Samsung takes a new step in the artificial intelligence chip market by announcing the development of a High Bandwidth Memory 3E 12H DRAM chip with advanced TC NCF technology, surpassing its competitors.
The high-bandwidth dynamic random-access memory consumes a low amount of power, with very wide communication lanes. This is achieved by vertically stacking memory chips to prevent processing bottlenecks caused by traditional memory chips.
TC NCF stands for the Non-Conductive Pressure-Insulating Material placed between stacked chip layers.
TC NCF provides thermal properties that improve cooling efficiency, and the method used in the new HBM3E 12H DRAM chip contributes to enhancing productivity.
The company has stated that it has begun offering samples of the HBM3E 12H to customers, and mass production is expected to begin in the first half of this year.
Samsung’s Executive Vice President of Memory Product Planning, Bai Young Chul, stated: “Artificial intelligence service companies require high-bandwidth dynamic random-access memory chips, and HBM3E 12H has been designed to meet these needs.”
The new chip is part of Samsung’s efforts to develop core technologies and achieve technological leadership in the high-bandwidth dynamic random-access memory market in the era of artificial intelligence.
Samsung believes that high-bandwidth dynamic random-access memory chips may be the most suitable option for future systems requiring additional memory, given the rapid growth in the artificial intelligence market.
Estimates show that the speed of artificial intelligence training can be increased by 34% when using HBM3E 12H in artificial intelligence applications compared to HBM3 8H chips, with an increase in concurrent users for inferencing services by over 11.5 times.
Samsung stated that the chip’s performance and capabilities allow customers to manage resources flexibly and reduce the overall costs of data center ownership.
HBM3E 12H provides a high bandwidth of up to 1280 gigabits per second with a capacity of 36 gigabytes, and it improves bandwidth and capacity by 50% compared to the HBM3 8H chip.
