Expedera provides customisable neural engine semiconductor IP that dramatically improves performance, power, and latency while reducing cost and complexity in edge AI inference applications.
Successfully deployed in over 10 million consumer devices, Expedera’s Neural Processing Unit (NPU) solutions are scalable and produce superior results in applications ranging from edge nodes and smartphones to automotive.
The platform includes an easy-to-use TVM-based software stack that allows the importing of trained networks, provides various quantisation options, automatic completion, compilation, estimator, and profiling tools, and supports multi-job APIs.
Headquartered in Santa Clara, California, the company has engineering development centres and customer support offices in the United Kingdom, China, Japan, Taiwan, and Singapore.
Origin™ is a neural engine IP line of products from Expedera that reduces memory requirements to the bare minimum and dramatically slashes processing overhead. Its unique packet-based architecture is far more efficient than the common layer-based architectures underlying other NPU implementations.
The architecture subdivides each layer into self-contained executable fragments that can be scheduled independently. This enables parallel execution across multiple layers, better resource utilisation, and deterministic performance.
It also eliminates the need for hardware-specific optimisations, allowing customers to run their trained neural networks unchanged, with no reduction in model accuracy. This innovative approach greatly increases performance while lowering power, area, and latency.
Siyad Ma is CEO, co-founder and former Expedera’s VP of Engineering. Previously, he led Algorithmic TCAM ASIC and IP teams for Cisco Nexus7k, MDS, Cat4k/6k. Siyad brings over 25 years of experience driving ASIC design and DFT at Spanslogic (Cisco), Zettacom (IDT), Chameleon, and AMD. He holds a PhD EE from Stanford.
