An artificial intelligence (AI) chip that automatically alters its response characteristics to match the changing speed of incoming data has been developed in Korea. The chip reduces prediction errors by up to 40 times compared with existing AI chips, and is expected to enhance the performance of next-generation AI devices such as autonomous vehicles and robots.
According to the scientific community on the 8th, a research team led by Distinguished Professor Choi Shin-hyun of the School of Electrical Engineering and the Graduate School of Semiconductor Engineering at the Korea Advanced Institute of Science and Technology (KAIST) recently developed a new AI chip device called the "Programmable Dynamic Memtransistor (PDM)," which adjusts its response speed to match the changing speed of input data.
Real-world data does not all change at the same speed. Some information changes very rapidly, such as heartbeats or voice, while other information changes slowly, such as weather or the condition of factory equipment. However, existing AI chips process data only at a fixed response speed once set, leaving them limited in their ability to simultaneously handle information that changes at different speeds.
In response, the research team combined the function of storing information with the function of adjusting response speed into a single chip, enabling it to operate at the speed best suited to the characteristics of the data.
Experimental results showed that the PDM could adjust its response speed within a range of about five times, and in time-series data prediction, it reduced prediction errors by up to 40 times compared with conventional fixed-type AI chips. A chip integrating multiple PDMs maintained accuracy similar to that of existing software-based AI while operating with less power.
The technology has the strength of high potential for mass production and commercialization, thanks to its non-volatile characteristic that retains settings even when the power is turned off and its compatibility with existing semiconductor processes (CMOS process). "This research has implemented an AI chip in which the semiconductor responds on its own in the most suitable manner according to the changing speed of data," Distinguished Professor Choi Shin-hyun said. "We expect it to become a core technology that enhances the performance of various AI devices such as autonomous vehicles, robots, and wearable devices while reducing power consumption."
Meanwhile, the research results were recognized for their achievements and published in the July issue of the international journal Nature Communications.
