Flow Computing’s Post

Arm Steps Deeper into Silicon: Implications for the Semiconductor Value Chain Arm is moving beyond IP licensing into full chip production with its AGI CPU. Discover its impact on the semiconductor value chain. https://lnkd.in/gpfCcF66

Hi folks! The "AI everywhere" trend just hit the entry-level x86 market, and it’s a massive win for those of us working with legacy industrial hardware. Congatec just dropped the conga-TC300, a COM Express module that finally brings a dedicated NPU to the Atom and Celeron-class space. Here’s the lowdown on why this is a major engineering shortcut: 𝗡𝗣𝗨 𝗳𝗼𝗿 𝘁𝗵𝗲 𝗠𝗮𝘀𝘀𝗲𝘀: We’re moving beyond relying solely on the CPU or GPU for inference. This module delivers up to 41 TOPS of total AI performance (with 18 TOPS coming straight from the NPU). It’s perfect for adding object recognition or voice processing without cooking your processor. 𝗧𝗵𝗲 𝗨𝗹𝘁𝗶𝗺𝗮𝘁𝗲 "𝗗𝗿𝗼𝗽-𝗶𝗻" 𝗨𝗽𝗴𝗿𝗮𝗱𝗲: Since it uses the standard COM Express Compact Type 6 form factor, you can theoretically swap out an old Intel Atom or Celeron module and upgrade your existing carrier board with minimal redesign effort. 𝗛𝘆𝗯𝗿𝗶𝗱 𝗔𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁𝘂𝗿𝗲: It’s using Intel Core Series 3 chips (2 Performance cores + 4 Efficient cores). This balance is great for handling real-time tasks on the "P" cores while letting the "E" cores manage background gateway functions. 𝗡𝗲𝘅𝘁-𝗚𝗲𝗻 𝗜/𝗢 & 𝗠𝗲𝗺𝗼𝗿𝘆: Support for 64 GB of DDR5 (6400 MT/s) with in-band ECC, plus USB4 and 2.5 GbE with TSN support for those time-sensitive industrial apps. 𝗦𝗼𝗳𝘁𝘄𝗮𝗿𝗲-𝗥𝗲𝗮𝗱𝘆: Through their "aREADY . COM" ecosystem, you can get preconfigured OS options like Ubuntu Pro or ctrlX OS with virtualization support. This lets you run HMI, AI, and real-time control in isolated "zones" on a single module. For firmware and systems engineers, this is a total lifesaver for extending the lifecycle of industrial edge devices without a full architectural overhaul. It’s basically a massive boost in "brainpower" while keeping the same thermal footprint. Are you planning to offload your edge AI tasks to a dedicated NPU this year, or are you still pushing the limits of your integrated GPUs? Read the full news at https://cstu.io/67fa20 and let’s swap notes in the comments! #EmbeddedSystems #IndustrialComputing #EdgeAI #x86 #Congatec #IntelCore #FirmwareEngineering

Semiconductor Insights Evolution of Chip Architecture From monolithic designs to chiplet-based architectures, the semiconductor industry is rapidly evolving to overcome scaling challenges and improve performance. What’s happening here? This image shows the transition from: Traditional Monolithic Chips 1st Gen Chiplet-Based CPUs Advanced Hybrid Multi-Die Architectures Why this matters? • Monolithic designs face yield and scaling limitations • Chiplets improve modularity, scalability, and cost efficiency • Advanced architectures separate compute (CCD) and I/O dies for better optimization This shift is powering modern high-performance processors used in servers, AI, and data centers. #Semiconductor #VLSI #ChipDesign #Chiplets #AMD #ProcessorDesign

AMD and Intel have partnered on ACE, a standardized matrix‑acceleration architecture for x86 CPUs aimed at boosting AI performance across vendors, signalling rare alignment between rivals as AI workloads demand broader ecosystem compatibility. #AMD #Intel #TechGiants #ChipMaker #AIAcceleration #x86 #Chips #Semiconductors #Processors #CPU #CPUs #Innovation #Technology #AI #TechnologyNews https://lnkd.in/dcAf2iYC

SiPearl has appointed Guillaume Etorre as COO ahead of the Rhea1 CPU launch, strengthening operations around processor design, reference systems, quality, and industrialisation. #Processors #HPC #Semiconductors

