Expedera Inc., a leader in scalable Neural Processing Unit (NPU) semiconductor IP, today unveiled its Origin Evolution NPU IP — a breakthrough designed to enhance Generative AI (GenAI) capabilities on edge devices. Origin Evolution addresses the significant challenges of running large language models (LLMs), convolutional neural networks (CNNs), and recurrent neural networks (RNNs) on resource-constrained hardware, enabling low-latency, secure AI inference beyond the cloud.
Features and Benefits
- Optimized Edge AI Performance
Origin Evolution supports scalable AI inference with up to 128 TFLOPS per core and can scale to PetaFLOPS with multi-core configurations. It optimizes power, performance, and area (PPA) to suit applications from smartphones and automobiles to data centers. - Packet-Based Architecture for Efficiency
Utilizing Expedera’s unique packet-based architecture, Origin Evolution reduces costly external memory movements by over 75% for LLM models such as Llama 3.2 1B and Qwen2 1.5B. This design improves processor utilization while significantly lowering memory and system power requirements. - Comprehensive Model and Network Support
The NPU IP natively supports popular AI networks, including Llama3, ChatGLM, DeepSeek, Qwen, MobileNet, Yolo, and MiniCPM, and can run custom or ‘black box’ layers without retraining or accuracy loss. - Advanced Software and Hardware Stack
Accompanied by a powerful software suite supporting frameworks like HuggingFace, Llama.cpp, and TVM, Origin Evolution offers full integer and floating-point precision, layer fusion/fission, and multi-core centralized control. Its high-speed external memory streaming interface supports the latest DRAM and HBM standards.
Strategic Impact
CEO and co-founder Siyad Ma emphasized, “Origin Evolution is a radical advancement providing an AI inference engine with out-of-the-box compatibility for LLM and CNN networks, ideal for diverse applications from smartphones to data centers. It embodies years of engineering progress and offers immense value to customers aiming to integrate GenAI into their products.”
With Origin Evolution, Expedera empowers companies to embed powerful GenAI directly into edge devices, overcoming the limitations of cloud-dependent AI models by reducing latency and addressing security concerns inherent to cloud computing.
