Serial Wire Debug (SWD) Protocol: Efficient Debug Interface For Arm-Based Systems


Modern embedded systems are becoming increasingly compact, power efficient, and feature rich. As SoCs integrate more functionality, developers need reliable debug access without increasing pin count or board complexity. Serial Wire Debug (SWD) addresses these needs by providing a streamlined alternative to JTAG, enabling high performance debug features using only two pins, making it ideal for t... » read more

Customizing Foundation IP For Ultra-Low-Voltage Designs


By Daryl Seitzer, Andrew Appleby, and Mohammad Tanveer Building a new system-on-chip (SoC) starts with assembling the right foundational elements—pre‑verified IP for logic, memory, I/O, and other essential functions. Standard IP solutions typically address most common design needs, but some projects call for more specialized approaches, especially when innovation is critical or when t... » read more

The On-Device LLM Revolution


The AI world is experiencing a fundamental shift. After years of cloud-centric inference dominated by massive data center GPUs, we're witnessing an accelerating migration of language models to edge devices. These are not the trillion-parameter behemoths that require server farms, but the "Goldilocks zone" models: 3B to 30B parameters — large enough to deliver genuinely useful AI capabilities,... » read more

AI Inference Needs A Mix-And-Match Memory Strategy


AI inference is no longer a single workload that can be served efficiently by a single type of accelerator or memory. From fast chat replies to 10M token codebases, inference spans wildly diverse workloads with very different limits on latency, bandwidth, capacity, and compute, as the figure below demonstrates.1 Source: Meta1 The AI inference spectrum of workloads includes: Inter... » read more

Formal Verification Fundamentals Remain Non-Negotiable In The New Verification Revolution


The semiconductor industry stands at a critical juncture. First-time silicon success rates have reached all-time lows, while design complexity continues to grow exponentially. System-on-chip designs now integrate billions of transistors, multiple processor cores, complex memory hierarchies, and sophisticated interconnect fabrics. In this environment, the stakes for verification accuracy have ne... » read more

AI, GPU, And HPC Data Centers: The Infrastructure Behind Modern AI


Artificial intelligence (AI) is stretching compute infrastructure well beyond what traditional enterprise data centers were designed to handle. Modern AI training requires massively parallel compute, low-latency networking, high-throughput storage pipelines, and facility engineering that can safely support higher rack power densities than legacy environments. These demands are fueling the eme... » read more

Power Leadership At 2nm: Foundation IP Optimized For Next-Gen Hyperscale SoCs


By Andrew Appleby and Daryl Seitzer As demand for data center compute accelerates, power efficiency has become the defining metric for modern CPUs, GPUs, and AI accelerators. Every watt saved directly impacts the massive operating costs of gigawatt-scale AI data centers, where power and cooling account for 40–60% of operational expenditures. To reduce energy consumption and strengthen t... » 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

Engineering After Orthogonalization: Why Verification Has Become A Lifecycle Discipline


Over the past several decades studying verification practices across the semiconductor industry, I’ve watched assumptions that once held up remarkably well begin to strain under the weight of modern system complexity. This is not a loss of engineering rigor. It is the result of systems that no longer conform to the boundaries earlier design models depended on. For much of the industry’s ... » read more

The Design Challenges Of Clock Integrity And Clock Jitter


Signal integrity is one of the many challenges faced by chip designers. Deep submicron technologies are unfriendly hosts for the nice, clean signals desired. The culprits that compromise signal integrity and introduce jitter include thermal effects, manufacturing flaws, signal crosstalk, IR (voltage) drop, signal loss over long runs, reflections, electromagnetic interference (EMI), ground bounc... » read more

← Older posts Newer posts →