IEC 62374-2007 pdf Semiconductor devices – Time dependent dielectric breakdown (TDDB) test for gate dielectric films

5.2 Pre-test The pre-test is performed for identifying initial failed samples. The gate current is measured at the applied use voltage. If the measured current is larger than the defined criterion, then that sample is rejected as an initial failed sample. When obtaining the defective distribution is necessary, the CVS test without pre-test may be effective. In this case the pre-test can be omitted. 5.3 Test conditions 5.3.1 General The following test condition is recommended for the TDDB test. The sample size should be selected to provide the necessary confidence level for the application. 5.3.2 Electric field V stress shall be decided by a trial test to get the TDDB lifetime data in a reasonable time. It is preferable to select at least 3 electric fields for estimating the field acceleration factor. 5.3.3 Temperature It is preferable to select at least 3 temperatures. The use-junction temperature should be in the test temperature range for estimating the temperature acceleration factor (activation energy).
6 Lifetime estimation 6.1 General The method to get the temperature and voltage acceleration factor is explained in this section. 6.2 Acceleration model 6.2.1 General E ox model and 1 /E ox model are widely used models. Both models usually contain the temperature acceleration model (Arrhenius model). To contrast the two models, the E ox model gives a more conservative estimation of lifetime than the 1 /E ox model gives. The 1 /E ox model should not be used for very thin oxide under 5 nm as an acceleration model. The Power-law model or Anode hole injection model (AHI- model) is advocated as an acceleration model for very thin oxides. The appropriate lifetime model should be used depending on the thickness ( t ox ). • In the case of t ox ≥ 5 nm, the E ox model should be used unless otherwise specified or the 1 /E ox model provides a better fit of the data. • In the case of 2 nm < t ox < 5 nm, the V g -model should be used unless otherwise specified or the Power-law model or AHI-model provides a better fit of the data. • In the case of t ox ≤ 2 nm, the Power-law model should be used unless otherwise specified or AHI-model provides a better fit of the data.