SK hynix Delivers 48 GB HBM4E Memory: Enhancements in Speed, Power Consumption, and Thermal Management

SK hynix has announced the commencement of deliveries of the first samples of HBM4E, an evolution of the high-bandwidth memory range designed for artificial intelligence systems. The modules have been sent to the company's key clients, marking the start of the validation phase that precedes large-scale production.

According to the South Korean manufacturer, samples of the 12-layer HBM4E version were delivered on schedule thanks to the experience gained in the development and production of previous generations of HBM memory. The goal is now to collaborate with partners to complete the path to commercialization.

The new memory is designed to meet the training and inference needs of large AI models. SK hynix indicates a maximum transfer speed of 16 Gbps per pin, along with an energy efficiency increase of over 20% compared to previous solutions. These improvements aim to increase the amount of data processable by accelerated systems while keeping consumption under control.

Among the innovations introduced is an updated interface and design optimization that allow for reduced latency in data transfers. The company claims that these modifications enable stable operation even in high bandwidth scenarios, a typical condition for AI data centers and high-performance computing platforms.

From a construction perspective, the memory uses proprietary Advanced MR-MUF (Mass Reflow Molded Underfill) technology, a process that includes the injection of protective material between stacked chips to enhance circuit protection and the structural robustness of the package. Thanks to this solution, SK hynix has managed to create a 12-layer configuration with a total capacity of 48 GB.

NVIDIA and SK hynix Team Up for Next-Generation AI Platforms

The company also highlights an improvement in thermal behavior. Compared to the previous HBM4 generation, the new HBM4E reduces thermal resistance by 17%, promoting more efficient heat dissipation and greater operational stability in high-end HPC and AI systems.

It is worth noting that Samsung has also announced the start of shipments of the first samples of HBM4E with characteristics very similar to those of SK hynix, anticipating the intention to introduce 32 GB variants with 8-layer stacks and 64 GB models with 16-layer configurations, depending on customer demands, in the future.