Sipeed M1s & M0sense – Low-cost BL808 & BL702 based AI modules (Crowdfunding)

Sipeed M1s & M0Sense

Sipeed has launched the M1s and M0Sense AI modules. Designed for AIoT application, the Sipeed M1s is based on the Bouffalo Lab BL808 32-bit/64-bit RISC-V wireless SoC with WiFi, Bluetooth, and an 802.15.4 radio for Zigbee support, as well as the BLAI-100 (Bouffalo Lab AI engine) NPU for video/audio detection and/or recognition. The Sipeed M0Sense targets TinyML applications with the Bouffa Lab BL702 32-bit microcontroller also offering WiFi, BLE, and Zigbee connectivity. Sipeed M1s AIoT module The Sipeed M1S is an update to the Kendryte K210-powered Sipeed M1 introduced several years ago. Sipeed M1s module specifications: SoC – Bouffalo Lab BL808 with CPU Alibaba T-head C906 64-bit RISC-V (RV64GCV+) core @ 480MHz Alibaba T-head E907 32-bit RISC-V (RV32GCP+) core @ 320MHz 32-bit RISC-V (RV32EMC) core @ 160 MHz Memory – 768KB SRAM and 64MB embedded PSRAM AI accelerator – NPU BLAI-100 (Bouffalo Lab AI engine) for video/audio detection/recognition delivering up […]

Quadric Chimera GPNPU IP combines NPU, DSP, and real-time CPU into one single programmable core

A typical chip for AI or ML inference would include an NPU, a DSP, a real-time CPU, plus some memory, an application processor, an ISP, and a few more IP blocks. Quadric Chimera GPNPU (general purpose neural processor unit) IP combines the NPU, DSP, and real-time CPU into one single programmable core. According to Quadric, the main benefit of such design is simplifying system-on-chip (SoC) hardware design and subsequent software programming once the chip is available thanks to a unified architecture for machine learning inference as well as pre-and-post processing. Since the core is programmable it should also be future-proof. Three “QB series” Chimera GPNPU cores are available: Chimera QB1 – 1 TOPS machine learning, 64 GOPS DSP capability Chimera QB4 – 4 TOPS ML, 256 GOPS DSP Chimera QB16 – 16 TOPS ML, 1 TOPS DSP Quadric says the Chimera cores can be used with any (modern) manufacturing process […]

TinyML-CAM pipeline enables 80 FPS image recognition on ESP32 using just 1 KB RAM

The challenge with TinyML is to extract the maximum performance/efficiency at the lowest footprint for AI workloads on microcontroller-class hardware. The TinyML-CAM pipeline, developed by a team of machine learning researchers in Europe, demonstrates what’s possible to achieve on relatively low-end hardware with a camera. Most specifically, they managed to reach over 80 FPS image recognition on the sub-$10 ESP32-CAM board with the open-source TinyML-CAM pipeline taking just about 1KB of RAM. It should work on other MCU boards with a camera, and training does not seem complex since we are told it takes around 30 minutes to implement a customized task. The researchers note that solutions like TensorFlow Lite for Microcontrollers and Edge Impulse already enable the execution of ML workloads, onMCU boards, using Neural Networks (NNs). However, those usually take quite a lot of memory, between 50 and 500 kB of RAM, and take 100 to 600 ms […]

Easily add face detection to your project with the Person Sensor module

It’s now much easier to AI features to your project thanks to better tools, but as we’ve experienced when trying out Edge Impulse machine learning platform on the XIAO BLE Sense board, it still requires some effort and the learning curve may be higher than some expect. But for common tasks like face detection, there’s no reason for the solution to be hard-to-use or expensive, and Pete Warden (Useful Sensors) has designed the $10 Person Sensor fitted with a camera module pre-programmed with algorithms that detect nearby faces and reports the results over an I2C interface.   Person Sensor specifications: ASIC – Himax HX6537-A ultra-low-power AI accelerator @ 400 MHz with 2MB SRAM, 2MB flash Camera Image Sensor – 110 degrees FOV Image scan rate – 7Hz with no facial recognition Image scan rate – 5Hz with facial recognition active Host interface Qwiic connector for the I2C interface up to […]

