Ceva and Mobilint have moved a key piece of edge AIartificial intelligence hardware from development into commercial use. Mobilint’s REGULUS edge AIartificial intelligence System-on-Chip combines Ceva’s SensPro AIartificial intelligence DSP1. differential signal processingSignal processing based on differences between two signals, often rejecting common interference.2. digital signal processing and NeuPro-Nano NPUNeural Processing UnitA processor optimized for neural-network inference workloads. and marks the first production deployment of Ceva’s NeuPro NPUNeural Processing UnitA processor optimized for neural-network inference workloads. family.
The SoC is positioned as a standalone processor for on-device AIartificial intelligence, aimed at applications that need local inference rather than cloud-side processing. It delivers 10 TOPSTera Operations Per SecondA metric used to describe the peak operation rate of an AI accelerator. of AIartificial intelligence performance at under three watts and supports up to eight gigabytes of memory, making it suitable for compact systems where power, space and response time are tightly constrained.
Its architecture splits work across specialized engines. SensPro provides programmable sensing and signal-processing capabilities, while NeuPro-Nano handles efficient neural processing for always-on inference. Mobilint’s own AIartificial intelligence processing architecture is integrated alongside those blocks, giving the platform a mix of sensing, signal processing and decision-making resources.
That partitioning matters for Physical AIartificial intelligence workloads, which typically involve several steps in sequence: preparing sensor data, recognizing objects, understanding the environment and reacting quickly. Matching those tasks to different processing engines gives product designers more flexibility as they trade off performance, efficiency and responsiveness.
REGULUS already appears in shipping hardware. It powers Mobilint’s MLD-R1 AIartificial intelligence USB accelerator, which was first previewed at COMPUTEX 2026 and is now shipping. The accelerator is designed to add on-device AIartificial intelligence capabilities to existing systems and targets computer vision, video analytics and generative AIartificial intelligence use cases.
Supported ecosystems include PyTorch, ONNX, Hugging Face, TensorFlow, TFLite and Keras. The potential application set includes smart factories, robotics, machine vision, security, medical devices and digital signage.
For Ceva, the deployment provides a production example of its AIartificial intelligence technology in commercial silicon. Mobilint contributes energy-efficient AIartificial intelligence semiconductors and SoC integration, while Ceva supplies silicon and software IP1. image placement2. Internet protocolInterface or communications concept for transmitting data between devices and systems.3. Intellectual PropertyIP denotes licensable semiconductor design blocks or technology elements. for intelligent edge applications. Together, the companies are pushing edge AIartificial intelligence closer to products that can process sensor data locally and respond in real time.