Joint development of UX Factory with KAIST Professor Yoo Hee-jun's research team

An artificial intelligence semiconductor that processes deep learning more efficiently has been developed by applying technologies such as variable artificial neural networks.

Professor Yoo Hee-jun's research team at the Korea Advanced Institute of Science and Technology (KAIST) jointly developed an artificial intelligence semiconductor with semiconductor startup UX Factory.

Implementing artificial intelligence on mobile devices requires high-speed computations at low power consumption. However, current software technologies utilize slow computation speeds and high power consumption, making the development of an AI acceleration processor essential.

The research team expanded the scope of application of artificial intelligence semiconductors by developing an artificial intelligence semiconductor (UNPU: Unified Neural Network Processing Unit) that can simultaneously process convolutional neural networks (CNNs) and recurrent neural networks (RNNs) on a single chip and can set energy efficiency and accuracy differently depending on the recognition target.

This time, we also developed an emotion recognition system that recognizes human facial expressions through a smartphone camera, automatically recognizes seven emotional states, including happiness, sadness, surprise, fear, and expressionlessness, and displays them in real time on the smartphone.

Despite announcing an early version of the TPU at the 'HotChips' conference in August last year, where IT companies present semiconductor chips they have developed, Google's TPUIt has attracted much attention for demonstrating energy efficiency up to four times higher than that of the (Tensor Processing Unit).

Professor Yoo Hoe-jun explained the significance of the research, saying, “This research is significant in that it developed a semiconductor that accelerates artificial intelligence on mobile devices with low power consumption. It is expected to be applied in various fields such as object recognition, emotion recognition, motion recognition, and automatic translation in the future.”

Yong Hong-taek, Director General of the Ministry of Science and ICT's Information and Communication Industry Policy Bureau, said, "In order to efficiently provide AI services, we are promoting the development of AI semiconductors worldwide," and "The Ministry of Science and ICT is also planning a large-scale project to develop AI semiconductor technology in cooperation with the Ministry of Trade, Industry and Energy, and we plan to apply for a preliminary feasibility study within this year."