Oxide Reliability

Last year, Tekmos introduced many 68HC05 and 68HC11 microcontrollers.  Most of these devices contained small amounts of EEPROM.  EEPROM is a non-volatile memory (NVM) that requires a special process to make.   And as with all NVMs, it is subject to two unique failure mechanisms that must be tested and verified.

The write and subsequent erase of a NVM stresses the oxide used to create the NVM.  After repeated write / erase cycles, the oxide will fail.  The minimum number of write / erase cycles is specified at 10,000 cycles at 85C.  The actual number of cycles you can do before failure is heavily dependent on the temperature and on the fab processing.  Checking the number of cycles is a destructive test, so it can only be done on a sample of parts.  We use oxide testing as part of our ongoing reliability monitors.

We have customers that want to use our parts at 175C.  This requires us to characterize the parts up to 200C.  I should note that what appears to be a more than adequate number of cycles at 85C is greatly reduced at the elevated temperatures.

The second failure mechanism is data retention, or charge storage.  We specify that a memory will remember for 10 years at 85C.  We guarantee that by writing a pattern into the processors, and then baking the wafers at 260C for 24 hours.  After the bake, we re-probe the wafers and insure that the pattern is still there.  The main failure mechanism for data retention is traces of poly left on the wafer after the poly etch.  These provide conductive paths that allow charge to be leaked from one bit to an adjacent bit. 

We verify the data retention by running parts at extended temperatures for long periods of time.  Like the oxide failures, the charge storage is a function of temperature, and so we evaluate it at higher temperatures. 

These tests are incorporated into our ongoing reliability monitors for the Tekmos products.