Researchers release deep learning compiler program for AI semiconductor development

By Park Sae-jin Posted : January 26, 2022, 13:12 Updated : January 26, 2022, 13:12

[Courtesy of ETRI]

SEOUL -- To help small-and-medium-sized enterprises develop artificial intelligence semiconductors with less cost and time, South Korean researchers have released a deep learning compiler that processes a particular programming language. The program was publicly released through Github, a global development platform.
AI semiconductors with neural processing units are chips specialized in processing AI-related applications. Each kind of chip is required to go through a delicate customization process for the operation of AI applications in sectors such as autonomous driving, facial recognition and internet of things (IoT) sensor management.
AI chips require different sets of AI deep learning compilers to create the best working code for maximized performance. Normally, major chipmakers develop and sell a set of AI semiconductors and compiler software but it is harder for smaller companies to focus on the development of hardware chips because it takes a long time and a lot of money to carefully develop and optimize system software and applications.
The Electronics and Telecommunications Research Institute (ETRI) said in a statement that its research team developed and released "NEST-C," a versatile AI compiler program that is compatible with AI chips, central processing units (CPUs), and graphic processing units (GPUs). The institute is also working with domestic software companies to develop services based on AI semiconductors.
"The compiler program is system software developed to vitalize South Korea's AI semiconductor industry," ETRI researcher Kim Tae-ho was quoted as saying. 
기사 이미지 확대 보기