DSP IP Suits Imaging and AI Applications
July 24, 2019
One of the primary inputs used for artificial intelligence is vision/imaging. And the better your images, the better your data, leading to better intelligence.
One of the primary inputs used for artificial intelligence is vision/imaging. And the better your images, the better your data, leading to better intelligence. To that end, Cadence Design Systems recently announced some new members joining its Tensilica Vision DSP product family.
The Vision Q7 DSP can deliver up to 1.82 tera-operations per second (TOPS), which is double the previous generation. The core is specifically optimized for simultaneous localization and mapping (SLAM), a technique that’s commonly used in the robotics, drones, and the mobile and automotive industries to automatically construct or update a map of an unknown environment. It’s also popular in the AR/VR world, where it’s used for inside-out tracking.
Features of the Vision Q7 DSP include a very long instruction word (VLIW) SIMD architecture and enhanced instruction set supporting 8/16/32-bit data types and optional VFPU support for single and half precision.
The flexible solution delivers 512 8-bit MACs, and it can be paired with the Tensilica DNA 100 processor to further boost performance. Further enhancements 3D DMA, compression and a 256-bit AXI interface. And because the Vision Q7 DSP is a superset of the Q6 DSP, existing software investments can be maintained and there’s an easy upward migration path.
The Vision Q7 DSP supports AI applications developed in the Caffe, TensorFlow, and TensorFlowLite frameworks through the Tensilica Xtensa Neural Network Compiler (XNNC), which maps neural nets into executable and optimized code. The DSP also supports the Android Neural Network (ANN) API for on-device AI acceleration in Android-powered devices. In addition, development tools and libraries are all designed to enable SoC vendors to achieve ISO 26262 automotive safety integrity level D (ASIL D) certification.