How High Can You Go

We recently characterized our high temperature 8-bit ADC.  While it was designed for 250C operation, we kept raising the temperature until we reached 320C.  We stopped there for a number of reasons, the first being that our oven could not go any higher.  It was rated to 350C, but we think that the test wires running out of the oven were providing a heat sink that lowered the temperature.  And we were concerned that our test setup or the package itself would fail.

All of this led to the question of how high can we go?

Ultimately, there is a temperature at which silicon becomes a conductor.  I recall from school that was 377C, but that was a long time ago, and I can't find information confirming that.

The digital logic was working well at 320C, and I suspect that it could go quite a bit higher, and perhaps up to 350C.  On the other hand, the transistor leakage was definitely affecting the analog circuitry, and if it was higher, it would affect the digital circuitry as well.

The fab has authorized baking bulk wafers at 310C for EEPROM reliability testing.  Somewhere above we may start to see threshold shifts in the silicon along with other process issues.

Our package should start coming apart around 300C.  The weak points are the solder used to hold on the lid, and the epoxy used to hold the die in place.  We did exceed this temperature in our testing, and we presume surface tension kept the package from falling apart.  This could be addressed with a gold die attach, and with a brazed on lid. 

The part we were working with has aluminum pads and gold wire bonding.  Over time, we should see corrosion between the gold and aluminum.  This can be addressed by using a gold - nickel - palladium coating on the bond pads.

Finally, our test board was rated for 300C.  It didn't appear to be damaged, but then the part was not at 320C for more than 15 minutes.  This can be addressed with a ceramic PC board.

So how high can we go?  The absolute limit will be set by the transistor leakage interfering with the circuit operation.  The exact temperature will depend on the design of the circuit, but will be in the range of 350C.

The answer also depends on how long we have to be there.  At 250C, the reliability goal is 10,000 hours, or just over a year.  Reliability halves with every 10C rise in temperature, and so the year of operation at 250C may be down to a few days at 320C, and to a few hours at 350C.  The exact failure could be either transistor parameter shifts, or bond pad corrosion.

Even though we can get to 350C, practical considerations will limit us to 310C.

We are going to start an experiment with our reliability monitor to see if semiconductor parameter shifts are occurring above 300C.  We will report the results in the future.

Date

2015-07-31