September 30, 2019 by Luke James
The DesignWare EV7x Vision Processors' architecture integrates vector DSP, vector FPU, and a neural network accelerator to provide a scalable solution that features more than a four-times performance increase for AI-intensive applications and emerging AI applications.
Earlier this month, Synopsys debuted its DesignWare ARC EV7x Embedded Vision Processor family that includes a deep neural network accelerator for AI and machine learning applications. Delivering what it claims to be a best-in-class 35 tera operations per second (TOPS) performance for AI-intensive applications.
The EV7x's cores have been designed with architectural enhancements. These include power gating to reduce power consumption and a more granular clock.
Synopsys has based its new processor on the ARCv2 RISC architecture and it features a 1, 2, or 4-core heterogenous architecture integrating vector DSP, vector FPU, and a neural network accelerator with integrated AES 256-bit encryption to enable a huge range of potential applications that cover everything from consumer to automotive.
The deep neural network (DNN) accelerator can scale from 880 to 14,080 MACs to deliver this 35 TOPS performance in a 16nm FinFET process technology. It also supports convolutional neural networks and batched long short-term memories for applications that need time-based results. The DNN works parallel to the vision engine, making the EV7x particularly useful in autonomous vehicles and ADAS applications where multiple input devices (e.g. cameras) and algorithms are operating at the same time.
The EV7x's processor has been designed for embedded vision tasks. Image courtesy of Synopsys.
Additionally, the new EV7x design uses clock and power gating technologies alongside architectural enhancements to cut down power consumption, and MetaWare EV Software Development Toolkit enhancements, combined with the processor, accelerate FPS throughput by 65% when compared with its predecessor, EV6x.
Each of the EV7x's VPUs features a 32-bit scalar unit and a 512-bit wide vector DSP that can be configured for 8, 16, or 32-bit operations.
The EV7x processor has an optional AES-XTS encryption engine that protects data passing from on-chip memory to the vision engine and DNN accelerator. This prevents valuable data (e.g. personal biometrics such as facial recognition data) from being intercepted and exploited by hackers.
Designers can achieve high levels of functional safety by using lockstep capabilities and safety monitors without having to lower performance or power. The EV7x can be purchased as ASIL B and ASIL D-compliant versions that accelerate ISO26262 certification of automotive systems on a chip (SoC). ASIL D safety requirements apply to applications such as steering systems, keyless entry, and automotive sensors.
Synopsys, Inc. is a Silicon to Software partner for innovative companies developing everyday electronic products and software applications. As a global leader in the electronic design automation and semiconductor space, Synopsys innovations such as the EV7x processor are helping bolster its leadership in the software security and space, too.