𝐀𝐑𝐌 𝐂𝐨𝐫𝐭𝐞𝐱 𝐌 𝐒𝐞𝐫𝐢𝐞𝐬 𝐂𝐏𝐔 𝐒𝐭𝐚𝐜𝐤 𝐃𝐞𝐬𝐢𝐠𝐧(𝐒𝐓𝐌32) Most people see a microcontroller as something that just runs code line by line. But inside an ARM Cortex M series CPU, there is a very interesting design that quietly makes real time systems possible. 𝐋𝐞𝐭’𝐬 𝐢𝐦𝐚𝐠𝐢𝐧𝐞 𝐭𝐡𝐢𝐬 𝐰𝐢𝐭𝐡 𝐚 𝐬𝐢𝐦𝐩𝐥𝐞 𝐬𝐭𝐨𝐫𝐲 When your embedded system is running normally, it is in a calm working state. In ARM terms, 𝐭𝐡𝐢𝐬 𝐢𝐬 𝐜𝐚𝐥𝐥𝐞𝐝 𝐓𝐡𝐫𝐞𝐚𝐝 𝐌𝐨𝐝𝐞. This is where your main application runs. Your FreeRTOS tasks also execute here. Now imagine something urgent happens. A timer interrupt fires, or a UART receives data, or a system exception occurs. At this moment, the CPU cannot continue normal work. It immediately switches to a special emergency handling state 𝐜𝐚𝐥𝐥𝐞𝐝 𝐇𝐚𝐧𝐝𝐥𝐞𝐫 𝐌𝐨𝐝𝐞. So now the question is how does the CPU manage memory safely between normal work and urgent events 𝐓𝐡𝐢𝐬 𝐢𝐬 𝐰𝐡𝐞𝐫𝐞 𝐭𝐡𝐞 𝐬𝐦𝐚𝐫𝐭 𝐩𝐚𝐫𝐭 𝐜𝐨𝐦𝐞𝐬 𝐢𝐧 ARM Cortex M uses two separate stack pointers. One is Main Stack Pointer, 𝐌𝐒𝐏. The other is Process Stack Pointer, 𝐏𝐒𝐏. Think of MSP as a system level stack. It is mainly used when the CPU is handling interrupts or exceptions in 𝐇𝐚𝐧𝐝𝐥𝐞𝐫 𝐌𝐨𝐝𝐞. Think of PSP as an application stack. It is used when your tasks and normal application code run in 𝐓𝐡𝐫𝐞𝐚𝐝 𝐌𝐨𝐝𝐞. 𝐍𝐨𝐰 𝐡𝐞𝐫𝐞 𝐢𝐬 𝐭𝐡𝐞 𝐢𝐧𝐭𝐞𝐫𝐞𝐬𝐭𝐢𝐧𝐠 𝐩𝐚𝐫𝐭 When an interrupt occurs, the CPU automatically saves important registers like R0 to R3, R12, LR, PC, and xPSR into the current stack before switching context. This ensures nothing gets lost. In FreeRTOS based systems, each task typically uses its own PSP. So when the CPU switches from one task to another, it saves the current task context using PSP and restores the next task context using its PSP. At the same time, interrupts always use MSP, so they remain isolated from task stacks and do not interfere with application execution. This clean separation between Thread Mode and Handler Mode, and between MSP and PSP, is what makes ARM Cortex M very stable for real time systems. It allows safe multitasking, predictable interrupt handling, and reliable system behavior even under heavy load. Without this dual stack design, managing real time systems would be far more complex and error prone. This architecture is one of the key reasons ARM Cortex M series is widely used in embedded and real time applications. Interestingly, other systems like ESP32 also run FreeRTOS, but their internal CPU architecture handles stacks differently and does not follow the same MSP and PSP separation model. With one year of experience in firmware development, I am currently exploring new opportunities to grow in embedded systems. Balemarthy Vamsi Krishna #opentowork #firmware

ARM’s CEO says agentic AI could push CPU core counts as high as 512, arguing that traditional GPU‑to‑CPU ratios are becoming less relevant as orchestration and autonomy workloads expand, highlighting a potential rebalancing of compute architectures. #ARM #AgenticAI #Chips #Processors #CPU #CPUs #Datacenters #Innovation #Technology #TechNews https://lnkd.in/gdaBFi24

Siemens’ Veloce Strato CS has supported verification of Arm’s AGI CPU, showing the scale of pre-silicon validation now required for AI compute platforms. #EDA #semiconductors #AI

China has unveiled a supercomputer capable of 2‑exaflops, which is built entirely with CPUs, packing 47,000 processors into 92 cabinets, as it looks to leapfrog US systems without relying on GPUs. #China #Tech #SuperComputers #Supercomputing #HPC #CPUs #Innovation #Technology #TechnologyNews https://lnkd.in/d7PTWUTk

Loongson Technology Corporation Limited Ships 1 Million 3A6000 CPUs in China Chip Push🚀 Chinese chipmaker Loongson Technology has shipped more than 1 million units of its 3A6000 desktop processor, marking a significant milestone in China’s drive for semiconductor self-reliance amid tightening US export restrictions. ✨ Highlights: ✔️ 3A6000 uses Loongson’s proprietary LoongArch instruction set and domestic supply chain ✔️ Chip performance is comparable to Intel desktop CPUs from around 2020 ✔️ Adopted under China’s government-backed XinChuang IT replacement program Read More: https://lnkd.in/d4XTu9Qa #Loongson #Semiconductors #ChinaTech #CPU #LoongArch