It’s no secret that CEVA has been pushing the envelope for over a decade to make intelligent machine vision a viable possibility in mass market embedded devices. With four generation of successful, widely adopted DSP cores behind us, including the award-winning CEVA-XM4, we have established our position as the industry leader in low-power, high-performance programmable IP imaging and vision engines.
Today, we are especially excited to unveil our fifth generation imaging and vision platform, delivering unprecedented performance thanks to cutting edge enhancements and innovations. Based on the new CEVA-XM6 DSP core, our latest platform makes it easier, faster, and lower-risk than ever to efficiently harness the power of neural networks and machine vision for smartphones, autonomous vehicles, surveillance, robots, drones and other camera-enabled smart devices.
All the Required Pieces for Deep Learning and AI on Embedded Systems
With the powerful and efficient CEVA-XM6 DSP at the heart of the solution, this platform includes all the critical components for a superior, intelligent vision engine. The platform includes the CEVA Deep Neural Network (CDNN) toolkit, function-specific accelerators, and a comprehensive software package, including OpenCV, OpenCL and OpenVX APIs, CEVA-CV computer vision library and a set of optimized, widely used algorithms. The CDNN toolkit is built on the highly-acclaimed CDNN2 software framework, including advanced neural network (NN) software and a network generator which ports NNs to an embedded environment at the push of a button. The toolkit is further augmented by the powerful new CDNN hardware accelerators built to complement the CEVA-XM6, together achieving the best combination of processing power, software flexibility and low power consumption.
Breakthrough Performance For Neural Networks and Advanced Computer Vision
Built on the strong foundations of the CEVA-XM4 and CEVA-MM3101 processors with over twenty-five design wins to date, the CEVA-XM6 introduces innovations and enhancements enabling unprecedented performance and surpassing anything we’ve done before. Compared to the previous generation CEVA-XM4 intelligent vision DSP, the new CEVA-XM6-based vision platform delivers up to 8x higher performance for neural network workloads and up to 3x performance improvement across all computer vision kernels. Key enhancements introduced in the new architecture include a new vector and scalar processing units and substantial enhancements to instruction set, memory bandwidth and direct memory access (DMA).
Our new vision platform also further extends CEVA’s performance advantage over leading GPU-based architectures when implementing neural networks. Compared to a leading GPU-based embedded system for computer vision and deep learning, our latest imaging and vision platform delivers more than 25x performance-per-watt efficiency and 4x faster processing for convolutional neural networks (CNNs) such as AlexNet and GoogLeNet.
The Future of Embedded Artificial Intelligence
CEVA is at the forefront of bringing superior machine intelligence to embedded systems, and to mass market consumer devices. Our new vision platform will assist in enabling a vast array of applications, from enhancements in computational photography, like the technology used in dual camera smartphones, machine vision for features like augmented reality and depth mapping, human machine interface enabling face recognition and eye tracking, and automotive applications that are revolutionizing mobility and transportation. These applications are bringing the future within reach and helping to realize our vision for a smarter, connected world.
You are more than welcome to find out more about the CEVA’s new vision platform:
- Join our live webinar on November 16th 2016: “Challenges of Vision Based Autonomous Driving and Facilitation of an Embedded Neural Network Platform” Register here
- Download CEVA-XM6 product note
You might also like
More from Deep Learning
In two of my previous posts, I discussed deep learning frameworks, features and challenges and the specific challenges of deep …