Transfer Learning-Based Artificial Intelligence-Integrated Physical Modeling to Enable Failure Analysis for 3 Nanometer and Smaller Silicon-Based CMOS Transistors

Abstract

Integral to the success of the semiconductor industry in keeping up with Moore's law is the importance of failure analysis (FA). Accurate and fast FA is vital in ensuring yield, reliability, and rapid production in the semiconductor industry. However, locating defects among tens of billions of transistors packed in the tiny modern microchip is not a trivial task. Not only the process technology has to achieve such high integration of devices evolved to become astoundingly sophisticated but also debugging for defects in these chips has become remarkably complex. With electrical nanoprobing, we show how artificial intelligence-integrated physical modeling can be effective in finding difficult proverbial needle-in-a-haystack defects based on the electrical responses of the devices. Moreover, the information learned on current devices can be transferred to the latest transistor technologies, enabling machine learning-based defect sleuths of the future. Notably, we achieved a defect region classification accuracy of 99.5% on well-studied fin nanoscale field-effect transistors (FET) using a defect dataset based on an experimentally calibrated TCAD digital twin model and an adaptive boundary refinement technique. With transfer learning, a defect region classification accuracy of 99.58% is also achieved on the next-generation gate-allaround FETs, overcoming the lack of crucial datasets and optimized labels to guide and accelerate the production process of emerging devices. The proposed technique is expected to be the next level of defect identification, an important stepping stone to accelerate the production process for advanced technology nodes beyond 3 nm.

Using Data From Similar Systems for Data-Driven Condition Diagnosis and Prognosis of Engineering Systems: A Review and an Outline of Future Research Challenges

“…Di et al [127] presented an ensemble-based approach using SAEs. Pan et al [128] designed an ensemble of random forests to transfer knowledge to the next transistor generation. Gribbestad et al [129] applied transfer learning to feed forward neural networks, LSTMs, and CNNs to predict the RUL of marine air compressors.…”

“…In [49], knowledge was transferred from circuit breaker simulations to a real-world experimental domain. Pan et al [128] transferred knowledge between different transistor generations. In detail, the source domain comprises data from fin field-effect transistors and the target domain from gate-allaround field-effect transistors.…”

Optical wafer defect inspection at the 10 nm technology node and beyond

“…Cheon et al used a single CNN model to extract effective features for identifying defect classes that were not seen using conventional automatic defect classification systems based on SEM images [196]. More literature on deep learning-based defect inspection can be found in [197,198]. However, to the best of our knowledge, most of the reported studies were based on raw images in which the defects are at least barely visible (for example, SEM images).…”

Active meta-learning for predicting and selecting perovskite crystallization experiments

“…Applications in computational chemistry include parameterization of ML forcefields 34 , in silico drug discovery 42 , and efficient metadynamics sampling in protein molecular dynamics simulations 43 . Applications to chemical experimentation are more limited, but examples include Tandem mass spec proteomics (with a task transfer from unmodified to post-translationally modified proteins) 44 , defect identification in silicon CMOS devices (with a task transfer between transistor gate geometries) 45 , and band gap and catalytic activation energy prediction (with transfer between DFT prediction results and experimental values) 46 .…”