Ceva has 200+ patents reflecting its 30 years of experience powering 19 billion devices. Ceva’s transformative semiconductor IP and embedded software offerings are used by the world’s top semiconductor and electronics companies to develop extraordinary and differentiated products that connect, sense, and infer – the three critical pillars of the rapidly evolving era of AI-enabled Smart Edge.
Built on that foundation, Ceva’s solutions enable a new generation of connected and distributed intelligence to make our lives safer, enjoyable, and more efficient. Ceva has established leadership positions in key technology areas, Including reliable and secure wireless connectivity for use in both infrastructure and end points; low-power and highly efficient audio and video/vision sensing and interface solutions; and scalable neural-network-based AI processing.
As Ceva points out is “empowering Smart Edge Innovation.”
IoT Evolution World: Ceva’s latest DSPs integrate support for AI inferencing. How are these capabilities being applied across different verticals like consumer, automotive, and industrial? And how open are these platforms for customer innovation and differentiation?
Moshe Sheier: Ceva’s processor solutions include NPUs for edge AI inferencing, and DSPs with AI support for a combination of signal processing and AI workloads. Our latest DSPs for computer vision and communications support AI inference and are applied across consumer, automotive, and industrial sectors. In consumer applications, they enhance communication modems and audio processing. In automotive, power ADAS and autonomous driving systems. In industrial, they enable predictive maintenance and machine vision.
Ceva's processors offer scalable AI performance – from 200 GOPS to 400 TOPS, customizable ISAs, software reusability, and innovative technologies like Dynamic Vector Threading, allowing customers to innovate and differentiate their products effectively.
IoT Evolution World: With the shift toward Software Defined Vehicles and increasing autonomy, how is Ceva evolving its DSP and sensing IP to meet the new requirements for real-time data processing, safety, and over-the-air upgradability in cars?
Moshe Sheier: Ceva is evolving its NPU and DSP IPs to meet these new automotive requirements. The NeuPro-M NPU family supports scalable AI performance of well over a 1000 TOPS, handling models with over a billion parameters, are automotive safety compliant, and enabling real-time, on-device inferencing.
Ceva's DSPs, like the Ceva-XC family, integrate real-time communication processing for 5G and V2X, to support over-the-air updates, ensuring systems can be updated with the latest AI models. These advancements enhance automotive safety, efficiency, and flexibility, supporting the shift toward software-defined vehicles and increasing autonomy.
IoT Evolution World: Ceva offers a broad range of wireless IP – from Bluetooth and Wi-Fi to UWB and 5G. What’s driving the demand for multi-protocol connectivity, and how do you approach bundling or customizing IP for customers developing Smart Edge devices?
Moshe Sheier: The demand for multi-protocol connectivity is driven by the need for versatile, high-performing, and cost-effective devices that can handle diverse use cases in consumer IoT, industrial, and personal computing markets. Ceva addresses this by offering the Ceva-Waves™ Links family, which integrates Wi-Fi, Bluetooth, UWB, and IEEE (News - Alert) 802.15.4 (for Thread/Zigbee/Matter) into a single IP package, which also includes the RF module.
This approach greatly simplifies development, reduces time-to-market, and provides optimized co-existence schemes for efficient concurrent communications. The modular architecture allows for easy customization to meet specific customer needs based on required protocols support.
IoT Evolution World: As more modules integrate AI at the edge, how do Ceva’s NPUs and ML-optimized DSPs support this trend? Can customers achieve meaningful AI performance without redesigning their system architecture?
Moshe Sheier: Ceva’s NPUs such as the NeuPro-Nano and NeuPro-M, support AI at the edge by offering scalable, efficient, and low-power solutions for embedded AI workloads. These NPUs handle neural network compute, feature extraction, and DSP tasks within a self-contained architecture, enabling high AI performance without the need for system redesign. The Ceva-NeuPro Studio SDK further simplifies development by providing an integrated software development environment for optimizing and deploying AI applications, ensuring customers can achieve meaningful AI performance efficiently.
IoT Evolution World: Over your decades in IP licensing, what lessons has Ceva learned from market shifts – such as the rise and fall of certain wireless standards – and how have those informed your current strategy in Smart Edge connectivity?
Moshe Sheier: One key lesson is the importance of diversification. Initially focused on the mobile (handset) market, Ceva expanded its portfolio to include multiple connectivity, sensing, and AI solutions for consumer IoT, automotive, and industrial applications, reducing reliance on any single market. This diversification has allowed us to adapt to changing market demands and technological advancements.
Ceva's current strategy in Smart Edge connectivity emphasizes flexibility and scalability. By offering a comprehensive range of wireless IPs, including 5G, Wi-Fi, Bluetooth, UWB, and 802.15.4, Ceva ensures its solutions can meet diverse customer needs. We focus on low power innovation and being first to market with the latest standards, and our programmable platforms support user’s customization to multiple standards, which helps maintain a competitive edge.
IoT Evolution World: What question should I have asked and what is the answer to that question?
Moshe Sheier: Possibly, “With multiple AI solutions out there, how is Ceva differentiating?”
Ceva differentiates itself by offering a holistic, self-contained processing solution for AI. Instead of just an AI accelerator, Ceva provides a standalone Neural Processing Unit (NPU) that delivers scalable performance and efficient neural network implementation. This solution includes an SDK, a model zoo of pre-optimized models, and a rich ecosystem of complementary AI solutions. This comprehensive approach ensures that customers have access to a robust and versatile IP offer, which many find highly beneficial for their AI needs.