Samsung, the largest chip maker in the world, is in urgent need to make chipsets that have dedicated cores for processing artificial intelligence-related data. Although the company has released its AI-powered digital voice assistant Bixby for the Galaxy S8 duo and the Galaxy Note 8, the Exynos 8895 chipset that powers the phones do not have a dedicated processing unit for AI-related tasks. Samsung has now embarked on research and development of AI chips and is planning to commercialize them in the next few years.

Huawei has announced that it will release its Mate 10 smartphone with its new Kirin 970 chipset that has a neural processing unit. Even Apple has launched three new smartphones – iPhone 8, iPhone 8 Plus, and iPhone X – which use the A11 Bionic SoC, featuring a “Neural Engine” that will help in machine learning tasks. “(Samsung) is in the middle of developing several types of chips that will be capable of processing massive data from AI applications on devices, eliminating the need to communicate with cloud servers,” an industrial source from one of Samsung’s partners told The Korea Herald.

Currently, data processed by AI apps such as voice recognition and machine learning are stored in the cloud and recalled when needed. With chips that have dedicated AI-related processing units, data can be stored and processed locally, which will improve the performance of AI features by 50 percent compared to current-generation chipsets. A lot of startups and tech giants are now focussing on improving the performance of their AI-powered devices. Market intelligence firm TrendForce says that AI will become the major growth factor in the future.

Samsung officially hinted at a technology forum that its new chipsets will have some AI-related features. Kim Ki-nam, president of Samsung’s semiconductor business, said, “The existing central processing unit and graphic processing unit chips make it hard to achieve efficiency in AI computing. The NPU will help address the efficiency challenge. However, NPUs so far have storage capacities that are equivalent to a thousandth of a human brain.” The company invested $300 million in Graphcore, a British AI chip startup, in order to overcome its limits.