Today, 94% of global automakers use Arm technology in their latest vehicle models. As the Arm platform becomes more prevalent across the automotive industry, implementations for major in-vehicle AI applications are presented by AWS, BigOne, and Sonatus.

Arm announced on the 13th that it is launching Arm Kleidi, a software and community engagement program for AI acceleration, to the automotive market.

“In the automotive industry, AI is often described as key to autonomous vehicles, but the reality is that there are already many vehicle applications leveraging AI, such as adaptive cruise control and personalized infotainment systems,” said Suraj Gajendra, vice president of products and solutions for Arm’s Automotive Business Unit.

On the other hand, as vehicle applications and functions become more sophisticated and AI demands increase, the industry is faced with challenges such as: increasing complexity of the development environment; need for cloud-based vehicle software development and validation infrastructure; compliance with stringent safety and security industry standards; and need for software scalability.

Arm has been addressing these challenges by leveraging AE technologies and collaboration via SOAFEE across the automotive ecosystem of virtualized platforms and standards-based software solutions.

Arm Kleidi provides software performance optimizations for AI inference workloads running on Arm CPUs. Kleidi integrates with AI frameworks such as ExecuTorch, llama.cpp, MediaPipe, and PyTorch, so developers can start building their applications and automatically see performance improvements on Arm-based platforms.

AWS developed an AI chatbot for SDV using Arm and Corellium's virtual platform. This chatbot allows drivers to interact directly with their vehicle to ask questions or inquire about dashboard alerts instead of having to flip through a physical vehicle owner's manual.

We have seen a 10x improvement in chatbot response time with KleidiAI integrated into the latest llama.cpp, which translates to a 1-3 second reduction in chatbot response time.

Additionally, VicOne used the TinyLlama-1 1B model to double the prefill prompt speed and improve token generation speed by over 60%. Sonatus ported AI Technician Builder to run on Arm-based AWS Graviton Amazon EC2 instances with a framework optimized for Kleidi.

Arm plans to provide performance acceleration to AI developers across all Arm markets, from cloud to edge, through Kleidi’s automotive expansion.