringing) versions of the approach.

by a die-average sensor

Click to expand...

It can be fast, like bringing out a separate emitter in the big BJT and using multiple sensing devices. The trick is indeed to make this non-averaging but instantaneous.

Bye,

Peter

FWIW, the Widlar-article is here:

Dynamic Safe-Area Protection for Power Transistors Employs Peak-Temperature Limiting
ROBERT J. WIDLAR AND MINEO YAMATAKE
(IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. SC-22, NO. 1, FEBRUARY 1987)

http://home.hetnet.nl/~chickennerdpig/FILES/0077widl.pdf

Bye,

Peter

high fidelity design with this scheme

Click to expand...

Hmm, fig 3 on page 3 of the pdf shows that for two of the three comparisons the THD actually gets worse at the left side of the graph... :shock:
It's only the load-situation that improves there - well, that'd be the most desirable case. :wink:

[quote author="clintrubber"]It's only the load-situation that improves there - well, that'd be the most desirable case. :wink:[/quote]

Well, since most of the amps I use spend the majority of their time at less than 50% power, it's vaguely attractive to me.

Runaway wouldn't be entirely controlled, but the diodes are aparently co-located near the emitter (IIRC) side fo the features, which would help track hotspots a lot better than if it was sitting over near the base bond pad.

There is obviously a lot of biasing stuff that's not show in the app note schematic. It struck me as either working (and obtuse and obscure), or incomplete. I think that the thermal tracking, since it is on-die, would be much closer to instantaneous than a device sharing the heatsink. The reduction of the thermal time constant surely provides an advanatge for the bias adjustment, no?

These parts are brand new. I can't get samples from my guy inside, which tells me that they just finalized the last Si spin. "Product Preview" status says a lot. I doubt we'll see anything in volume until Q2/Q3. Knowing ON, there will be a whole host more data on the relative tracking in the future. They don't even mention the diode i the prelim datasheet.

-dave