The world’s leading semiconductor manufacturers are in Taiwan, so in order to reach the goal of transforming Taiwan into an AI power in 2022, Ministry of Science and Technology initiated the "Semiconductor Moonshot Project" in 2018. The research team from the college of electrical engineering and computer at NCHU has executed the project since then and won the 2021 Future Technology Awards under the category “Semiconductor AI Edge Core Technology Competitiveness” recently. The award includes "Expandable and Modularized AI Hardware Accelerator", "computation aware AI Model architecture optimization", and "Real-time Computing and data labeling System". The Principal Investigator, Professor Yin-Tsung Hwang said, “To put it simply, our project is to integrate AI technologies into robots such that ‘sensing, positioning, navigation, and cloud analysis in order to make them perform at the same time. Through technology customization and robot development, it provides services under various application scenarios, especially those crowded environments such as supermarkets and airports. By means of real-time consumer data collection, the robot can even provide, in a rather precise manner, ‘shopping suggestions’ or ‘helpful services.’ to the customers.”
Service robots are currently the research and development focus in various countries. The robot developed in NCHU supports three major functions, i.e., "sensing (image, radar, LiDAR), cloud analysis, and navigation." In particular, the computing kernel of this robot is powered by a self-developed AI chip featuring "expandable" and "modularized" designs. This chip can process four to five times the amount of data achievable by the traditional device. In addition, through the optimization of "deep learning architecture", all environment sensing and object recognition tasks can be completed in real time at the edge (local) site, which minimizes the data traffic with the cloud. In other words, with the help of this AI chip, not only the robot becomes smarter and more helpful but it can be retargeted to different applications easily by simply reprogramming the AI model.
The development of this smart service robot prototype lasts for three and a half years. The robot equipped with AI chips can quickly detect objects in the surrounding, identify customers, predict their behaviors, and then provide the most appropriate services for them. The "age", "gender" and even the "visual direction" information of the customer can be analyzed “locally” without resorting to the cloud analysis, which results in cost reduction, shorter response time, and quick assessment of customers’ preferences. Another distinctive feature of this robot is “social aware” navigation. In other words, it observes basic human social norms when passing through a place with people around. For example, the robot can detect if two people along its navigation path are talking to each other and it will avoid cutting through them rudely. “Besides technological advantages, cost factor is a main concern when AI technology is employed. Only those provide "plentiful and cheap" solutions excels in business competition. This is also the goal that our team has been constantly paying attention to,” said Professor Tsung-Ying Huang with a smile.
Professor Chung-Pin Wu from the Department of Electrical Engineering, NCHU, who is in charge of the development of another "Future Technology Award" winning technology, "hardware chip accelerator", pointed out that this accelerator design meets the expectation of low power consumption, high performance, and cost reduction requirements altogether. This is a considerable breakthrough among the numerous teams worldwide actively devoted to this research. This technology largely enhances the AI chip performance in aspects such as speed, cost and computing complexity.
As the applications of AI become more and more widespread, how to achieve "maximum performance" with "minimum operation" subject to a limited data bandwidth is the key issue. Professor Chih-Peng Fan, one of the team members of the "Semiconductor Moonshot Project" from NCHU, further explained that through these awarded technologies, i.e., “Expandable and Modularized AI Hardware Accelerator", "computation aware AI Model architecture optimization", and "Real-time Computing and data labeling System", the industry does not need to redesign the "architecture". Instead, by simply changing the model parameters, the production line can quickly reconfigure for new productions. This helps industry quickly adapt to the ever-changing consumer markets.
At present, through various technology promotion and matching events sponsored by the MOST, the research teams in NCHU are gradually building up their partnership with the industry. Especially, five out of the eight award winning teams this year are from the college of EECS. Professors with research backgrounds in EE, CS, opto-electronics, and communication, are summoned to work together closely, brainstorm for creativity, and aim at achieving "scientific breakthrough" and "industrial applicability”. Professor Gu-Chang Yang, Dean of College of Electrical Engineering and Computer Science said, “‘unity is strength’.” They support each other and inspire each other. As the academic leader in central Taiwan, the college of EECD, NCHU continues to provide the strongest technical supports for industry while facing international competition.”
The research team from the collage of Electronic Engineering and Computer Science, NCHU