Expedera launches origin evolution NPU IP for generative AI on edge devices

Expedera  has introduced its Origin Evolution NPU IP. The technology is designed to run Large Language Models (LLMs), Convolutional Neural Networks (CNNs), and Recurrent Neural Networks (RNNs) on resource-constrained edge devices, addressing the computational demands of Generative Artificial Intelligence (GenAI).
The Origin Evolution NPU IP uses a packet-based architecture to manage power, performance, and area (PPA) requirements, latency, and memory constraints. It supports applications in smartphones, automotive systems, and data centers, and is available now, with production-ready and silicon-proven designs in customer use.

Running LLMs on edge devices reduces latency and addresses security concerns compared to cloud-based solutions. However, large model sizes and computational needs present challenges for edge systems. Origin Evolution incorporates hardware and software optimizations, including model compression, hardware optimization, attention optimization, and dedicated frameworks to manage computational and energy constraints.
The NPU IP is scalable to 128 TFLOPS in a single core and PetaFLOPS with multiple cores. It reduces external memory movements by over 75% for models like Llama 3.2 1B and Qwen2 1.5 B, and supports high memory-bound use cases, delivering thou...