Dear Raghunathan,

'Burn-in' means, that something is done at high temperature, so high, that it almost 'burns'.
Theoretical background is, that there are some physical processes, where the reaction rate is dependent on temperature: Physical processes can be accelerated by increasing temperature. Typical processes are failuring processes and drift processes. So, e.g., silicon chips show accelerated aging when crystal temperature is increased.

In the Arrhenius Model this acceleration factor 'F' can be estimated, if normal temperature 'Tn', test temperature 'Tt' and thermal activation energy 'Ea' for the process which causes aging are given:

F = exp [Ea / K x (1/Tn - 1/Tt)]

Tn and Tt are absolute temperatures, K is Boltzmann constant.

An example:
With Ea = 0.7eV, K = 8.63 x 10**-5 eV/K, Tn = 273K + 25K = 298K, Tt = 273K + 125K = 398K F = 933 results. So, this process 'runs' 933 times faster at T = 398K than at T = 298K.

The idea of 'burn-in' is now to skip over the first several hundred hours, where 'infant mortality' of device occurs, while acceleration factor F helps to achieve this in a much shorter time period: Introduce a burn-in at about 125°C for about 100 hours and if the device is not damaged by this, it can be assumed, that it will not suffer any more from 'infant mortality'.

Correct 'burn-in' is of course much more difficult than stated above, when you go into detail. Often, the exact activation energies and parameters like 'infant mortality', etc. are not known at the beginning of fabrication process. Burn-in makes only sense, if there's already statistical material available about according device: Burn-in must be so demanding, that infant mortality is surely 'survived', but not so demanding, that each device is predamaged after the burn-in phase...

It's interesting to note, that not only 'infant mortaility' is the subject of burn-in, but also 'initial drift'. Especially analog devices show a certain increased drift performance right at the start of lifetime, so called 'initial drift'. Many drift processes need a certain time to stabilize. After the initial drift period device shows a much lower drift. This is e.g. interesting for a designer of analog circuits, where a low offset voltage is crucial: As long as the device suffers from increased 'initial drift', circuitry must all the time be recalibrated.

Here 'burn-in' also helps. 'Initial drift' time period can also be accelerated by burn-in. Introduce a burn-in at about 125°C for about 200 hours and afterwards device will show a much lower drift. Circuitry need not so often to be recalibrated or the need for recalibration can even be omitted.

In the following picture, which is from an older databook from ANALOG DEVICES, you can see long term drift performance of OP-27 at 125°:



By the way, each LM117 voltage regulator from NATIONAL SEMICONDUCTORS has passed a 100% Burn-in!

What causes 'infant mortality' or 'initial drift'?
Think of the movement of mobile charges on top of or within the oxide-protective die surface, which are present directly after manufacturing. As these charges accumulate, they can create leakage paths and other unwanted parasitic action. Accumulation of these charges results in a rise of potential which causes the decrease and finally the stop of further charge movement. Such a process can cause 'infant mortality' or 'initial drift'...

It's rather difficult to choose the right temperature and duration of burn-in process for complexer systems. In most cases temperature cannot be increased so high, that there's a relevant acceleration factor introduced. Then, even test periods of several hundreds of hours are rather useless. It's then better to distinguish and to introduce different burn-in procedures for the different parts.

Bye,
Kai