VeriSilicon NPU IP powering 100 million+ AI-enabled chips worldwide
Date: 10/03/2024
VLSI IP vendor VeriSilicon has achieved a significant milestone with its Neural Network Processor (NPU) IP now integrated into over 100 million AI-enabled chips across various sectors worldwide.
VeriSilicon NPU IP is designed for high-performance design, offering a low-power, programmable, and scalable architecture. It can be tailored to different chip sizes and power budgets, making it a cost-effective solution for accelerating neural network computations. Supported by a comprehensive Software Development Kit (SDK), the NPU IP ensures swift product deployment.
The latest VIP9000 series NPU IP provides scalable and high-performance processing capabilities for Transformer and Convolutional Neural Network (CNN) models. With support for major frameworks like PyTorch, ONNX, and TensorFlow, coupled with 4-bit quantization and compression technologies, the IP addresses bandwidth constraints, enabling advanced AI algorithms deployment on embedded devices.
VeriSilicon's FLEXA technology facilitates seamless integration of the VIP9000 with Image Signal Processors (ISPs) and video encoders, forming low-latency AI-ISP and AI-Video subsystems without the need for DDR memory. This customizable solution suits power- and space-constrained environments in deeply embedded applications.
