Established in 1983, SKHynix manufactures semiconductors, such as DRAM and NAND flash memory products and CMOS image sensors, for the IT industry. The company has grown rapidly along with the digital sector, especially in the emerging markets of China and India. It is ranks eighth in the world in the semiconductor market and employs around 24,000 employees in more than 30 offices, worldwide.
SKHynix wanted to reinforce its capabilities by researching, developing, and manufacturing innovative, new semiconductor products. It needed to implement a robust and scalable infrastructure that would support rapid and challenging changes within the company and in the global semiconductor market.
The company implemented Oracle Exadata Database Machine and Oracle’s Sun ZFS Storage 7420 appliance to improve the performance, stability, and efficiency of its next-generation fault detection and classification (FDC) system. It has cut transaction response times by 77%, compressed data storage by more than 50%, lowered CPU usage by around from 44% to 17.5%, and can complete daily batch processing tasks up to 13x faster.
SKHynix’s FDC system is a core component of the company’s manufacturing execution system, which it uses to track the semiconductor manufacturing process. The FDC system enables engineers to analyze data from manufacturing processes (including equipment sensor data), detect defects and abnormal conditions, and trace them back to the machinery causing the problem. Engineers can then perform corrective repairs, and proactively schedule maintenance before the equipment fails. This level of advanced production control is especially important in the complex semiconductor manufacturing industry, where each semiconductor goes through several hundred production steps.
The FDC system needs to process up to 18,000 transactions per second (TPS) and approximately 1.5 billion transactions each day, drawn from around 40TB of data. SKHynix wanted to improve the FDC system’s data processing performance and reduce the transaction response time, enabling the company to more efficiently monitor its manufacturing processes as the size of its operations grew.
“We wanted a highly available, stable infrastructure for our FDC system, with multiple database and application servers and clusters,” said Song Jae Young, senior manager, head of automation team I, SKHynix. “This would allow the system to easily process thousands of transactions a second and scale to accommodate an increase in data volumes in the future without experiencing performance issues or unscheduled downtime.”
SKHynix wanted to reduce and stabilize the high rates of CPU, memory, and I/O resource usage in its legacy database. It also wanted to improve the storage capacity of its FDC system. This would ensure the system would remain cost-effective and efficient as the volume of data about semiconductor manufacturing increased.
By implementing Oracle Exadata, SKHynix has reduced the FDC system’s data transaction response time by 77%.
With the legacy infrastructure, response speeds had reached an average of more than 30 milliseconds for each transaction, which was affecting how quickly the FDC system could analyze data from manufacturing processes and detect faults. By taking advantage of Oracle Exadata features, such as Smart Flash Logging, the FDC system can process transactions in about 7 milliseconds.
“We’ve experienced a dramatic improvement in the performance of our FDC system,” said Song. “It can now easily process up to 1.5 billion transactions each day, ensuring we can quickly detect any equipment issues that would affect our manufacturing process.”
In addition, proof-of-concept tests showed that running on Oracle Exadata enables the FDC system to process 8x more transactional load without any performance degradation. This will allow SKHynix to increase its semiconductor manufacturing without needing to add more hardware.
SKHynix can now complete its daily batch processing tasks twice as fast on average and up to 13x faster in some instances, by using Oracle Exadata software features, such as Smart Scan. Smart Scan performs real-time, targeted data searches, completing queries in the storage layer and only returning relevant data.
“Daily batch processing time has been reduced from four hours and 14 minutes to two hours and 23 minutes on average, and the minimum batch processing time is now 19 minutes,” said Song. “The SQL execution speed for key units of data has increased by an average of 2.5 times, which means we can more quickly retrieve, analyze, and report on information about the performance of our manufacturing equipment.”
Oracle Exadata has lowered the CPU usage of the FDC system’s database by approximately 44% to 17.5%. The database machine has also cut memory usage from 215 GB to 67 GB, reducing memory usage from 84%, down to a rate between 65% to 70%.
“Using Oracle Exadata’s Resource Manager feature, we can efficiently monitor CPU and memory usage, as well the usage of I/O resources for different processes,” said Song. “The resource manager controls the large volume of database requests that were previously executed without the administrators’ knowledge, and prioritizes data processing tasks. This maintains I/O usage at an average of 5%, which ensures stable processing speeds for the FDC system.”
Oracle Exadata’s advanced hybrid columnar compression function has helped reduce the storage consumption of the FDC system by compressing the system’s entire data table to less than half of its previous size. This will also ensure the FDC system can scale without requiring extra hardware.
SKHynix has also improved the analytical accuracy of its FDC system by retaining three months of historical data rather than one. This ensures the company can more efficiently monitor the performance of different pieces of equipment over time, schedule maintenance in advance, and improve the overall efficiency of semiconductor manufacturing.
SKHynix is using Oracle Exadata’s Automatic Storage Manager and real application clusters to ensure the FDC system is always available, by providing a dual infrastructure which balances data processing across its four database servers and two pairs of standby-active servers.
In addition, the company has deployed Oracle’s Sun ZFS Storage 7420 appliance to provide faster backup and restore speeds.
SKHynix chose Oracle Exadata as it was impressed by the database machine’s overall throughput, SQL execution, application processor response speeds, and ability to dramatically improve the performance of its batch processing.
The company was attracted to Oracle Hybrid Columnar Compression technology in Oracle Exadata, as it would maximize the efficiency of its storage resources and ensure the FDC system could accommodate business growth without needing extra hardware. It also wanted to take advantage of Oracle Exadata’s real application clusters and automatic storage manager features to ensure high availability.
“In addition, we found Oracle Exadata was superior to competing products as it was easier to transfer existing databases and applications to it,” said Park. “After also considering its scalability and flexibility, we decided Oracle Exadata was the best platform to handle and manage our vast amounts of data.”
SKHynix chose Oracle Exadata X2-2 Half Rack in May 2012 after conducting an extensive proof-of-concept.
The company needed to complete Oracle Exadata implementation in a short timeframe. It installed the machine in just a few days, before migrating the database and testing applications in two phases in June 2012.
SKHynix then completed a 10-day introductory and stabilization period before going live in July 2012. The FDC system has been operating stably and efficiently since then.
Oracle Specialized Partner Youngwoo Digital worked with Oracle and SKHynix to provide consistent and stable support throughout the proof-of-concept, product selection, and implementation. The company also provides SKHynix with postimplementation assistance.
“We’re happy with how smoothly Youngwoo Digital carried out the implementation, including the simple and hassle-free data transfer and integration process,” said Song.