home
Search

A new era for eye tracking with AI algorithms on NPU

  • Blog
  • by Erland George-Svahn
  • 6 min

NPU illustration

Earlier this year, Tobii launched Tobii Nexus, an AI powered webcam-based eye tracking software library for device and application integrations. As we push the boundaries of what's possible with eye tracking technology, the processing power needed to run AI algorithms efficiently has become more critical than ever.

In addition to using traditional CPUs, we are now utilizing new hardware, such as Neural Processing Units (NPU), when available. This shift brings unique opportunities for device manufacturers and developers to integrate eye tracking into their products. Here’s why this advancement matters.

Host NPU: Turbocharging AI Performance

NPUs accelerate AI workloads, offering a significant performance boost when running deep learning models and complex vision algorithms. