IS Heat Sink Required?

[Winfield Hill]

Phil Allison wrote...

** The test was like this.

Input: 26 volts DC average with 2.2 volts p-p ripple at 50 Hz.
Load: 68 ohm, 10 watt WW resistor.
Device dissipation: 2.3 watts.

I get 2.43 watts.

Thermometer: K-type bead, coated in thermal grease, wedged
into device mounting hole.
CRO and DMM monitoring output voltage.
Device sitting on bench in free air.

Sample #1: Teledyne LM7815 CSP ( 8446 )
Sample #2. ST LM7815 CV ( 99244 )

Thermal shutdown ( TS ) indicated by sudden rise in output
ripple and DC voltage drop > 0.2 volt.

RESULTS:

Sample #1: TS at 145 C in 90 seconds, voltage then dropped
to 6.7 volts.

Sample #2: TS at 151 C in 100 seconds, voltage then dropped
to 0.33 volts.

If we take the thermal resistance to be 4C/W, then the shutdown
Tj was about 10 degrees higher, or 155 and 161C resp. We can
assume a bit of thermal lag between the case temp and your TC
reading, a few more degrees may be added, yielding an estimated
Tj = 160 and 165C for these two samples. Nice. This confirms
Spehro's guess of "160 to 170°C" three days ago. It also means
one shouldn't use a Tj = 150C maximum for designs, as Anthony
Fremont was asserting, because that's much too close to the
unspecified shutdown temp. The 78xx datasheet's 125C maximum
"operating" temperature spec includes a safety margin against
shutdown, along with prolonging chip life.

Ah, Phil, old chap, while you're at it, could you measure up a
few more parts to build up some statistics for us?

You could also lower the input voltage, along with changing the
load resistor to maintain the same dissipation, to test your
suggestion that shutdown Tj changes with input-output voltage
difference. A glance at NSC's LM7815 schematic shows that the
shutdown Tj should increase for input-output voltages below
6.5V, when Q11 to D1 becomes saturated. Perhaps to allow more
dissipation in the safer regions of the pass-transistor's SOA.

In fact, it appears for some low value of input-output voltage,
perhaps below 3.5V, there's effectively no overtemp shutdown!,
and an NSC LM7815 is only protected by its current limit. Even
though this drops as Q14's temperature increases, it may still
be over 1.5A, according to the datasheet curves, which *cough*
would be over 5 watts for 3.5V, which means (I'm speculating)
that using a ripple-free 18.5V source, a 10-ohm load, and *no*
heatsink, you could force an LM7815 die into excessively-high
temperatures, certainly over 200°C, and eventually destroy it.

NSC's LM340, and most other manufacturer's '7815's, including
ST's L7800, TI's uA7815, On Semi's MC7815, LinFinity's SG7815,
JRC's NJM7815, etc., use a different overtemp shutdown scheme,
which is not affected by the input-output voltage differential.
Ditto for the original Fairchild's uA7815 (1978 datasheet).
(I couldn't find a copy of Teledyne's '7815 datasheet here.)

Hey, it's even possible NSC no longer ships the LM7815 circuit
we see in their datasheet. I note their combined LM340-
LM78xx datasheet shows the alternate scheme, like everyone else.

These could be rather interesting tests.

 

[Winfield Hill]

Jim Thompson wrote...

Two ways Tj is measured accurately:

(1) Run device at operating conditions, momentarily reduce current
to where series resistance doesn't alter vBE, and measure the vBE.

(2) On ASIC's I often include a separate junction just for measuring
chip temperature.

A third way, employed when Vbe isn't available, etc., is to remove
power from the device, and pass a small current, say 10mA, in the
reverse direction across pins where a substrate diode, etc., may
be accessed. A two-current ratio test should be even better. The
test can be performed quickly, say 10us, and the part immediately
reconnected for normal operation. If desired, the subject diode
can be calibrated beforehand in an oven. Motorola shows complex
circuits to do this for power BJTs in some of their old app notes.

 

[Phil Allison]

"Winfield Hill"

I get 2.43 watts.


If we take the thermal resistance to be 4C/W, then the shutdown
Tj was about 10 degrees higher, or 155 and 161C resp. We can
assume a bit of thermal lag between the case temp and your TC
reading, a few more degrees may be added, yielding an estimated
Tj = 160 and 165C for these two samples. Nice. This confirms
Spehro's guess of "160 to 170°C" three days ago. It also means
one shouldn't use a Tj = 150C maximum for designs, as Anthony
Fremont was asserting, because that's much too close to the
unspecified shutdown temp. The 78xx datasheet's 125C maximum
"operating" temperature spec includes a safety margin against
shutdown, along with prolonging chip life.

** The other *inescapable* conclusion is that you cannot expect a LM7815
etc to dissipate anything like 2 watts with no heatsink or fan cooling in a
REAL application.

