Aeva adds Cadence DSP IP for LiDAR

Aeva has licensed Cadence Tensilica Vision DSP IP to accelerate signal processing in its 4D LiDAR systems for industrial robotics and automotive applications.

The design will bring Cadence’s programmable DSP technology into Aeva’s next-generation LiDAR processing pipeline, supporting real-time perception systems that require high performance, low latency, and efficient power consumption. Aeva’s 4D LiDAR technology detects velocity and position simultaneously, enabling autonomous machines to interpret movement as well as distance.

Cadence’s Tensilica Vision DSPs are designed for programmable vision and signal processing workloads. In this application, the IP will support LiDAR processing pipelines that involve point cloud processing, neural network execution, simultaneous localisation and mapping, radar-style signal operations, and computer vision workloads.

James Reuther, chief engineer of Aeva, said: “Cadence’s DSP technology provides the flexibility and performance uplift we need to push the boundaries of perception and deliver scalable solutions to our industrial and automotive customers.”

LiDAR systems are evolving from optical front ends into integrated perception platforms. Modern designs increasingly combine sensing, signal processing, AI acceleration, and software libraries inside the sensor architecture itself. That places greater demand on processor IP that can be configured around specific workloads while remaining within embedded power and latency limits.

Aeva’s 4D LiDAR approach measures both range and velocity, allowing autonomous systems to distinguish moving objects from static surroundings. Automotive automation is one target market, but the same capability also applies to industrial robotics, warehouse automation, smart infrastructure, and factory inspection, where dynamic environments can make conventional sensing more difficult.

Programmable DSP IP gives sensing system designers flexibility as perception algorithms change. Fixed-function hardware can provide efficiency for stable workloads, but autonomy and physical AI systems are still developing quickly. Configurable signal-processing capability allows critical parts of the pipeline to be optimised without forcing every function into immutable hardware.

Cadence’s Tensilica Vision DSPs include support for software libraries covering neural networks, computer vision, SLAM, radar, and point cloud processing. They also support the NeuroWeave SDK, a third-generation neural network compiler intended to execute newer AI networks used in LiDAR-related applications.

Industrial sensing environments test perception systems in ways that differ from controlled road scenarios. Factory automation and robotics deployments often involve vibration, variable lighting, reflective surfaces, narrow aisles, moving machinery, and unpredictable human activity. Velocity data can improve object interpretation in these environments, but only when the signal chain can process the additional information fast enough for control decisions.

The automotive market adds its own engineering constraints. ADAS and automated-driving systems require perception stacks that are accurate, power-efficient, certifiable, and scalable across vehicle platforms. The shift towards software-defined vehicles means sensors must also fit broader compute and update architectures, making programmable processing inside the sensing path more valuable.

Processor IP selection has become part of the competitive design of advanced sensors. A LiDAR system is now judged on the quality of its full perception pipeline: optics, detectors, signal processing, AI execution, software tooling, power profile, and integration with the wider machine architecture. The Cadence licence gives Aeva another processing element for that pipeline as industrial and automotive autonomy platforms become more software-led.