Researchers have developed a statistical method that could improve the prediction of semiconductor insulator lifetimes and overcome limitations in conventional reliability testing.
Researchers from TU Wien, in collaboration with IBM and the National University of Singapore teams, have developed a new statistical approach that could improve the estimation of the lifetimes of insulation materials utilised in semiconductors. According to a recently published paper in Nature Electronics, traditional reliability testing methods for innovative transistor insulators lead to incorrect results while the newly introduced approach allows the expected lifetime of electronic components to be estimated using practical measurement data.
The lifetime of transistor insulators depends on the insulating material used and the operating voltage. Current reliability testing techniques suppose that the thinner layer of insulator fails at the same voltage level as the thick one because it involves only the scaling of the breakdown voltage. But it was found that this assumption is invalid when applied to nanometre thin insulators because dielectric breakdown is a random, time-dependent phenomenon whose failure behaviour depends on the test conditions.
“For thick insulating layers, you can simply apply a voltage, keep increasing it and measure the voltage at which it breaks down,” says Prof. Tibor Grasser from the Institute for Microelectronics at TU Wien . “The voltage at which the material fails is then converted accordingly. Roughly speaking: If the insulating layer in the transistor is only one-tenth as thick, one assumes that it will break down at the same field strength and therefore withstand only one-tenth of the voltage.”
The method of performing lifetime tests for high accuracy currently involves exposing samples with varying surface areas to several fixed voltages, with the test taking a very long time before results are gathered statistically. To solve this problem, the group suggests that the voltage should be increased using several ramp rates, after which the analysis is carried out using a special statistical procedure. Their results show that at least three voltage ramp rates have to be considered in order to relate voltage acceleration with failure probability.
“This allows us to say, what maximum voltage can a certain number of components of a certain area withstand over a certain period of time? That tells us which maximum operating voltage may be used for these components,” says Grasser. “This makes it possible to compare different materials reliably—without having to fear discovering years later that the wrong materials were used, and that they reach the end of their lifetime sooner than predicted.”
