Machine Learning Based V-ramp VBD Predictive Model Using OCD-measured Fab Parameters for Early Detection of MOL Reliability Risk

Abstract

In this paper, we propose for the first time a breakdown voltage $(\mathrm{V}_{\text{BD}})$ prediction method using structural parameters measured in-process for early detection of reliability risks in Middle-Of-Line (MOL). $\boldsymbol{\mathrm{V}_{\text{BD}}}$ of the MOL is proportional to the distance of the Gate (PC) to Source/Drain-Contact (CA). Since PC to CA space can be calculated using MOL-related structural parameters at the early stage of the process, we created and validated models predicting V-ramp $\boldsymbol{\mathrm{V}_{\text{BD}}}$ using five fab parameters measured in-process by optical critical dimension scatterometry (OCD). And we compared three modeling methods. The first is the geometrical calculation model (GCM), the second is multiple-linear-regression (MLR) method, and the last is the Multi-Layer Perceptions (MLP) model based on the machine learning (ML). We found the highest predictive consistency $\boldsymbol{\mathrm{R}^{2}0.6}$ in ML method, and it is expected to contribute to the early prediction of MOL V-ramp $\mathrm{V}_{\text{BD}}$ through additional consistency improvements.