Early HBM4 Validation Points The Way For Next Generation AI And HPC Systems


As AI and high‑performance computing systems continue to scale, memory bandwidth has emerged as a primary system‑level constraint. Larger models, higher compute density, and increasingly complex multi‑die designs are driving the need for memory interfaces that can deliver extreme bandwidth while operating within tight power and signal‑integrity margins. High‑Bandwidth Memory (HBM) has... » read more

Router-in-a-Package Design Combining HBM4, Chiplets and In-Package Optics (Technion, Berkeley, UCSD)


A new technical paper "Scaling Routers with In-Package Optics and High-Bandwidth Memories" was posted by researchers at Technion, UC Berkeley and UC San Diego. Abstract "This paper aims to apply two major scaling transformations from the computing packaging industry to internet routers: the heterogeneous integration of high-bandwidth memories (HBMs) and chiplets, as well as in-package optic... » read more

HBM4 Sticks With Microbumps, Postponing Hybrid Bonding


The next generation of high-bandwidth memory, HBM4, was widely expected to require hybrid bonding to unlock a 16-high memory stack. A JEDEC move made that unnecessary with this generation, but it’s merely a postponement, not a cancellation. HBM has been in high demand for AI in data centers — especially for training. Data movement dominates energy consumption, and high-bandwidth memories... » read more

High Bandwidth Memory (HBM): Everything You Need To Know


In an era where data-intensive applications, from AI and machine learning to high-performance computing (HPC) and gaming, are pushing the limits of traditional memory architectures, High Bandwidth Memory (HBM) has emerged as a high-performance, power-efficient solution. As industries demand faster, higher throughput processing, understanding HBM’s architecture, benefits, and evolving role in ... » read more

Critical Factors For Storing Data In DRAM


DRAM is becoming more complicated to develop, and more difficult to manage inside AI data centers. In the past, latency, bandwidth, and capacity were the primary considerations. But as the amount of data that needs to be processed, moved, and stored continues to rise, a whole new set of factors is emerging. Steven Woo, fellow and distinguished inventor at Rambus, talks about latency under load,... » read more

HBM4 Memory: Break Through to Greater Bandwidth


Delivering unrivaled memory bandwidth in a compact, high-capacity footprint, has made HBM the memory of choice for AI training. HBM4 is the fourth major generation of the HBM standard, with new power management and RAS features. The Rambus HBM4 Controller provides industry-leading performance to 10.0 Gb/s, enabling a memory throughput of over 2.5 TB/s for training systems, generative AI and oth... » read more

Scaling DRAM Technology To Meet Future Demands: Challenges And Opportunities


Since the invention of the 1T1C bit cell more than 50 years ago, DRAMs have become the main memory of choice for processors in computer systems and many consumer electronics devices. As new use computing paradigms have been created, including 3D graphics, cloud computing, smart phones, and AI processing, specialized processors and DRAM memories have been developed that are optimized for these u... » read more

What’s Different About HBM4


Memory bandwidth is limiting the flow of huge datasets that are needed to train AI models. There is much more data to process, store, and retrieve, but the speed at which that data moves through high-bandwidth memory (HBM) stacks is significantly lower than the speed at which data can be processed. Frank Ferro, group director for product management at Cadence, talks about the new HBM4 standard,... » read more

LLM Inference: Core Bottlenecks Imposed By Memory, Compute Capacity, Synchronization Overheads (NVIDIA)


A new technical paper titled "Efficient LLM Inference: Bandwidth, Compute, Synchronization, and Capacity are all you need" was published by NVIDIA. Abstract "This paper presents a limit study of transformer-based large language model (LLM) inference, focusing on the fundamental performance bottlenecks imposed by memory bandwidth, memory capacity, and synchronization overhead in distributed ... » read more

HBM Roadmap: Next-Gen High-Bandwidth Memory Architectures (KAIST’s TERALAB)


A new technical paper titled "HBM Roadmap Ver 1.7 Workshop" was published by researchers at KAIST’s TERALAB. The 371-page paper provides an overview of next-generation HBM architectures based on current technology trends, as well as many technology insights. Find the technical paper here or here.  Published June 2025. Advising Professor : Prof. Joungho Kim. Fig. 1: Thermal Manag... » read more

← Older posts