Memory Distortion? and some new beginnings.

[Onvinyl]

... and here is a closeup of a square wave @20kHz...

yes, it's the pure input pair without anything else causing this...
Rüdiger

[ionomolo]

Brilliant thread. The most innovative and constructive i've read so far.

Has anybody tried with mosfets instead of bjt's?

These can be biased at zero-tempco currents. If the amplifier was in positive tempco we should get unsymmetrical distortion, while at zero tempco thermal distortion should be lower and more symmetrical.

Since i don't expect other parameters from the mosfet to drift much at this point, a sudden decrease in distortion or improovement in quality would give a good clue.

Anybody thinks from memory on a small/medium power mosfet to do the job?

[VHF man]

Originally posted by Christer


Of course.

And by posting it you spolied the fun for everyone else.

The answer is BLUE.

Because global warming has melted the sea ice at the magnetic pole so the poor polar bear is actually swimming in near freezing water.

[tankcircuitnoise]

My nose was severely rubbed into bipolar thermal runaway
(a form of extreme distortion) when debugging silicon for
a CAN transceiver IC. At normal VDD (< 5v), the common-emitter
PNP pulldown was robust. Part of IC certification require surviving
Maximum Automobile Battery voltage of 40volts. In every test,
as output was forced from 5volts upward toward 40volts,
the PNP output device self-destructed. Of 5 output transistor
stripes on the IC, the middle stripe showed a burned spot.
The overseas silicon consultant designer balked at considering thermal
runaway as the cause of the failure, so some math was brought
to bear (you may enjoy this).

Assuming bipolars have safe-operating-area failures,
we will compute the failure point (an extreme form of distortion).
This is a case of positive feedback; we'll find the set of
parameters where Gain = 1.000, and we'll know to avoid
operating anywhere near that set of values.

Math:

delta_Temp_result=delta_Temp_in*delta_energy*therm al_resist/watt

By setting delta_Temp_result == delta_Temp_in,
we have the case of Gain = 1.000 as desired.

Now we need delta_energy/dT = delta_Ie/dT * ( Vce * Ie )

delta_Ie/dT = (2milliVolts/1degC) * (10% delta_Ie/4milliVolt delta_Vbe)
and delta_energy = delta_Ie * OperatingPointPower

Now some equation combining:

dT = dT*(2mv/1degC)*(10% delta_Ie/4mv)*(Vce * Ie/100%)*Rthermal
or
dT = dT * 1/20 * Vce * Ie * Rthermal

The current per PNP stripe was 6milliAmps;
I assumed 1,000degC/watt Rthermal, because I had no other value (yet).

dT = dt * 1/20 * 0.006 * Vce * Rthermal
and setting up for Gain = 1.000

1 = 1 * 1/20 * 0.006 * Vce * Rthermal
or
1 = 0.0003 * Vce * Rthermal
we find

3,333 / Rthermal = Vce_runaway

For Rthermal (my assumption) of 1,000 degC/watt,
the Vce needed to cause Gain = 1.000 is 3.3 volts.

However, the measured failure voltage was significantly greater.

Discussing this with an inhouse IC-process-development leader,
I was soon handed an Italian paper guiding me to use, for
the 4micron by 100 micron emitter stripe of the vertical PNP,
a new value for Rthermal of ---- 200 degreeC/watt.

Substituting that, we have 3,333 / 200 = 16.5 volts.

And the consistent Vce of failure, in lab tests, was 15.3 volts.

The team was very happy. The design consultant quickly reversed
their stand about "no. its not bipolar thermal runaway". Some poly
resistors were added, and the CAN Transceiver IC was released.

What is our lesson? If Ie is uncontrolled/unlimited, then bipolar
devices can be operated Vce & Ie values providing ThermalGain = 1.00
and bad things happen to good transistors.

Please note there is no lower limit on the positive feedback.
Any combination of Vce/Ie/Rthermal is + Feedback.
We always get that distortion.

Now consider a bipolar differential pair.
The maximum power is limited because the tail current is limited.
Does that mean there is no thermal distortion?

[tankcircuitnoise]

These SPICE+thermal resistance
seem to use a FIXED thermal resistance,
whereas as stated by others
the correct thermal model is a distributed (3-D at worst, 1-D in some cases)
R*C*R*C*R*C circuit.