The Pulse
Ceva’s NeuPro NPU Reaches Production in Mobilint’s REGULUS
Ceva says Mobilint has deployed its NeuPro-Nano NPU and SensPro AI DSP in the REGULUS edge AI chip, marking the first production deployment of Ceva’s NeuPro NPU family. The announcement describes REGULUS as a 10-TOPS processor and says it i
AI.info Team · Written with AI · fact-checked ·
Ceva calls REGULUS its first production deployment
Ceva says Mobilint has deployed its NeuPro-Nano neural processing unit in the REGULUS edge AI system-on-chip, marking the first production deployment of Ceva’s NeuPro NPU family. In its November 2, 2026 announcement, Ceva said Mobilint had also deployed Ceva’s SensPro AI digital signal processor in the chip. REGULUS combines those components with Mobilint’s processing architecture.
Ceva describes REGULUS as delivering 10 trillion operations per second, or TOPS, at less than 3 watts and supporting up to 8GB of memory. The release lists robotics, drones, machine vision, smart cameras and internet-connected devices among the intended applications, where processing must fit within power and space constraints.
“REGULUS brings together the specialized processing capabilities needed to run sophisticated AI workloads directly on the device within very tight power constraints,” said Dongjoo Shin, CEO of Mobilint.
A prior report put mass production on a later schedule
A September report from The Elec said Mobilint’s mass-production schedule for REGULUS had been delayed by six months. Production was then expected to begin no earlier than June 2027, following design work with Asicland to improve chip performance during optimization for the foundry process. The report said Mobilint had delivered all samples requested by customer Shinsegae I&C; it separately reported that quality verification was underway, without identifying who was conducting it.
Ceva’s November announcement does not address the schedule reported by The Elec or state how many REGULUS chips have been produced. It says the chip is already enabling Mobilint’s MLD-R1 AI USB accelerator, which the release describes as now shipping. The announcement provides no unit volumes or shipment totals.
Two Ceva blocks handle sensing and inference
Ceva describes SensPro as a programmable processor for sensing and signal-processing tasks, while NeuPro-Nano handles neural-network processing for always-on inference. The company says the multi-engine architecture allows different sensing and AI workloads to run on the processing engine best suited to each task. Its announcement does not provide workload-specific benchmark results or detail how much of the stated 10 TOPS comes from each processor.
Ceva’s release says the MLD-R1 supports computer vision, video analytics and generative AI workloads. It lists PyTorch, ONNX, Hugging Face, TensorFlow, TensorFlow Lite and Keras among the supported AI frameworks and ecosystems. The release describes the accelerator as designed to add on-device AI to existing systems, with applications including smart factories, robotics, machine vision, security, medical devices and digital signage.
Ceva gains a commercial reference for NeuPro
For Ceva, the REGULUS deployment is a first production use of its NeuPro NPU family in another company’s chip. Yaron Galitzky, Ceva’s executive vice president and general manager of its AI Division, said the deployment demonstrated progress in bringing Ceva’s AI technology into commercial silicon.
The announcement establishes that Ceva IP is part of a product the company says is shipping, but leaves measures of commercial scale unanswered. Neither the announcement nor The Elec report provides REGULUS production quantities or shipment totals. Ceva’s release identifies the MLD-R1 as a commercial product using REGULUS, but does not report sales or revenue tied to the design.