Surely it can't be too much longer before something comes our way as a result of this new partnership between Arm and Cerence, where low-power, real-time LLM inference at the edge is the central challenge?
As we all know, Akida excels at:
- Keyword spotting
- Natural language intent classification
- Time-series pattern recognition
While Arm and Cerence are working on optimizing LLMs on traditional CPU/GPU pipelines, the bottlenecks of power, latency, and thermal limits in vehicles still remain. Akida, being a neuromorphic processor would be capable of delivering sub-milliwatt operation for AI inference, event-based, real-time processing, on-device learning capabilities and ultra-low latency for audio and language data streams.
What's not to like about that? These would be ideal traits for in-vehicle voice assistants and LLM use cases, where responsiveness, power efficiency, and privacy really matter.
It says here that "CaLLM Edge operates fully on Arm-based chipsets" and we know Akida is compatible with the Arm product family as has been successfully demonstrated with Cortex M85.
I could easily imagine a Cerence voice assistant enhanced by Akida doing real-time voice analysis and decision-making, entirely offline, with a power budget that’s EV-battery friendly.
Arm should be asking: "How can we future proof this stack for in-cabin AI by 2026-2027 when compute demands will surge but battery and thermal budgets won't".
Cerence AI and Arm push LLM boundaries with on-device AI for smarter cars
Jun 6, 2025 |
Stephen Mayhew
Categories
Edge Computing News |
Hardware
Cerence AI has partnered with semiconductor manufacturer,
Arm to enhance its embedded small language model (SLM), CaLLM Edge, using Arm’s Kleidi software library.
The collaboration aims to optimize CPU and GPU performance for real-time language processing at the edge, improving speed, efficiency, and privacy highlighting the growing importance of
edge computing and generative AI in the automotive industry.
Arm’s Kleidi technology accelerates machine learning and neural network operations on Arm-based devices, addressing the challenges of limited compute power in vehicles. CaLLM Edge operates fully on Arm-based chipsets, enabling advanced in-car AI capabilities without relying on cloud connectivity.
“We are excited to partner with Arm to take CaLLM Edge to the next level, setting new standards for performance and efficiency in edge computing in the car,” says Nils Schanz, EVP, Product & Technology,
Cerence AI. “By combining our expertise in AI-powered language models with Arm’s innovative library, we are continuing our journey to create a new era of voice-first experiences and next-generation AI applications in the automotive space, empowering consumers with smarter, faster, and more responsive in-car assistants.”
This partnership supports automakers in delivering smarter, faster, and more responsive AI-powered user experiences for drivers and setting new standards for in-car AI applications, enhancing safety and connectivity.
https://www.edgeir.com/cerence-ai-and-arm-push-llm-boundaries-with-on-device-ai-for-smarter-cars-20250606