EmbeddingGemma 2: Offline Multimodal Search for Edge Boards
Google's EmbeddingGemma 2 maps text, code, images, video and audio into one vector space in 740M parameters. Here is what it means for embedded Linux boards, how on-device RAG works,…
Alif Semiconductor Acquisition: What ADI’s $1.35B Deal Means
Analog Devices is buying Alif Semiconductor for $1.35 billion, adding Ensemble and Balletto AI microcontrollers with Arm Ethos NPUs. Here's what the deal signals for edge-AI MCUs and the skills…
TI TinyEngine NPU: Edge AI Comes to Cortex-M0+ and M33 MCUs
TI's TinyEngine NPU puts a small fixed neural accelerator into a sub-dollar Cortex-M0+ MCU and a Cortex-M33 motor-control family. Here is what it runs, how to use it, and where…
ESP32-S31 in Mass Production: Linux, Specs and When to Pick It
Espressif's ESP32-S31 is now in mass production and can boot Linux. Here is how the dual-core RISC-V AIoT SoC compares to the ESP32-S3, ESP32-P4 and Raspberry Pi, and when it…
STM32N6 NPU in 2026: Boards, YOLO, and a Getting-Started Guide
The STM32N6's 600 GOPS Neural-ART NPU now has a real ecosystem, from MicroPython and Arduino boards to official YOLO support. Here's what changed in 2026 and how to deploy your…
Arduino VENTUNO Q: Physical AI Robot Arms Running SmolVLA Locally
A $299 Arduino VENTUNO Q ran Hugging Face's SmolVLA locally to drive an SO-101 robot arm. Here's how the dual-brain board and vision-language-action models work, and how it compares with…
BitNet ESP32 Cluster: 1-Bit LLMs Reach $5 Microcontrollers
A seven-board BitNet ESP32 cluster and Cactus Compute's 2-bit Needle model show ultra-low-bit LLMs reaching cheap hardware. Here is how ternary weights work and where the limits are.
LLM on a Microcontroller: How Generative AI Fits in 512 KB SRAM
A 28.9M-parameter LLM now runs on an ESP32-S3 and a diffusion model draws faces on a $1 RP2350. Here is how memory tiering, quantization and DMA streaming make it work,…