Tech Shifts You Can’t Ignore: AI, Embedded & Industrial Insights

At Tech Teamz, we keep a close eye on groundbreaking innovations in embedded engineering, AI acceleration, and next-gen computing, as we are always thrilled to support these companies in their growth.

Let’s do a quick recap!

GenAI NPU: The AI Accelerator That’s Beating Nvidia, Intel, and Google

Spanish startup RaiderChip has introduced the GenAI NPU, a fully hardware-based accelerator designed to enhance generative AI inference at the edge. The novelty about this is that GenAI NPU integrates all large language model (LLM) operations into hardware, eliminating processing latency and ensuring real-time performance.

Key features and performance

• Achieves 2.4× faster token generation per memory bandwidth by reducing hardware-software communication latency.
• Uses cost-effective DDR/LPDDR memory instead of high-bandwidth memory (HBM), cutting costs and energy consumption.
• Outperforms major AI accelerators—37% more efficient than Intel’s Gaudi 2, 28% better than Nvidia’s cloud GPUs, and 25% ahead of Google’s TPU v5e.
• Handles both full-precision (FP32, FP16, BF16, FP8) and quantized (Q4_K, Q5_K) models, boosting inference speed by 276% while reducing memory usage by 75%.
• Supports FPGA and ASIC implementations, allowing customization for different AI workloads.

The GenAI NPU builds on RaiderChip’s GenAI v1, a hardware accelerator optimized for transformer-based models like Meta’s Llama 2/3 and Microsoft’s Phi series. This previous version offloaded 99.99% of inference computations to hardware, relying on a lightweight software layer (<200 KB) that functions autonomously.

Is this a step toward scalable AI acceleration?

By eliminating reliance on cloud-based inference, RaiderChip aims to enable high-performance AI processing in environments where latency, security, and energy efficiency are critical. Their hardware-first approach could drive broader adoption of generative AI across embedded and industrial applications.

ArcelorMittal: EU needs to reinforce trade protections for European industry

ArcelorMittal is urging the EU to strengthen trade protections, increasing support for environmentally conscious investments. High costs and cheap imports are straining the European steel industry.??

ArcelorMittal paused parts of its decarbonization plan due to high energy costs and weak demand. While sales of its XCarb low-carbon steel doubled, they remain a small fraction of total shipments. The company sees government support as crucial for modernization efforts and may seek EU Green Deal funding.

The European Commission recently announced a Steel and Metals Action Plan to adjust Green Deal regulations while maintaining climate goals, with a full plan expected in spring 2025.

Toshiba's New Motor Control Tech Raises the Bar

Toshiba has launched new motor-control solutions, including Arm Cortex-M4-based microcontrollers and a 50-V brushed DC motor driver IC, to enhance efficiency and precision in high-power applications. These additions simplify system design and support industrial electrification trends.

The M470 and M4K microcontrollers offer advanced features such as floating-point units, memory protection, and Toshiba’s programmable motor driver. The M470 series supports dual-motor control, while M4K devices provide additional I2C and amplifier features, catering to diverse motor control needs.

Toshiba’s TB67H482FNG DC motor driver delivers precise current regulation with multiple torque levels and built-in safety features. With these innovations, Toshiba aims to meet growing demands for high-performance motor-driven applications.

At Tech Teamz, we continue to track these advancements, exploring how they impact the embedded and industrial AI landscape. Stay tuned as we analyze what’s next in AI acceleration and edge computing.