Ceva Enhances TinyML Optimized NPUs for AIoT Devices with New IP

As the demand for efficient and specialized AI in IoT devices continues to rise, the market for TinyML is experiencing rapid growth. Research from ABI Research suggests that by 2030, over 40% of TinyML shipments will be powered by dedicated TinyML hardware, marking a significant shift in the industry. In response to these trends, Ceva has introduced the Ceva-NeuPro-Nano NPUs, designed to address the specific performance challenges of TinyML and make AI more accessible across a variety of applications.

The Ceva-NeuPro-Nano Embedded AI NPU architecture is a breakthrough in the field, offering full programmability and efficient execution of Neural Networks, feature extraction, control code, and DSP code. This versatile architecture supports a wide range of machine learning data types and operators, including native transformer computation, sparsity acceleration, and fast quantization. By optimizing power efficiency and performance, Ceva-NeuPro-Nano NPUs outperform existing processor solutions used for TinyML workloads, setting a new standard for AI integration in IoT devices.

Chad Lucien, vice president and general manager of the Sensors and Audio Business Unit at Ceva, highlighted the potential of Ceva-NeuPro-Nano in enabling companies to incorporate TinyML applications into low-power IoT SoCs and MCUs. This advancement aligns with Ceva's commitment to enhancing smart edge devices with advanced connectivity, sensing, and inference capabilities. By leveraging their expertise in wireless IoT connectivity and audio and vision sensing, Ceva aims to empower customers to unlock the full potential of TinyML and drive innovation in the IoT space.

Industry Analyst Paul Schell from ABI Research commended Ceva-NeuPro-Nano as a compelling solution for on-device AI in smart edge IoT devices. The NPU's ability to meet the power, performance, and cost requirements for always-on use-cases on battery-operated devices opens up opportunities across various markets, from consumer electronics like TWS earbuds and wearables to industrial sensors and home automation devices. Ceva-NeuPro-Nano is poised to revolutionize the integration of TinyML in energy-constrained AIoT devices.

With two configurations available — the Ceva-NPN32 and Ceva-NPN64 — the Ceva-NeuPro-Nano NPUs cater to a wide range of TinyML workloads. The Ceva-NPN32 is optimized for voice, audio, object detection, and anomaly detection, while the Ceva-NPN64 offers enhanced performance for more complex AI use cases such as object classification, face detection, and health monitoring. Both configurations benefit from Ceva-NetSqueeze technology for efficient processing of compressed model weights, ensuring superior performance and memory footprint reduction.