Expedera is a leading provider of semiconductor intellectual property, specializing in ultra-efficient AI inference engines. Their technology is designed to optimize AI processing tasks, tailoring solutions that fit precisely to the customers' performance, power, and area (PPA) requirements. This customization allows Expedera's products to eliminate unnecessary silicon waste, minimize memory needs, and improve overall efficiency. With a focus on addressing both current and future AI network standards, their innovations are finding applications in a variety of fields, including mobile devices, automotive, virtual reality, smart home, and industrial sectors.
At the core of Expedera's offerings are their scalable neural processing units (NPUs) that adopt a unique packet-based architecture. This architecture allows for parallel execution across multiple layers, enhancing performance while reducing power consumption, area, and latency. Expedera's technology enables clients to run their trained neural networks unchanged, maintaining model accuracy without needing hardware-specific tweaks. This leads to an increase in performance and a substantial decrease in power usage, making their solutions ideal for edge devices that operate under tight PPA constraints.
The company's Origin line of neural engine products reflects Expedera's commitment to efficiency and scalability. These include the Origin E1, E2, E6, and E8, each optimized for different performance tiers and applications. The offerings range from small-scale implementations in consumer electronics to high-performance solutions in data centers and autonomous vehicles. They also offer TimbreAI T3, a low-power AI interface focused on audio noise reduction for devices such as headsets and wearables.
Expedera's dedication to power efficiency is a highlight of their architecture, achieving up to 18 TOPS/W, which is among the highest in the industry. Their technology supports a wide range of AI models including CNN, RNN, LSTM, and emerging models like large language models (LLMs) and stable diffusion. Their solutions are field-proven with deployment in over 10 million consumer devices, providing a solid foundation for future innovation and expansion.
Read more
