The University of Michigan- Shanghai Jiao Tong University Joint Institute (UM-SJTU JI) stands at the forefront of technological innovation with its unique academic atmosphere and advanced scientific explore. JI boasts a strong research team, which keeps innovating and strives to produce results that will lead the industry and ultimately benefit the human beings.
In order to help the public discover the mystery of scientific research, JI has launched a series of mini-workshop to introduce faculty’s research.
At a recent mini-workshop, Professor Weikang Qian presented “Approximate Computation: A low-power computing paradigm for fault-tolerant applications.”
In today’s information age, integrated circuits have become indispensable in our lives. Their applications range from smartphones, notebook computers, cars, to modern architecture.
In the professional field, the current integrated circuit design often requires a smaller chip with faster computing speed and lower power consumption. On the other hand, with the development of artificial intelligence and multimedia technology, there are increasingly wider applications in pattern recognition, machine learning, image processing, video processing, etc. These applications have one thing in common: their results do not need to be completely accurate; a small margin of error is acceptable. For example, in image processing, small deviations in pixel values are not perceivable; in web search, it is difficult for users to find the difference between a most relevant result and a highly relevant result.
Professor Weikang Qian presenting “The Application of Approximate Computation”
In this context emerged the concept of approximate calculation of the circuits. Due to the fault tolerance of the application, it is not necessary for the calculation results of the circuit to be consistently exact with the target. Therefore, one can sacrifice some accuracy to reduce the chip area, improve the speed of the chip and reduce the chip’s power consumption. In fact, some studies have shown that by cleverly introducing errors, one can greatly improve the circuit area as well as speed while reducing power consumption.
Professor Qian’s group is currently focusing on two aspects of the research in the approximate calculation of the circuits. Firstly, some arithmetic circuits, such as adders and multipliers, are used to design low power approximation circuits. Secondly, the algorithm of automatically generating the optimal approximation computation circuit is explored and an optimal approximation circuit is generated for any given target circuit and error constraint.
Professor Weikang Qian is an assistant professor at JI. He received his bachelor’s degree in automation from Tsinghua University in 2006 and his doctoral degree in electrical and computer engineering from the University of Minnesota in 2011. HIs research direction is mainly computer-aided design of integrated circuits and circuit design for new technologies. He received the Best Paper Nomination for the 2009 International Computer Aided Design Conference and the Best Paper Nomination for the 2016 International Symposium on Logic and Synthesis.
