AI chips work as the human brain serving Samsung’s next-gen hidden card
In the near future, you would not need to clarify pronunciation in refined manner to your mobile device. Three to five years later, AI assistants will be commercialized to carry out various tasks if you talk to them in a friendly and informal way. It may not sound like rocket science but it takes Neural Processing Unit (NPU) technology to calculate task-performing more than 10 trillion times per second. NPU is called a next generation semiconductor that works in the same way as the human brain. The so-called “AI chips” learn information and process it into data on their own just as the human brain does.
Samsung Electronics has embarked on its journey to strengthen next-gen NPU technology with the aim of becoming the world best in system semiconductor by 2030. The Korean semiconductor producer announced on Tuesday to put company-wise energy into NPU business by increasing the number of related employees from 200 to 2,000 within the next 10 years.
Parallel computing is the key to processing multiple calculations simultaneously just as human neural networks do. NPU is considered integral in processing parallel calculations efficiently and swiftly, thus realizing AI accordingly.
Samsung Electronics already released Exynos 9 (9820) equipped with independently developed NPU last year. It is a result of ceaseless research effort done by Samsung System LSI Business and Samsung Advanced Institute of Technology. Exynos 9 is based on system-on-chip (SoC) system, which puts various features in a single chip. It enables “on-device AI” that processes AI calculations on a mobile device, which has been done by exchanging data with cloud servers.
On-device AI does not need data to pass via clouding systems, thus helping protect privacy. Likewise, it responds quickly and consumes low power and cost even without networks. It can be combined with 5G telecommunications technology to connection surrounding AI devices to clouding systems, which will open a new world of reasoning and learning, according to Samsung Electronics.
Dong-Jun Heo firstname.lastname@example.org