The Japan News by The Yomiuri Shimbun

Home > Research > Exploring the potential of semiconductors Seeking world leading computational performance


Knowledge Co-Creation - Profiles of researchers

Exploring the potential of semiconductors
Seeking world leading computational performance

Dajiang Zhou
Assistant Professor, Faculty of Science and Engineering (Graduate School of Information, Production and Systems), Waseda University

A coherent system from design to implementation

I specialize in semiconductor design research and development. I have been conducting research on the design of highly efficient, low energy consumption VLSI (very large-scale integrated) systems that can process high volume video data for next-generation televisions. In 2014, I was extremely honored to receive the Waseda Research Award (High-Impact Publication).

I started conducting semiconductor design research when I began my master’s degree program at Shanghai Jiao Tong University, after graduating from with a degree in electronic engineering. While thinking about my future and the most significant topics in information and electronics of the next 40 to 50 years, artificial intelligence came to mind. There was no doubt in my mind that artificial intelligence would exceed the limits of human brainpower, and it would be necessary to develop powerful artificial intelligence in VLSI, not software. Compared to improvements in performance through software development, the development of VLSI can achieve one hundred to one thousand times greater performance. I strongly desired to take on such as challenge.

I first met Professor Satoshi Goto (who retired in March 2015) from Waseda’s Graduate School of Information, Production and Systems (where I currently reside) while presenting a paper at an international conference held in the U.S., during my master's program at Shanghai Jiao Tong University. Professor Goto is a globally respected researcher on semiconductor design and had a close relationship with my supervisor at Shanghai Jiao Tong University. Therefore, while applying for a doctorate, I asked Professor Goto to be my advisor and left China to study at Waseda’s Graduate School of Information, Production and Systems located on Kitakyushu Campus.

Semiconductor design starts from algorithms and architecture to circuit design and implementation. Few research institutions in the world are involved in the research and development of this entire process. However, developing semiconductors with the highest possible performance is not possible without a coherent research process for this system. Full-fledged implementation requires, in addition to research capabilities, research grants in the form of governmental projects or university-business alliances. Massachusetts Institute of Technology (MIT), National Taiwan University and National Chiao Tung University, which are strong in semiconductor research, and Waseda's Graduate School of Information, Production and Systems have established a coherent research and development system. This was the biggest motivation for me to study abroad.

Graduate School of Information, Production and Systems, Waseda University (Kitakyushu Campus) (excerpt from the graduate school's leaflet)

Received the best student paper award of VLSI Circuits Symposium 2010 while completing my doctorate (I met Professor Goto, my advisor at the 2011 symposium)

Presenting a series of world leading chips

My current research theme is not artificial intelligence, but multimedia video processing. The world of video processing is progressing at a remarkable rate, and every year I take on the challenge of designing and developing a new VLSI chip. Over the last eight years since I came to Waseda, I have designed and developed seven semiconductor circuits and implemented them into chips with Professor Goto and as a member of Professor Shinji Kimura's.

Although 4K resolution televisions (3840 × 2160 pixels) are now commercially available, there is now research that focuses on the next-generation 8K standard (7680 × 4320 pixels). This new standard can process huge volumes of data and can handle data volume that is over 100 times the size of current-high definition videos and has an video compression ratio that is 100 to 200 times the quality of current standards. To maximize its power, there is no other way but to implement a high performance VLSI system with dedicated video compression and decompression functionality. Although a standard compression/decompression algorithm exists, it is very difficult to maximize performance by implementing them into semiconductors. Researchers around the world are competing against each other to accomplish this goal.

Figure 1. The progress of video compression technology and the remarkable improvement in video resolution

Figure 2. Video decompression process

It is not possible to develop a single semiconductor chip by yourself. I am working with excellent team members. With luck and timing on our side, we have been able to present a series of the world leading chips. In 2012, we developed the world’s first top performance standard chip for 8K resolutions, which reporters of Japan’s semiconductor industry selected for the Semiconductor of the Year Award. Reporters usually select commercial VLSI chips for business, and it is quite unusual for a university team to receive the award. I believe this reflects the consistent achievements of the university's integrated research and development structure.

Semiconductor research is fun during the phase where we come up with design ideas. However, as we go through the implementation phase, we experience many difficulties and come under much pressure. Once we send the specification of a semiconductor chip, in which several million transistors and devices are integrated, to a company for implementation and production, we cannot easily make modifications when a problem arises. It costs over 10 million yen to create a prototype of one chip and even one mistake may become a serious issue. We are most happy when we find out our completed chip runs without any problems. This validates and proves the value of our ideas.

