At embedded world 2026, on the DigiKey booth, Paige Hookway speaks with Raemin Wang, Vice President, Segment Marketing at Lattice Semiconductor about small, low power FPGA enabling physical AI.
Wang began by introducing himself and explaining that his team focuses on physical AI solutions and ecosystems, working closely with partners in industrial, automotive, and consumer sectors.
He provided an overview of Lattice Semiconductor, noting the company’s expertise in low-power, small-size FPGAs (field-programmable gate arrays), which offer flexible I/O, multiple interfaces, and parallel processing capabilities. These FPGAs are programmable at the hardware level, allowing deployment in a range of applications including robotics, autonomous vehicles, 5G communications, and data centres.
The conversation then shifted to the challenge of managing the increasing variety of sensors in Edge devices. Wang highlighted the proliferation and diversity of sensors, such as LiDAR, radar, IMUs, and RGB cameras, used in industrial automation, robotics, and consumer devices.
Lattice FPGAs can efficiently ingest and process high-bandwidth multi-sensor data due to their flexible interfaces, enabling engineers to handle complex inputs at the Edge rather than relying solely on central processors. This “sense AI” approach allows AI inferencing to occur closer to the sensor, which improves performance, reduces latency, and minimises power consumption.
Wang said that Lattice’s Sense AI solutions provide pre-built, tiny AI models optimised for high accuracy with minimal resource requirements. These models facilitate the transition from AI prototypes to real-world deployment by addressing constraints such as limited power, space, and computational resources. The models can be deployed on Lattice FPGAs as well as other MPUs or NPU accelerators, supporting applications like human-machine interfaces, defect detection, and multi-object recognition.
The discussion also covered Lattice’s integration with NVIDIA’s Holoscan platform. Lattice FPGAs simplify sensor integration and pre-processing, enabling low-latency transfer of data into NVIDIA’s AI ecosystem. The company offers reference designs that allow developers to quickly modify sensor configurations without redesigning the entire system, accelerating time-to-market and supporting a robust partner ecosystem.
Security and safety were emphasised as critical considerations in AI-driven industrial applications. Wang described FPGA-based approaches to security, noting their advantages over traditional CPU-based architectures. Lattice FPGAs support hardware roots of trust, monitor system buses for anomalies, and enable rapid recovery using golden boot images, reducing downtime. He also discussed “cyber agility,” highlighting that FPGAs can adapt to evolving security standards, including post-quantum cryptography (PQC) to counter future quantum computing threats. Lattice’s 55TDQ FPGA, capable of PQC today, was showcased at the event and recognised with a “Best of Show” award.
Find out more in the interview below.
