WiseEye^TM

View the repository on GitHub

Ultra low Power Endpoint AI Processors

Himax offers DRAM-less, ones-to-tens mill-watt AI processors to perform real time analytics in the endpoint itself.

The processors are designed to enable battery-powered intelligence vision, voice, and vibration applications.

WiseEye 2 AI Processor (WE2)

WE2 provides up to 50 GOPS with asymmetric multiprocessing architecture and DVFS power management.

Maximum ML computing

- High performance 400MHz Cortex-M55

- High efficiency 150MHz Cortex-M55

- microNPU 400MHz Ethos-U55

- Internal 2MB SRAMs and 512KB TCM

- External 100MHz QSPI Flash, Max 16MB(128Mb)

Optimum energy efficiency

- High-efficiency DC-DC

- Internal power islands

- DVFS power management

System-wide security

- PUF-based hardware root-of-trust

- PUF-based unique device ID

- PUF-based TRNG

- CryptoCell-312 crypto engine

- TrustZone security

- Secure debug

Rich peripherals

- MIPI CSI RX and TX

- DVP and SDI

- PDM and I²S

- SD and SDIO

- SPI, I²C, and I³C

- UART, PWM and GPIO

Power Consumption Measurement - Power Modes

Note:

1. Measurement condition is in room temperature.

2. Facial landmark demo video showed in https://himaxwiseeyeplus.github.io/

3. Additional an on-board power switch needed to disable WE2 power source, except VDDIO_AON and VDDA_DCDC sources for GPIO wake up.

Chip Package WLCSP65 (2.3mm x 5.6mm)

Open-Source Unified Software Toolchain

An open-source unified toolchain for Ethos-U55 and Cortex-M55 simplifies developer creation of intelligence applications.

Optimized software flow for embedded and ML

- TensorFlow Lite for Microcontrollers

- Open-source Vela NN optimizer tool for Ethos-U55

- Open-source CMSIS-NN library for Cortex-M55

- Unified software flow for processor and microNPU integration

Unified software flow to accelerate ML performance

- NN operators are accelerated by Ethos-U55 driver by default

- Fallback to CMSIS-NN, then reference kernel by Cortex-M55

Himax WE2 board kit

Himax WE2 board kit information

- Board kit online sales: https://www.seeedstudio.com/Grove-Vision-AI-Module-V2-p-5851.html

- Board kit wiki: https://wiki.seeedstudio.com/grove_vision_ai_v2/

- AI model assistant (Seeed’s SenseCraft ): https://wiki.seeedstudio.com/grove_vision_ai_v2_software_support/#-no-code-getting-started-with-sensecraft-ai-

- Vision-based AI examples: Connect to open-source and web-based cloud: https://wiki.seeedstudio.com/connect_vision_ai_v2_to_ha/

- Vision-based AI examples: Connect to phone-based app (Seeed’s SenseCap Mate App, iOS/Android): https://wiki.seeedstudio.com/connect_vision_ai_v2_to_ha/

Demo Videos

These demo videos are captured from WE2 silicon.

Facial landmark (12fps)

Human pose (7fps)

Face Mesh (7fps)

YoloV8n (9fps)

YoloV8n Pose(10fps)