VectorMesh™ NPU IP - Inventec AI Center

High performance, low power consumption, and a flexible architecture

「 VectorMesh™ 」 NPU IP

Designed to revolutionize traditional processor architecture, our solution incorporates unique and innovative technologies that optimize the timing of data transfer and enhance resource utilization. This not only achieves a multi-fold increase in computational capabilities but also significantly reduces the gate count by four times compared to other equivalent products under identical processing and computational conditions.

The architecture’s flexibility enables support for computations with different data precisions (8~16 bits) and allows for customized designs based on chip specifications, resulting in a substantial reduction of unnecessary additional resources.

Efficiency Takes Precedence: The architecture is conceived from the outset with a primary focus on achieving exceptional PPA (Power/Performance/Area) metrics, prioritizing efficiency as the foremost consideration. This results in a NPU that combines high performance with low energy consumption.
Tailor-Made Solutions: The flexible architecture features high compatibility and scalability, allowing for customized designs tailored to the specific requirements of the system.

Cost-Effective, Energy-Efficient

Solutions for Diverse, High-Performance AI Applications.

Meticulously developed in-house, our architecture is fully flexible and empowers AI computation with optimized costs and power consumption. Our solution supports a wide range of AI models, including MobileNet, SSD, YOLO, Wav2vec, TDNN, and more, making it highly applicable in various domains such as visual perception, image processing and enhancement, speech processing, and biometric identification. Positioned as the leading NPU IP for edge AI computing, our solution stands out as the industry’s premier choice.

Customized Services: Leveraging 6 years of practical experience from the Inventec AI Center, we offer professional AI model design and training to meet diverse application scenarios.
In the field of computer vision, we excel in object detection, face detection, face recognition, gesture recognition, pose estimation, video analysis, and video inference.
In the field of image processing and enhancement, we specialize in image denoising, image rectification, semantic segmentation, image style transfer, and super resolution.
In the realm of speech processing, our capabilities include speech detection, speech recognition, speech feature analysis, and speech command recognition.
For biometric identification, we provide solutions for human body shape recognition, gait recognition, and more.

One-stop Service

Streamlining Verification Processes for Swift Market Deployment

In the past, following the acquisition of NPU licensing, accessing AI model training and compilation services may have required additional time and manpower. In contrast, the VectorMesh™ Embedded Neural Network Processor provides simultaneous access to a stable and efficient software development toolchain.

From the design and training of AI models to the compilation and integration of these models into SoC chips, our one-stop service facilitates comprehensive system integration. This not only reduces cross-validation time but also enhances cost-effectiveness, accelerating the mass production of AI chips to seize market opportunities.

Prioritizing Efficiency: Strategically focused on efficiency, our one-stop IP integration service expedites system integration processes and effectively mitigates risks.
Optimizing Resources: Conserve valuable human and time resources by entrusting tasks to the Inventec VectorMesh™ development team — renowned for prompt customer support and specialized expertise.