Chip Industry Week in Review


The IEEE ISSCC conference was held this week in San Francisco. Among the highlights: IBM detailed an AI accelerator based on its new inferencing dataflow architecture. CEA-Leti presented a chip-scale, ultra-fast, battery-operated EPR spectrometer. QuTech introduced a cryo-CMOS SoC with NV centers in diamond. UTokyo showed its low-jitter PLL architecture for beyond 5G/6G. Imec d... » read more

Blog Review: Feb. 18


Synopsys' Raja Tabet anticipates deployment of an agentic AI workforce within the next 12 to 24 months that can take on different engineering personas, such as a digital implementation agent, a verification agent, or an analog agent, to run experiments in parallel, generate and triage tests, and propose fixes. Cadence's Reela Samuel dives into power usage effectiveness in data centers and wh... » read more

Chip Industry Week In Review


Geopolitics U.S. lawmakers are urging tighter export controls on advanced semiconductor manufacturing equipment (SME) to China, warning existing loopholes threaten national security. "China is working to build domestic SME by exploiting access to U.S. and allied subcomponents required to produce tools," states the letter, which also says better coordination with allies is essential. The U.S.... » read more

UCIe’s Major Technical Components Are Now In Place


Key Takeaways UCIe 3.0 doubles bandwidth and enhances manageability, addressing new use cases and following an annual update cycle since 2023. The growing demand for chiplet-based architectures in AI data centers is driven by the limitations of monolithic chips, making inter-chiplet communication and connectivity crucial. While UCIe was initially seen as feature-heavy, many of its ma... » read more

Scaling llama.cpp On Neoverse N2: Solving Cross-NUMA Performance Issues


This blog post explains the cross-NUMA memory access issue that occurs when you run llama.cpp in Neoverse. It also introduces a proof-of-concept patch that addresses this issue and can provide up to a 55% performance increase for text generation when you run the llama3_Q4_0 model on the ZhuFeng Neoverse system. Cross-NUMA memory access problem In llama.cpp, performance drops when the number o... » read more

5 Strategic Decisions for Building a Scalable Compute Platform for Now and the Future


Artificial intelligence (AI) is no longer a “nice-to-have” technology—it’s a central driver of competitive advantage and business innovation. Across industries, enterprises are moving beyond experimentation and embedding AI into all their products, workflows, and customer experiences. But as organizations scale, many are discovering a stark reality: their compute infrastructure was not ... » read more

Chip Industry Week in Review


Intel hired ex-Qualcomm GPU guru Eric Demers for the company's high-performance GPU push, setting the stage for a three-way battle with Nvidia and AMD. The key targets for Intel and AMD will be better power efficiency and a programming model that rivals CUDA, but don't expect Nvidia to stand still. Acquisitions Texas Instruments plans to acquire Silicon Labs for ~$7.5B cash to enhance i... » read more

Blog Review: Feb. 4


Siemens' Tova Levy examines thermal management in 3D-IC, including why heat behaves differently in vertical stacks and how to analyze and manage thermal risk earlier and more predictably to ensure a design can meet performance, reliability, and time-to-market targets. Cadence's Reela Samuel finds that known-good-die strategies, standardized die-to-die test access, and vertical reliability mo... » read more

Blog Review: Jan. 28


Synopsys' Dana Neustadter and Vincent van der Leest argue that a hardware-based approach to security is required to fully address the risks introduced by modern AI architectures and the distributed workloads they support. Siemens EDA's Tova Levy examines multiphysics challenges in 3D-IC designs and outlines three design imperatives to identify risks earlier and support PPA, reliability, and ... » read more

Blog Review: Jan. 21


Keysight's Armando Valim considers the impact of AI on the memory market as AI infrastructure pressure widens the gap between high-performance memory and lower-margin consumer memory and SSD, forcing manufacturers to make strategic decisions and define which markets to serve. Cadence's Reela Samuel breaks down the major 3D-IC packaging methods used today, from wafer stacking flows to hybrid ... » read more

← Older posts Newer posts →