Ceva, Inc. announced that it has extended its Ceva-NeuPro family of Edge AI NPUs with the introduction of Ceva-NeuPro-Nano NPUs. These highly-efficient, self-sufficient NPUs deliver the power, performance and cost efficiencies needed for semiconductor companies and OEMs to integrate TinyML models into their SoCs for consumer, industrial, and general-purpose AIoT products. TinyML refers to the deployment of machine learning models on low-power, resource-constrained devices to bring the power of AI to the Internet of Things (IoT).

Driven by the increasing demand for efficient and specialized AI solutions in IoT devices, the market for TinyML is growing rapidly. According to research firm ABI Research, by 2030 over 40% of TinyML shipments will be powered by dedicated TinyML hardware rather than all-purpose MCUs. By addressing the specific performance challenges of TinyML, the Ceva-NeuPro-Nano NPUs aim to make AI ubiquitous, economical and practical for a wide range of use cases, spanning voice, vision, predictive maintenance, and health sensing in consumer and industrial IoT applications.

The new Ceva-NeuPro-Nano Embedded AI NPU architecture is fully programmable and efficiently executes Neural Networks, feature extraction, control code and DSP code, and supports most advanced machine learning data types and operators including native transformer computation, sparsity acceleration and fast quantization. This optimized, self-sufficient architecture enables Ceva-NeuPro-Nano NPUs to deliver superior power efficiency, with a smaller silicon footprint, and optimal performance compared to the existing processor solutions used for TinyML workloads which utilize a combination of CPU or DSP with AI accelerator-based architectures. Furthermore, Ceva-NetSqueeze AI compression technology directly processes compressed model weights, without the need for an intermediate decompression stage.

This enables the Ceva-NeuPro-Nano NPUs to achieve up to 80% memory footprint reduction, solving a key bottleneck inhibiting the broad adoption of AIoT processors. The Ceva-NeuPro-Nano NPU is available in two configurations - the Ceva-NPN32 with 32 int8 MACs, and the Ceva-NPN64 with 64 int8 MACs, both of which benefit from Ceva-NetSqueeze for direct processing of compressed model weights. The Ceva-NPN32 is highly optimized for most TinyML workloads targeting voice, audio, object detection, and anomaly detection use cases.

The Ceva-NPN64 provides 2x performance acceleration using weight sparsity, greater memory bandwidth, more MACs, and support for 4-bit weights to deliver enhanced performance for more complex on-device AI use cases such as object classification, face detection, speech recognition, health monitoring, and others. The NPUs are delivered with a complete AI SDK - Ceva-NeuPro Studio - which is a unified AI stack that delivers a common set of tools across the entire Ceva-NeuPro NPU family, supporting open AI frameworks including TensorFlow Lite for Microcontrollers (TFLM) and microTVM (µTVM). The Ceva-NeuPro-Nano Key Features: Flexible and scalable NPU architecture: Fully programmable to efficiently execute Neural Networks, feature extraction, control code, and DSP code; Scalable performance by design to meet a wide range of use cases; MAC configurations with up to 64 int8 MACs per cycle; Future proof architecture that supports the most advanced ML data types and operators; 4-bit to 32-bit integer support; Native transformer computation; Ultimate ML performance for all use cases using advanced mechanisms; Sparsity acceleration; Acceleration of non-linear activation types; Fast quantization; Edge NPU with ultra-low memory requirements: Highly efficient, single core design for NN compute, feature extraction, control code, and DSP code eliminates need for a companion MCU for these computationally intensive tasks; Up to 80% memory footprint reduction via Ceva-NetSqueeze which directly process compressed model weights without the need for an intermediate decompression stage; Ultra-low energy achieved through innovative energy optimization techniques: Automatic on-the-fly energy tuning; Dramatic energy and bandwidth reduction by distilling computations using weight-sparsity acceleration; Complete, Simple to Use AI SDK: Ceva-NeuPro Studio provides a unified AI stack, with an easy click-and-run experience, for all Ceva-NeuPro NPUs, from the new Ceva-NeuPro-Nano to the powerful Ceva-NeuPro-M; Fast time to market by accelerating software development and deployment; Optimized to work seamlessly with leading, open AI inference frameworks including TFLM and µTVM; Model Zoo of pretrained and optimized TinyML models covering voice, vision and sensing use cases; Flexible to adapt to new models, applications and market needs; Comprehensive portfolio of optimized runtime libraries and off-the-shelf application-specific software.