Study Suggests Possibility of AI Implementation on Hardware

 

Research Published in IEEE Transactions on Industrial Electronics

 

A Korean research team has developed a neuromorphic hardware system for visual pattern recognition with a memristor array and CMOS neurons.

 

                            

                                     Prof. Byung-geun Lee                                Mr. Myonglae Chu

 

The research was directed by Prof. Byung-geun Lee (corresponding author) and led by Mr. Myonglae Chu (first author) of GIST School of Mechatronics, and was supported by the Pioneer Research Center Program of the National Research Foundation of Korea. The result was published on September 9, 2014 in the on-line version of the IEEE Transactions on Industrial Electronics, a highly prestigious research journal in electronic engineering. (Title: Neuromorphic Hardware System for Visual Pattern Recognition with Memristor Array and CMOS Neuron)

 

A neuromorphic system refers to a very large-scale system of integrated circuits that mimic neuro-biological architectures present in the nervous system. A memristor is a two-terminal passive device that modifies or retains its resistance according to the time integral of current flowing through it, and a memristor array is the leading candidate for synapse implementation in a neural network.

 

Although a similar technology had earlier been achieved using software, it is the first time that a neuromorphic system for visual pattern recognition has been implemented on hardware. As such, the study is considered a milestone and is expected to contribute to the development and commercialization of artificial intelligence systems.

 

Also participating in the research were Prof. Hyunsang Hwang of POSTECH (Department of Materials Science and Engineering), Prof. Byoung Hun Lee of GIST (School of Materials Science and Engineering) and Prof. Moongu Jeon of GIST (School of Information and Communications). The outstanding result was made possible through convergent collaboration of researchers with different backgrounds over the past three years.

 

                    

 

Prof. Byung-geun Lee said, “The research demonstrated that a memristor array can actually replace the synapse in a neural network.” “It is expected to contribute to the development of the artificial intelligence system, which is currently faced with an obstacle as the CMOS fabrication process approaches its technological limits,” he added.

 

For further information or inquiries, please contact isso@gist.ac.kr.