Ceva, Inc. has extended its Ceva-NeuPro family of edge AI NPUs with the launch of Ceva-NeuPro-Nano. These highly-efficient NPUs claim the power, performance and cost efficiencies needed to integrate TinyML models into SoCs for consumer, industrial and general-purpose artificial intelligence of things (AIoT) products.
The Ceva-NeuPro-Nano embedded AI NPU architecture is fully programmable and supports most advanced machine learning (ML) data types and operators including native transformer computation, sparsity acceleration and fast quantization.
(Source: Ceva, Inc.)
Addressing the specific performance challenges of TinyML—the deployment of ML models on low-power and resource-constrained devices for IoT applications, Ceva said, the Ceva-NeuPro-Nano NPUs enable edge AI everywhere with the aim to make AI ubiquitous, economical and practical for a range of use cases, including voice, vision, predictive maintenance and health sensing in consumer and industrial IoT applications.
By 2030 over 40% of TinyML shipments will be powered by dedicated TinyML hardware rather than all-purpose microcontrollers, according to ABI Research.
“Ceva-NeuPro-Nano is a compelling solution for on-device AI in smart edge IoT devices. It addresses the power, performance, and cost requirements to enable always-on use-cases on battery-operated devices integrating voice, vision, and sensing use cases across a wide array of end markets,” said Paul Schell, industry analyst at ABI Research, in a statement. “From TWS earbuds, headsets, wearables, and smart speakers to industrial sensors, smart appliances, home automation devices, cameras, and more, Ceva-NeuPro-Nano enables TinyML in energy constrained AIoT devices.”
The embedded AI NPU efficiently executes neural networks, feature extraction, control code and DSP code, Ceva said, and these features eliminate the need for a companion MCU for computationally intensive tasks.
Ceva also noted that the self-sufficient architecture enables Ceva-NeuPro-Nano NPUs to deliver high power efficiency, with a smaller silicon footprint, and optimal performance compared to the existing processor solutions, a combination of CPU or DSP with AI accelerator-based architectures, used for TinyML workloads.
The NPUs feature the Ceva-NetSqueeze AI compression technology that processes compressed model weights, without the need for an intermediate decompression stage, enabling up to an 80% memory footprint reduction. This solves a key bottleneck limiting the broad adoption of AIoT processors, the company said.
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. The Ceva-NPN32 is optimized for most TinyML workloads targeting voice, audio, object detection and anomaly detection use cases, while the
Ceva-NPN64 delivers enhanced performance for more complex on-device AI use cases such as object classification, face detection, speech recognition and health monitoring. The Ceva-NPN64 provides 2× performance acceleration using weight sparsity, greater memory bandwidth, more MACs and support for 4-bit weights.
The edge NPU also delivers ultra-low energy thanks to energy optimization techniques, such as automatic on-the-fly energy tuning and by distilling computations using weight-sparsity acceleration.
The Ceva-NeuPro-Nano NPUs are available for licensing now. They 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. Offering an easy click-and-run experience for users, the SDK supports open AI frameworks including TensorFlow Lite for microcontrollers (TFLM) and microTVM (µTVM). It provides a Model Zoo of pretrained and optimized TinyML models covering voice, vision and sensing use cases and a portfolio of optimized runtime libraries and off-the-shelf application-specific software.
Advertisement