The truth is, we once failed to create a prototype for the world's first chip for 8K resolution. Due to a small error made by the partner company to which we outsourced implementation and production, the planned release of a chip for 8K resolution in 2011 following the release of a chip for 4K resolution in 2010 was delayed by a year. When the problem occurred, I visited the partner company in China and desperately searched for the cause of the problem. After discovering the source of the problem, it took half a year to recreate the prototype through a minor modification. Fortunately, there are many direct flights from Kitakyushu and Fukuoka to China every day and the distance to China is about the same as that to Tokyo. This means I can leave Japan in the morning and have lunch in China the same day. It is very convenient.

After the H.264 standard for video coding—the next-generation following MPEG2—was established in 2004, organizations began research and development based on new coding. Since 2013 when HEVC, an even newer generation of coding, was established, organizations proactively began research and development on chips for high compression coding of HEVC. Our team also has successfully developed a chip for HEVC in 8K, and we intend to be the first to present a research paper at the International Solid-State Circuits Conference (ISSCC). The ISSCC is the “Olympics” of semiconductors and will be held in February 2016.

Figure 3. Recent example of success in VLSI development design
8K UHDTV H.264/AVC Encoders (2009 to 2012)

Full-scale launch of artificial intelligence VLSI research

Going forward I would like to continue my research on video processing and begin research on artificial intelligence. Compared to when I first began my activities as a researcher, there have been remarkable research developments in artificial intelligence and its waves of practical application centered on deep learning. In addition to semiconductor design based on ideas at the forefront of artificial intelligence, I would like to devise unique and optimal algorithms for semiconductors. I have already begun gradually conducting research on deep learning functionality with semiconductors.

Advanced research on deep learning was originally conducted at Stanford University and other institutions of higher education. These days, large Internet corporations such as Google and Baidu are leading the field. However, their research is software based and uses several thousands of servers to operate, resulting in power inefficiencies. Many dedicated VLSI processors are used in the field of video processing and in the coming era, VLSI will lead the field of deep learning. My goal is to develop dedicated VLSI design and achieve high-level results.

Everyone knows that VLSI is vital in developing next-generation artificial intelligence. In practice, however, those familiar with algorithms for deep learning do not understand VLSI design well while those familiar with VLSI design do not understand artificial intelligence well. The pursuit of the best performance possible requires people who understand both. By taking advantage of my research experience, I would like to engage in semiconductor design for artificial intelligence. In addition to algorithms for artificial intelligence, a challenge of implementing algorithms will be successfully enabling massive data communication between an VLSI chip for computation and a memory chip. To address this issue, we can apply our technical prowess.

Getting students interested in research

One reason why we can produce great research results is the excellent post-doctoral researchers and graduate school students at this university. Ninety percent of students involved in semiconductor design research at Kitakyushu Campus are students from overseas, and most of them are from China. Their employment opportunities are very promising and are often hired as developers. They are playing key roles at prominent companies including major Japanese manufacturers. Unfortunately, however, most of them transfer to foreign-owned companies in Japan or institutions in the United States and other countries within several years.

While I give lectures and advise research, I try to get students interested in research as much as possible. I believe the best way to inspire researchers is to help them find a research topic that interests them. This was the case for me.

At the site of IEEE International Conference on Image Processing 2014 with other members of Satoshi Goto's lab

Many awards for papers and other awards received individually and as a team member

Dajiang Zhou
Assistant Professor, Faculty of Science and Engineering (Graduate School of Information, Production and Systems), Waseda University

Dajiang Zhou was born in Shanghai in 1983 and graduated from the School of Electronic, Information and Electrical Engineering. He completed his master's course at Shanghai Jiao Tong University and his doctorate degree at the Graduate School of Information, Production and Systems at Waseda University, receiving his Ph.D. in Engineering. After serving as a Doctoral Research Fellow at the Japan Society for the Promotion of Science and Junior Researcher at Waseda University, he was appointed to his current post in 2013. He has published more than 90 papers in top journals and at international conferences in the semiconductor field and has received many awards, including the best student paper award of VLSI Circuits Symposium 2010, the design contest award of ACM ISLPED 2010, the 2014 paper award of the Institute of Electronics, Information and Communication Engineering (IEICE), and the award for excellent Chinese students studying abroad. He also received the Waseda Research Award (High-Impact Publication) in 2014.
→See the summary of Dr. Zhou's research history