SNUG Silicon Valley
Sondrel are excited to be presenting at SNUG Silicon Valley this year on the 30th of March. We hope to see you there!
Piyush Singh, SoC Architect at Sondrel
Abstract of the presentation
While transformer inspired networks have been shown to outperform CNNs in terms of accuracy for autonomous vehicle perception tasks, they generally have not been as efficient as CNNs in terms of inference time.
This presentation features Sondrel's SoC architecture framework and a case study demonstrating the use of Synopsys Platform Architect to model a typical automotive HPC application with the Vision Transformer neural network architecture. The performance of a proof-of-concept SoC platform for ViT inference is validated using Platform Architect.
The presentation also outlines the challenges in designing SoCs for ViT applications and presents a methodology that addresses these challenges. The effectiveness of the proposed framework, along with joint algorithm-accelerator co-design, in mitigating uncertainty early and validating concept architecture is discussed.
Click view more to register for the event!