In her inauguration speech on May 20th, 2020, President Ing-wen Tsai declared information and digital industries as one of Taiwan’s six core strategic industries and promised that the government would take advantage of Taiwan's strengths in the semiconductor and information and communication technology (ICT) industries to promote the development of internet of things (IoT) and artificial intelligence (AI). Governed by the Ministry of Science and Technology (MOST), the Taiwan Semiconductor Research Institute (TSRI), National Applied Research Laboratories, is the hub of semiconductor research in Taiwan. In accordance with MOST policies, the TSRI supports academic research and development of AI chips in Taiwan, helps university spin-off and cultivates the required talents. On November 24th, 2020, the TSRI signed a contract for Arm® Flexible Access for Research with Arm, the leading semiconductor intellectual property (IP) company. This makes the TSRI the first institution in Asia and the world’s first government-endowed academic institution to join the Flexible Access for Research program.
AI has been ubiquitous for a period of time and is moving from the cloud to the edge. The cloud AI computing is done by connected supercomputers, while the edge AI computing is done by the AI chip which can be unconnected. With the aid of AI chips with high privacy protection and real-time response features, on-device AI applications such as smart voice assistant, smart medical diagnosis, and facial recognition can be realized.
However, AI chip design is extremely complex. AI applications can be run with various AI processors, including the central processing unit (CPU), graphic processing unit (GPU), and neural-network processing unit (NPU) inside the AI chip. Designing these AI processors from scratch requires expertise in multiple domains, such as big data analysis, deep learning algorithm development, hardware integrated circuits (IC) design, implementation, verification and testing. It’s labor-intensive, time-consuming and costs a lot.
Flexible Access is an IP licensing model announced by Arm in July 2019 that allows SoC (system-on-a-chip) design teams to initiate research before they license IP and pay only for what they use at production. Arm then announced Flexible Access for Research in 2020, which enables academic research teams to access and tape out with a wide range of IP and tools without any fees or costs. Accessing this real-world, commercially proven IP helps academic research specialists innovate their projects and focus on resolving critical research issues.
Through the partnership with Arm, the TSRI introduces AI processor CPU IP, which enables academics to access various AI processors for multiple use cases and thus focus on developing the core of AI chips, namely the AI accelerator circuits. This makes AI chip design easier and reduces the time to tape-out. It also helps research teams in universities to spin-out. It’s possible to use IP introduced from Arm by TSRI for the university startup's product development as long as the startup signs the contract for the similar Flexible Access for Startups program.