Bits + chips
HBM2 memory solutions gaining popularity
Alex Yang, DIGITIMES Research, Taipei
Wednesday 27 December 2017

The popularity of HBM2 (second-generation high bandwidth memory) solutions has been increasing recently due to the adoption of such products by Nvidia's Tesla P100 accelerator and Quadro GP 100 graphics card as well as AMD's Radeon Vega chip, Digitimes Research has found.

In addition to HBM-based products, there is a variant version of HMC (hybrid memory cube) chips. While both HBM and HMC solutions are made with a similar vertical stacking process to integrate multilayer DRAM devices on a base logic die, the HBM is more like a 2.5D stacked-die packaging solution, and the HMC is closer to a 3D stacked-die packaging process.

While SK Hynix and Samsung Electronics are using wafer-level packaging (WLP) process to fabricate HBM2-based chips, Taiwan Semiconductor Manufacturing Company (TSMC) is adopting CoWoS (chip-on-wafer-on-substrate) process for HBM2 devices. On the other hand, the Micron Technology/Intel camp focuses mainly on HMC-based products.

However, Intel's release of its deep-learning chip, Lake Crest, which came following its acquisition of Nervana, has come with HMB2. This indicates that HBM-based architecture will be the main development direction of memory solutions for HPC solutions by GPU vendors.

The high packaging cost and compatibility are likely to continue to affect the applications and popularity the HBM2 chips, Digitimes Research believes.

Realtime news
© DIGITIMES Inc. All rights reserved.
Please do not republish, publicly broadcast or publicly transmit content from this website without written permission from DIGITIMES Inc. Please contact us if you have any questions.