Samsung Electronics Co. said Tuesday it was ramping up capacity for second-generation, or 8GB (bytes per second) high-bandwidth memory (HBM2) - costly, high-performing, and power-efficient memory mostly used to power machine-learning data centers and supercomputers - to meet growing demand for their applications in graphic cards and network systems.

The world’s largest memory chipmaker that became the first in the industry with capacity of mass-producing the second-generation HBM, a multi-stack of DRAM dies, in June last year will make the chips take up more than half of the total output in high-performance memory by the first half of next year. It also would release an upgraded version in the 8GB HMB2 based on new designing and processing technology in the second half of this year.

“We are timely meeting our capacity with the needs and applications for high-memory solution,” said Han Jae-soo, executive vice president of the Memory Sales & Marketing team at Samsung Electronics.

The interface stacks up eight 8GB (bits per second) core dies on a buffer die and is vertically interconnected by through-silicon via (TSV) holes and microbumps. A single die contains over 5,000 TSV holes - 36 times that of a DDR4 DRAM of the same GB - to deliver superfast data transmission performance to take up tasks like image recognition and natural language processing. A single 8GB HBM2 package hosts 40,000 TSVs.

The memory can transmit 13 20GB UHD motion pictures in a second.

The chip is also designed to prevent overheating from multi-tasking.

The package is the same size as 4GB HBM2 but with double the space and capacity and twice better power efficiency.

By Lee Dong-in

[ⓒ Pulse by Maeil Business Newspaper & mk.co.kr, All rights reserved]