Search

In this research, we employed distributed systems to explore the similarities in parking ticket records using unsupervised machine learning algorithms on a large dataset. Using 37 million ticket records (9 GB) collected by the New York City Department of Finance, we applied an algorithm to cluster existing tickets and dive deeper to find the distribution of precincts within different clusters. Amazon Web Services including S3, EC2 and EMR, and tools like MongoDB and Apache Spark were used in this endeavor. In this study, computational time and cost for different EMR settings were evaluated. We conclude that there are significant computational advantages to using distributed systems when implementing unsupervised learning on a large dataset as well as storing and managing data. We also observed that it is time efficient for a cluster with more workers instead of fewer workers with large memory space for the utilized data set. However we observed a trade-off between the execution time and the total cost for the cluster configuration.

Typical controlled collapse chip connection (C4) bump structurally composed of solder bump, Ni UBM and Cu/Ti sputtered film on the top side of Si wafer in which polyimide film patterned and coated onto wafer as a buffer layer between substrate and UBM for stress relief. C4 bump structure is widely used in high-performance device of semiconductor package industries for many years and has achieved good reliability performance. In the past experience, stress-induced polyimide crack would happen on low-k IC package and be associated with inter-layer dielectric (ILD) cracking during thermal process and resulting in reliability problem. In order to resolve polyimide cracking problem, this preliminary study is attempted to optimize the bumping process based on the experimental of C4 bump structure with Copper pad as polyimide underneath to simulate the key condition of bumping process. Three key parameters in process, curing temperature, reflow peak temperature and reflow cooling rate, were considered to differentiate the influence degree related to the thermal stress from coefficient of thermal expansion (CTE) mismatch. Focus ion beam (FIB) and finite element methods were separately employed for cranny observation and stress modeling. The result exhibits the cured polyimide cranny formed under the condition of higher curing temperature, higher reflow cooling rate. The dominant factor is polyimide curing temperature (PCT), and followed by reflow peak temperature (RPT) and reflow cooling rate (RCR), sequentially. The cranny on polyimide propagated to and around the round-shaped Ni undercut spot after reflow due to the thermal stress arising from CTE mismatch between the polyimide and contained metals. Low curing temperature (350°C) with lower modulus of elasticity, low reflow temperature (240°C) and slow cooling rate (2.9°C/s) allow polyimide to absorb more strain energy and flexibly adjust internal stress far from fracture stress.

In this paper, we describe an environment monitoring and control system for fish farming applications. This system can receive sensor data in proprietary format and map it to a standard data model. All sensor data and control commands can be accessed via well defined HTTP RESTful interfaces. With this interface, historical data and statistics can be easily accessed and application development time can be greatly shortened.