The rapid advancement of artificial intelligence (AI) and Internet of Things (IoT) technologies has significantly enhanced the intelligence of IoT devices. As AI-powered smart terminals, such as smart home devices and AI toys, gain market traction, system-on-chip (SoC) solutions are poised for new growth opportunities.
Flyingchip, a subsidiary of SmartSens, recently introduced the A1, its first high-performance edge visual AI SoC for AIoT applications. The A1 integrates a high-performance AI image signal processor (ISP), a 0.8 TOPS@INT8 self-developed neural processing unit (NPU), and 1Gb DDR3L memory. It offers superior image processing, efficient edge computing, low power consumption, and compact packaging. The A1 supports multiple CMOS image sensor resolutions, forming a complete edge AI SoC and sensor system. This enables high-precision, low-latency real-time intelligent imaging for consumer and industrial visual modules, accelerating upgrades in applications like AI toys, smart home devices, industrial scanning, and night vision modules.
High-Performance AI ISP for Low-Light Imaging
The A1, Flyingchip¡¯s inaugural AIoT edge visual AI SoC, features a proprietary AI ISP supporting resolutions from 0.3MP to 8MP. Its core image processing capabilities include:
Advanced Low-Light Noise Reduction
The A1¡¯s low-light noise reduction and image enhancement algorithms outperform traditional ISPs, delivering clearer, high-contrast, low-noise images in ultra-low-light conditions. This improves the accuracy and efficiency of image capture and processing for smart visual devices in dark environments, such as outdoor night settings.
Three-Frame HDR Synthesis
Supporting up to three-frame HDR synthesis with a 120dB dynamic range, the A1 mitigates overexposure or loss of detail in challenging lighting conditions, such as strong or backlit scenes, producing detailed and balanced images.
RGB-IR Image Processing
The A1¡¯s RGB-IR ISP module enables simultaneous RGB and IR image output, meeting the needs of robotic depth modules for color image processing and depth calculations, ensuring precision and responsiveness in tasks like motion, navigation, and obstacle avoidance.
Dual Sensor Input and Synchronization
The A1 supports up to two image inputs with synchronized processing, compatible with various sensor types and shutter mechanisms, enabling flexible applications like binocular stereo vision for industrial and consumer multi-camera systems.
Lightweight NPU for Edge Processing
The A1¡¯s 0.8 TOPS@INT8 self-developed NPU supports mainstream CNN architectures, optimized for lightweight neural networks and visual tasks. Compared to similar products, the A1 improves inference efficiency by approximately 97% for classification models and 260% for detection models, enhancing the accuracy and speed of visual data processing in intelligent applications.
The A1 also prioritizes low power consumption and compact packaging. Its advanced low-power technology reduces overall camera module power draw and heat generation. Available in 9mm x 9mm (QFN) and 7mm x 7mm (BGA) packages, the A1 supports miniaturized camera module designs, saving space in smart devices.
Edge AI SoC and Sensor Solutions
To meet the performance and market demands of smart terminals, Flyingchip offers A1-based edge AI SoC and sensor combinations for consumer and industrial visual modules, enabling compact, high-performance intelligent visual applications.
Smart Hardware
In emerging smart hardware like AI pet robots, visual perception camera modules provide precise, real-time image capture and processing for tasks such as navigation and recognition. The A1, paired with 0.3MP-2MP global shutter CMOS sensors, supports single or dual-camera modules. It delivers high-quality RGB or IR imaging and runs visual algorithms for face or object detection, enabling real-time user expression and emotion perception while reducing backend processing loads. Single-camera modules combining video and detection capabilities optimize system design and space in compact devices.
Smart Home
The A1, paired with 3MP-6MP low-power fast-start CMOS sensors and fisheye lenses, forms a wide-field-of-view, low-power visual perception solution. This system quickly activates upon detecting people, using low-light noise reduction, distortion correction, and human detection algorithms to provide fast, accurate recognition. This enables smart home devices like air conditioners or fans to adjust airflow direction and speed based on occupant positions, enhancing user comfort.
Industrial Scanning
For industrial machine vision, the A1 with 0.3MP-2MP global shutter CMOS sensors processes high-frame-rate inputs in real time via its AI ISP, improving image clarity. It supports lightweight algorithms for barcode reading and counting, enabling fast, accurate scanning and detection at the edge, boosting efficiency and cost-effectiveness in automated production lines.
Night Vision Modules
Paired with 4MP-8MP night vision full-color CMOS sensors, the A1 delivers high-performance night vision modules. Its AI ISP optimizes high-quality real-time images for superior low-light performance, providing clear, detailed, low-noise full-color night vision for various camera applications.
The A1 AI SoC is now in mass production and available for sampling.
