AMD Unveils Monolithic 6nm FPGA for Edge AI and Automotive

AMD has recently unveiled its latest innovation in the form of the second generation of Versal 'adaptive SoC' designed specifically for edge AI applications. This new iteration marks a shift back to a monolithic chip architecture, departing from the chiplet design previously employed.

The Versal AI Edge Series Gen 2 and Versal Prime Series Gen 2 adaptive SoCs are groundbreaking in that they integrate pre-processing, AI inference, and post-processing functionalities into a single 6nm device. This consolidation allows for end-to-end acceleration of AI-driven embedded systems, all while ensuring ASIL-D levels of functional safety.

One of the early adopters of the AMD Versal AI Edge Series Gen 2 is Subaru, which has announced plans to utilize the technology for an ADAS vision system that relies solely on cameras, eliminating the need for radar or lidar sensors. The applications of this chip extend beyond automotive use, targeting robotics systems and medical devices as well.

Salil Raje, senior vice president and general manager of the Adaptive and Embedded Computing Group at AMD, emphasizes the importance of considering the entire pipeline rather than just the inference process. He highlights the necessity of adaptive computing to accommodate evolving models and sensors, stressing the significance of software in conjunction with heterogeneous platforms.

The Versal Series Gen 2 devices are tailored for functional safety applications, boasting features such as up to eight ARM Coretex-A78AE automotive cores and 10 ARM Cortex-R52 real-time cores. With a maximum of 100 AI engines, each functioning as a VLIW processor, the chip facilitates seamless connectivity through streams supporting multiple data types, compute arrays, and memory tiles.