Scalar, Vector DSP & CNN Processing Units for Highly Accurate & Fast Vision Processing
The DesignWare® EV61, EV62 and EV64 Embedded Vision Processors deliver up to 4.5 TeraMACs/sec when implemented in 16-nm processes under typical conditions and support multiple camera input with resolutions up to 4Kp. The processors are fully programmable and configurable and combine the flexibility of software solutions with the high performance and low power consumption of dedicated hardware.
The EV6x family integrates a high-performance 32-bit scalar core with a 512-bit vector DSP, and an optimized convolution neural network (CNN) engine fast for accurate object detection, classification and scene segmentation. The EV6x supports any CNN, including popular networks such as AlexNet, VGG16, GoogLeNet, Yolo, Faster R-CNN, SqueezeNet, and ResNet. Designers can run CNN graphs originally trained for 32-bit floating point hardware on the EV6x’s 12-bit CNN engine, significantly reducing the power and area of their designs while maintaining the same levels of detection accuracy. The engine delivers power efficiency of up to 2,000 GMACs/sec/W when implemented in 16-nm FinFET process technologies (worst-case operating conditions). The EV6x’s CNN hardware also supports neural networks trained for 8-bit precision to support lower memory bandwidth and power use.
The EV61 features a single vision CPU (32-bit scalar RISC with a 512-bit vector DSP), the EV62 features a dual-core vision CPU and the EV64 has a quad-core vision CPU. All of the cores can be configured with the optional CNN engine that can be configured with 880, 1,760 or 3,520 MACs.
To speed application software development, the EV6x processor family is supported by a comprehensive software programming environment based on embedded vision and neural network standards including OpenCV, OpenCL™ C, OpenVX™ with the DesignWare ARC MetaWare EV Development Toolkit. The toolkit includes a CNN mapping tool that analyzes neural networks trained using popular frameworks like Caffe and Tensorflow, and automatically generates the executable for the programmable CNN engine. For maximum flexibility and future-proofing, the tool can also distribute computations between the vision CPU and CNN resources to support new and emerging neural network algorithms as well as customer-specific CNN layers.
DesignWare EV61, EV62 and EV64 Processors Datasheet
Downloads and Documentation
- Optimized for high frame-rate and video resolution embedded vision applications
- Fast, accurate object detection with a programmable CNN engine
- CNN engine:
- Delivers up to 4.5 TeraMACs/sec in typical operating conditions when implemented in 16-nm FinFET processes
- Offers power efficiency of up to 2000 GMACs/sec/W in worst-case operating conditions when implemented in 16-nm FinFET processes
- Supports both coefficient and feature map compression/decompression to reduce data bandwidth requirements and decrease power consumption
- High-performance vision CPU with 512-bit wide SIMD vision DSP and 32-bit scalar CPU
- Vision CPU scales from 1 to 4 vector DSPs and operates in parallel to the CNN engine
- Supports data- and task-level parallelism
- Runs full range of vision algorithms for HD resolutions up to 4Kp
- Works with all host processors for vision offload
- High productivity MetaWare EV Development Toolkit supports C/C++, OpenCL C, OpenCV and OpenVX
- CNN mapping tool automatically dispatches processing tasks to available hardware resources for faster execution
|EV61 processor for embedded vision applications with a single-core HS VDSP vision CPU||STARs
|EV62 processor for embedded vision applications with dual-core HS VDSP Vision CPU||STARs
|EV64 processor for embedded vision applications with quad-core HS VDSP Vision CPU||STARs