Domestic researchers are studying a software development platform capable of easily and conveniently automatically generating and integrating various artificial intelligence (AI) services embedded in smart devices, such as visual intelligence, emotional intelligence, object intelligence, and healthcare intelligence. This is expected to be a significant boost to the revitalization of the domestic AI semiconductor ecosystem and the securing of technological supremacy in smart devices in the future.

The Electronics and Telecommunications Research Institute (ETRI) announced on the 14th that it is developing a smart edge device software (SW) development platform.

The smart edge device software development platform currently under development by ETRI is a next-generation platform technology that facilitates the development of AI services customized for edge devices at low cost and high efficiency.

By utilizing this platform, the performance of AI models running on edge devices can be improved by up to 25 times, while reducing storage space to less than 1/10.

To implement AI services on edge devices, it is necessary to lighten and optimize AI models to suit the device environment.

Because the terminal itself has lower performance and storage space compared to the central serverIt is a door.

Previously, AI services faced limitations in areas closely related to daily life due to the need for large-capacity storage and high-performance GPU processors, but this platform easily resolved this issue.

This technology can easily optimize AI models for smartphones, wearable devices, and IoT devices, and is expected to contribute significantly to the creation of various lifestyle-oriented AI services such as smart homes and smart health.

The research team explained that AI technology can be easily incorporated into application software without massive cost investment, and that it is easy to distribute and manage via 4G and 5G mobile communications.

In addition, it supports more than eight types of artificial intelligence frameworks, such as TensorFlow and PyTorch, and provides a development environment that enables automatic compilation of AI models for edge devices and integration into application software.

Once this technology is developed, it is expected to open the way for providing mobile AI services in various fields, such as visual intelligence, emotional intelligence, object intelligence, and healthcare.

As the utilization of domestic AI processors installed in edge devices increases, it is expected that the revitalization of the AI semiconductor ecosystem and the securing of technological supremacy will gain momentum.

ETRI stated that once this technology is developed, it will produce: an integrated framework for deep learning model lightweighting technology; a lightweight edge deep learning model based on deep learning acceleration hardware; a deep learning compiler based on multi-stage intermediate representations; a deep learning operation information profiler supporting multiple deep learning acceleration hardware; an automatic edge deep learning model generation platform based on deep learning acceleration hardware; and an integrated development environment system for edge device application software based on deep learning acceleration hardware.

The research team stated that the development of this technology follows the 'AI Model Compiler NEST-C' and 'AI Acceleration HW SoC' developments previously carried out by ETRI. They explained that this is possible because they possess source technology resulting from the development of prior art.

Na Jung-chan, Head of the Seoul SW-SoC Convergence R&BD Center at ETRI, stated, “It is highly significant that we commenced development of this technology, which was considered a promising technology, at an early stage. We will develop this technology to a world-class level to contribute to Korea leading the way in AI software development platforms for edge devices.”

The technology to be developed is planned to be transferred to smart edge device companies, AI semiconductor development fabless companies, artificial intelligence edge solution companies, cloud companies, AI SaaS, etc. in the future.

This technology is being developed as a project under the Ministry of Science and ICT's 'Smart Edge Device Technology Development Project' to develop SW platform technology for the development of 5G convergence services based on smart edge devices using domestically produced artificial intelligence semiconductors, with Therromind Co., Ltd., Nextchip Co., Ltd., Open Edge Technology Co., Ltd., and Korea University of Technology and Education participating in the technology development until 2026.