To meet the demanding requirements for pattern fidelity, critical dimension and placement errors on advanced masks, the use of multi-beam mask writers together with using low sensitivity resists became necessary and inevitable. To reach the targeted throughput on such low sensitivity resists, an increase of the beam current is necessary which results in two problems. Worse beam stability control increases the risks of pattern errors and thus leads to higher yield loss. In addition, stronger resist charging and thermal effects also result in more unpredictable displacement errors which in turn make overlay control much more difficult. Here we present a new method that utilizes machine learning to detect tool abnormalities and trigger immediate exposure interruption which significantly reduces the mask yield loss. To reduce and compensate stronger charging and thermal effects from a higher beam current, we introduce hardware modifications and software corrections as well as an exposure sequence optimization that in combination minimize the yield loss and overlay problems and enable mask making in 3nm and beyond.
The successful development of Actinic Pattern Mask Inspection (APMI) has enabled the high-volume manufacturing of advanced chips, such as N5 and N3, due to the production of defect-free masks by tsmc's mask shop. This accomplishment can be attributed to the utilization of an innovative Extreme Ultraviolet (EUV) inspector and Graphics Processing Unit (GPU)-based defect detection with Artificial Intelligence (AI) assistance. The application of EUV inspector unleashed pellicle inspection to prolong mask operation periods in wafer foundries. Besides, the improving in the manufacturing efficiency via automation also boost the productivity in the mask shop. According to our previous report in BACUS 2023, the improvement by performing various approaches in the novel Laser-Produced Plasma (LPP) system enabled tsmc to capture EUV image with high stability. The continual improving in the system in later keep reducing the vibration of the crucible and hence improve the tin stability. Furthermore, tsmc developed a GPU-based inspection system, which allowed for flexible algorithm development compared to FPGA. The ML-based rendering model aligned features with tool images and reduced image residue. Therefore, the final inspected image could be possessed with high SNR in advanced node and aggressive OPC compared to DUV inspector. Additionally, the final inspection results will be processed via a Deep Learning (DL) based model, reducing false positives, and implementing auto-defect classification. By combining these contributions, the actinic tools were able to streamline the manufacturing flow and fulfill the requirements for massive production significantly.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.