The AMD Versal™ adaptive system-on-chips (SoCs) bring together scalar, adaptable, and intelligent engines, along with cutting-edge memory and interfacing technologies. These components work in unison to provide powerful heterogeneous acceleration for a wide range of applications. An essential feature of Versal adaptive SoCs is their focus on being programmable and optimizable by data scientists, software developers, and hardware engineers. To facilitate this, a comprehensive set of tools, software, libraries, IP, middleware, and frameworks are provided to support standard design processes.
These SoCs, built using TSMC's 7 nm FinFET process technology, represent a groundbreaking platform that combines software programmability and domain-specific hardware acceleration. The portfolio comprises six different series of devices, each designed for scalability and AI inference capabilities across a diverse set of markets, including cloud, networking, wireless communications, edge computing, and endpoints.
The Versal architecture integrates various engine types with robust connectivity and communication capabilities, all connected via a network on chip (NoC) to ensure seamless memory-mapped access throughout the device. The Intelligent Engines consist of SIMD VLIW AI Engines for adaptive inference and advanced signal processing, while DSP Engines perform fixed point, floating point, and complex MAC operations. Adaptable Engines combine programmable logic blocks and memory to maximize computational density. Scalar engines include Arm® Cortex® -A72 and Cortex®-R5F processors, tailored for intensive computational tasks.
The AI Engine system software driver is a critical component in the AI Engine system, serving as the bridge between the hardware and software layers. This user guide outlines the key features and functions of the software driver, explaining how it aids in configuring, debugging, and tracing AI Engine operations. This guide is intended for software developers, system architects, and engineers who work with AI Engine-based systems and need to understand the AI Engine software driver's design, capabilities, and features.