Application of artificial neural networks to the identification of weak electrical regions in large area MIM structures

Thumbnail Image
ExtendedESREF21_Final.pdf(864.24 KB)
Accepted Version
Muñoz-Gorriz, J.
Monaghan, Scott
Cherkaoui, Karim
Suñé, Jordi
Hurley, Paul K.
Miranda, Enrique
Journal Title
Journal ISSN
Volume Title
Elsevier Ltd.
Research Projects
Organizational Units
Journal Issue
Large area metal-insulator-metal (MIM) structures are prone to exhibit weak electrical regions when they are subjected to severe stress conditions. Although the root cause of this problem is hard to identify, it has been attributed to non-uniform oxide thickness, variable dielectric permittivity, correlated defect generation, and fringe effects in capacitors. In this paper, we explore the application of artificial neural networks (ANNs) to the spatial localization of such weak regions. To this end, HfO2-based MIM structures were electrically stressed with the objective of generating a large number of breakdown spots. These spots are statistically distributed both in size and location over the device area. Two-input/two-output ANNs with different number of neurons and hidden layers were assessed with the purpose of identifying the best and simplest option for detecting where the most severe damage occurs. The obtained results are compatible with previous studies based on spatial statistics techniques. The method can be applied to other systems that exhibit multiple localized failure events.
Neural networks , Perceptron , Spatial statistics , Dielectric breakdown , Reliability , MIM
Muñoz-Gorriz, J., Monaghan, S., Cherkaoui, K., Suñé, J., Hurley, P. K. and Miranda, E. (2021) 'Application of artificial neural networks to the identification of weak electrical regions in large area MIM structures', Microelectronics Reliability, 126, 114312 (7pp). doi: 10.1016/j.microrel.2021.114312
Link to publisher’s version