Free air conditions on the bench do not typify actual use conditions, let
alone any sort of worst case scenario where other heat sources are
contributing to the local ambient, convection is restricted and heat
sensitive parts like electro caps are sited adjacent to the regulator IC.

If one also allows for a high room ambient ( ie 40C ) conditions and the
possibility if the unit being in direct summer sunlight - then even the
suggested figure of 0.6 watts may be too high.





.......... Phil

 

[Spehro Pefhany]

If one also allows for a high room ambient ( ie 40C ) conditions and the
possibility if the unit being in direct summer sunlight - then even the
suggested figure of 0.6 watts may be too high.

Exactly, it's just a rule of thumb. If it looks like it might be
close, then a more detailed evaluation may be in order.


Best regards,
Spehro Pefhany

 

[Jim Thompson]

Jim Thompson wrote...

A third way, employed when Vbe isn't available, etc., is to remove
power from the device, and pass a small current, say 10mA, in the
reverse direction across pins where a substrate diode, etc., may
be accessed. A two-current ratio test should be even better. The
test can be performed quickly, say 10us, and the part immediately
reconnected for normal operation. If desired, the subject diode
can be calibrated beforehand in an oven. Motorola shows complex
circuits to do this for power BJTs in some of their old app notes.

That only works for discrete BJT's that have a flyback diode
structure, since BJT's are classically 3-layer, as opposed to IC's
which have "substrates" for isolation.

"A two-current ratio test" is optimum... usually you can do that
without ANY calibration.

...Jim Thompson

 

[Winfield Hill]

Jim Thompson wrote...

That only works for discrete BJT's that have a flyback diode
structure, since BJT's are classically 3-layer, as opposed to
IC's which have "substrates" for isolation.

It's not necessary to invoke a substrate, for example I'm sure
the base-collector diode may be used.

"A two-current ratio test" is optimum... usually you can do that
without ANY calibration.

Yes.

 

[John Larkin]

That only works for discrete BJT's that have a flyback diode
structure, since BJT's are classically 3-layer, as opposed to IC's
which have "substrates" for isolation.

"A two-current ratio test" is optimum... usually you can do that
without ANY calibration.

...Jim Thompson

Any available junction will work. We do this on mosfets a lot, but any
substrate or esd diode can be calibrated.

John

 

[Anthony Fremont]

If we take the thermal resistance to be 4C/W, then the shutdown
Tj was about 10 degrees higher, or 155 and 161C resp. We can
assume a bit of thermal lag between the case temp and your TC
reading, a few more degrees may be added, yielding an estimated
Tj = 160 and 165C for these two samples. Nice. This confirms
Spehro's guess of "160 to 170°C" three days ago. It also means
one shouldn't use a Tj = 150C maximum for designs, as Anthony
Fremont was asserting, because that's much too close to the

Whoa there, let's hang on a second here. I never meant that value
should be used as a design guideline, I just said that's what the data
sheet had as a max Tj. Even I wouldn't attempt to operate parts at
their absolute maximum ratings. If I was trying to design something
for production, I'd probably assume that "ambient" meant the trunk of a
car in houston in August. When I'm tinkering "ambient" is 21C with a
good breeze. ;-)

 

[Chris Jones]

Any available junction will work. We do this on mosfets a lot, but any
substrate or esd diode can be calibrated.

John

Analog Devices makes some ICs which measure the forward voltage at three
different currents, and then they compensate for the ohmic resistance in
series with the diode. (ADT7461)

I'd like to build a thermometer readout using one of those to measure the
temperature of the chips I'm tormenting. It would be nice if it would work
with one side of the diode grounded but I'm not sure if it will or not. If
all else fails I might have to make a floating supply for the ADT7461 IC so
that I can ground either end of the diode under test, and then I can use
any old ESD diode inside the chip being punished.

Chris

 

[John Larkin]

Analog Devices makes some ICs which measure the forward voltage at three
different currents, and then they compensate for the ohmic resistance in
series with the diode. (ADT7461)

I'd like to build a thermometer readout using one of those to measure the
temperature of the chips I'm tormenting. It would be nice if it would work
with one side of the diode grounded but I'm not sure if it will or not. If
all else fails I might have to make a floating supply for the ADT7461 IC so
that I can ground either end of the diode under test, and then I can use
any old ESD diode inside the chip being punished.

Chris

We normally just calibrate the test junction at a few points, by
heating the chip with power off, and graph temp versus junction drop.
Then in operation, we drop the operating power, bias up the test
junction for, say, a few hundred us, and use an oscilloscope to
accurately measure the diode drop. You can watch a junction cool off,
too, and observe the complex thermal time constant. In that case, it's
easier to use the simple junction drop curve rather than the more
complex multiple-current technique.


John