Advanced FPGA technology trend based on patent analysis with link mining

This paper provides the trend analysis of the technology development on FPGA with Machine Learning based on the public information. In recent years, demands for the computing power are expanding due to reform of industrial structure typified by the Industry 4.0 and explosive epidemic of IoT (Internet of Things) and AI (Artificial Intelligence). Due to the restriction of power supply, the challenge of edge device is to achieve a good balance between the performance and the power consumption. GPU is not suitable for embedding to the edge devices due to high power consumption and heat generation. On the other hand, the devices using only CPU do not satisfy required performance. For the above condition, FPGA has drawn attention as CPU accelerator. Xilinx and Intel (former Altera) are FPGA suppliers. Hence, it is important to grasp their technology development trends for conducting business. In this paper, we reveal the technical development trend of the target companies from the patent information with the Machine Learning. Knowledge extraction from the patent information has been made so far, since the conventional patent analysis methods depend on the personal heuristic knowledge. It was hard to extract the technological structure from the amount of the target classifier. Thus, we are focusing on classification codes in the patent and Link Mining method is employed as the analytical method. Link Mining is aiming to abstract the elements and their relations in focus object into the nodes and the edges to search for the structural features. We analyze the patent information submitted from the three companies above. The graphs consist of the classification codes of the patents of every 6 months which are made for revealing the patents' technology structure. With the proposed method, we were succeeded in revealing the companies' focused technology areas, their technological transition, and their differences and common points from the results of extracting the graphs' features.