Google KataOS – A secure OS for embedded systems written in Rust (mostly)

Google Research has been working on its own Rust-based operating system called KataOS and designed to secure embedded systems that run Machine Learning (ML) applications. There has been a lot of talk about the Rust programming language in recent times, since it offers about the same level of performance as C programming but helps programmers write more secure code with built-in prevention against buffer overflows for instance. It has gained a lot of traction over the years, and Linux 6.1 will be the first kernel release to include Rust code. Google Research noticed that system security is often treated as a feature that can be added to existing systems either by software or an extra security chip. But in a world, where more and more of our private data is exposed to the world through the Internet, it is not good enough, so the company developed KataOS open-source, secure operating […]

Nuvoton NuMicro MA35D1 Arm Cortex-A35/M4 microprocessor to power Linux edge IIoT gateways

Novoton NuMicro MA35D1 microprocessor features two Arm Cortex-A35 cores, one Arm Cortex-M4 real-time core, and two Ethernet interfaces for Linux-based edge IIoT gateway. The SoC also is offered in variants supporting external DDR memory or integrated up to 512MB RAM, 154 or 208 GPIOs, and an optional “Enhanced ADC”. The MA35D1 also comes with a TFT interface for up to 1920×1080 displays, several hardware security features, and the company says the microprocessor facilitates Tiny AI/ML for edge computing despite not integrating an AI accelerator. Nuvoton NuMicro MA35D1 key features and specifications: CPU sub-system 2x Cortex-A35 cores running at up to 800 MHz Cortex-M4 real-time core at up to 180 MHz Memory sub-system On-chip 384 KB SRAM (Cortex-A35 256 KB + Cortex-M4 128 KB) Multi-Chip Package (MCP) DDR up to 512MB External DDR interface for MA35D16A087C SKU Storage Quad SPI NAND Flash Controller Secure Digital Host Controller (SDHC) Display and Video […]

$7 Lolin S3 ESP32-S3 board ships with MicroPython firmware

Lolin S3 is the first ESP32-S3 board from the company, but instead of using the more compact D1 mini form factor, the board features a longer design with two rows of 20 pins offering up to 31 GPIOs. Based on ESP32-S3-WROOM-1 module, the board features 16MB QSPI flash, 8MB SPRAM, two USB Type-C OTG and UART ports, a Lolin I2C port, an RGB LED, as well as Reset and user buttons. Lolin S3 specifications: Wireless module – ESP32-S3-WROOM-1 module with: Espressif Systems ESP32-S3 dual-core Tensilica LX7 @ up to 240 MHz with vector instructions for AI acceleration, 512KB RAM, 2.4 GHz WiFi 4 and Bluetooth 5.0 LE with support for long-range, up to 2Mbps data rate, mesh networking 16MB QSPI flash 8MB PSRAM PCB antenna USB – 2x USB Type-C ports, one OTG port, one UART port for programming and debugging Expansion 2x 20-pin headers with up to 31x GPIO, […]

TinyMaix is a lightweight machine learning library for microcontrollers

Sipeed TinyMaix open-source machine learning library is designed for microcontrollers, and lightweight enough to run on a Microchip ATmega328 MCU found in the Arduino UNO board and its many clones. Developed during a weekend hackathon, the core code of TinyMax is about 400 lines long, with a binary size of about 3KB, and low RAM usage, enabling it to run the MNIST handwritten digit classification on an ATmega320 MCU with just 2KB SRAM and 32KB flash. TinyMax highlights Small footprint Core code is less than 400 lines (tm_layers.c+tm_model.c+arch_O0.h), code .text section less than 3KB Low RAM consumption, with the MNIST classification running on less than 1KB RAM Support INT8/FP32 model, convert from keras h5 or tflite. Support multi-architecture acceleration: ARM SIMD/NEON, MVEI, RV32P, RV64V (32-bit & 64-bit RISC-V vector extensions) User-friendly interfaces, just load/run models Supports full static memory config MaixHub Online Model Training support coming soon Sipeed says there […]