ETRI researchers on the NPU via the Nest compiler
We are demonstrating that deep learning image classification technology is in operation.


'NEST-C' Unveiled to Enhance AI Deep Learning Performance
Helps SMEs and startups reduce design and development costs

The Electronics and Telecommunications Research Institute (ETRI) took the lead in revitalizing the domestic AI semiconductor ecosystem by unveiling core technologies that reduce the time and cost of developing artificial intelligence semiconductors.

ETRI announced on the 26th that it has developed 'NEST-C,' a deep learning compiler and core AI system software. In addition, it released it on the web (Github) along with its self-developed AI semiconductor hardware so that developers can easily utilize it.

'NEST-C' has also been established as a standard by the Korea Information and Communications Technology Association (TTA), and by resolving issues of compatibility and scalability between hardware and software, it is expected to accelerate the development of AI semiconductors.

'NEST-C' is expected to shorten application development and optimization time, particularly for small and medium-sized enterprises and startups that have found it difficult to focus their capabilities on semiconductor design.

To run applications such as autonomous driving, the Internet of Things (IoT), and sensors, optimized AI semiconductors must be designed for each. Generally, manufacturers develop and sell AI semiconductors, system software, and applications together.

Previously, compilers had to be developed for each 'type of deep learning platform,' but now the Nest compiler can replace that.

In particular, this is the first time that both software and hardware for AI semiconductor development have been unveiled.

ETRI plans to expand the scope of support for deep learning compilers in collaboration with domestic companies and is also pursuing the commercialization of AI semiconductor application services through specialized software companies.

They also announced plans to contribute to the creation of new services by improving the performance and convenience required for the development of AI application services.

Kim Tae-ho, Head of the Next-Generation System Software Research Lab at ETRI, said, “The open-source release of the standard deep learning compiler is system software developed to revitalize the domestic AI semiconductor ecosystem. We are currently collaborating with various AI semiconductor companies to apply it.”