Where Power Is Spent In HBM


HBM is gaining ground because of a spike in the amount of data that needs to be processed quickly, but big reductions in power are possible if that processing can be moved closer to the HBM modules, and if more can be done in each compute cycle without sending data back and forth to memory as frequently. Steven Woo, fellow and distinguished engineer at Rambus, talks about what can be done to bo... » read more

Improving Performance And Power With HBM3


HBM3 swings open the door to significantly faster data movement between memory and processors, reducing the power it takes to send and receive signals and boosting the performance of systems where high data throughput is required. But using this memory is expensive and complicated, and that likely will continue to be the case in the short term. High Bandwidth Memory 3 (HBM3) is the most rece... » read more

Efficiently Process Large RM Datasets In Underlying Memory Pool, Disaggregated Over CXL (KAIST)


A technical paper titled "Failure Tolerant Training with Persistent Memory Disaggregation over CXL" was published (preprint) by researchers at KAIST and Panmnesia. "TRAININGCXL can efficiently process large-scale recommendation datasets in the pool of disaggregated memory while making training fault tolerant with low overhead," states the paper. Find the technical paper here. or here (IEE... » read more

Choosing The Correct High-Bandwidth Memory


The number of options for how to build high-performance chips is growing, but the choices for attached memory have barely budged. To achieve maximum performance in automotive, consumer, and hyperscale computing, the choices come down to one or more flavors of DRAM, and the biggest tradeoff is cost versus speed. DRAM remains an essential component in any of these architectures, despite years ... » read more

Ferroelectric Memories: The Middle Ground


The first article in this series considered the use of ferroelectrics to improve subthreshold swing behavior in logic transistors. The prospects for ferroelectrics in logic applications are uncertain, but ferroelectric memories have clear advantages. The two most common commercial memories lie at opposite ends of a spectrum. DRAM is fast, but requires constant power to maintain its informat... » read more

FPGA-Based Prototyping Framework For Processing In DRAM (ETH Zurich & TOBB Univ.)


A technical paper titled "PiDRAM: A Holistic End-to-end FPGA-based Framework for Processing-in-DRAM" was published by researchers at ETH Zurich and TOBB University of Economics and Technology. Abstract "Processing-using-memory (PuM) techniques leverage the analog operation of memory cells to perform computation. Several recent works have demonstrated PuM techniques in off-the-shelf DRAM dev... » read more

Choosing The Right Memory At The Edge


As the amount of data produced by sensors in cars and phones continues to grow, more of that data needs to be processed locally. It takes too much time and power to send it all to the cloud. But choosing the right memory for a particular application requires a series of tradeoffs involving cost, bandwidth, power, which can vary greatly by device, application, and even the data itself. Frank Fer... » read more

Week In Review: Semiconductor Manufacturing, Test


SEMI , SEMI Europe and European Commission representatives, in consultation with semiconductor industry stakeholders, proposed initiatives to overcome the skills shortage in Europe’s microelectronics industry: Create an industry image campaign to raise public awareness on how technology is shaping the future, and how workers can establish careers in the semiconductor industry. Remove ... » read more

Week In Review: Design, Low Power


RISC-V The European Union said it will spend the equivalent of $286.5 million on a high performance computing ecosystem based on RISC-V. According to the call for proposals, the aim of the project is to “establish a partnership between the EuroHPC JU and a consortium of industry, research organizations and institutions in HPC to the development of innovative HPC hardware and software technol... » read more

Rowhammer Mitigation: In-DRAM Mechanism Scaling The Number of Refreshes With Activations (ETH Zurich)


A technical paper titled "REGA: Scalable Rowhammer Mitigation with Refresh-Generating Activations" was written by researchers at Computer Security Group (COMSEC), ETH Zurich and Zentel Japan. The paper will be presented at IEEE's Symposium on Security and Privacy in May 2023. "With REGA, we propose the first fully in-DRAM mitigation capable of protecting devices independently from their blas... » read more

← Older posts Newer posts →