Finding a Nano-Scale Needle in a Peta-Scale Haystack – Anomaly Detection

KLA is a leader in process control using advanced inspection tools, metrology systems, and computational analytics, supporting chip manufacturing for a broad range of device types, including advanced logic and memory (3D NAND, DRAM, MRAM, etc.), power devices, RF communications devices, LEDs, photonics, MEMS, and more. Chip manufacturers use KLA’s array of products to help accelerate their development and product ramp cycle, and improve overall profitability in the chip manufacturing process.​

To detect anomalies within the manufacturing process, inspection systems must be capable of identifying defects as small as 10nm and do so quickly while minimizing false positives. KLA chose DDN to architect a data management solution with the write speed necessary to keep up with enormous data processing demands and safeguard the efficiency and reliability of their manufacturing process.

Finding a Nano-Scale Needle in a Peta-Scale Haystack – Anomaly Detection

Challenges

The multi-step workflow involves capturing high-resolution images, performing image processing and segmentation, and analyzing data to identify and classify defects. The raw data generated by this process per wafer can be up to 300TB, which is analyzed against large data sets with the goal of completing a run on a wafer in under an hour. The data-intensive nature of this process necessitates a robust and scalable data infrastructure that can deliver fast write performance while delivering massive amounts of data to the computing systems and increasing data center efficiency.​

Solution

引用

We have limited access to our system so we really needed something that was a single platform that would scale with ease.​

Krishna MurikiHPC Systems Design & HPC Architect KLA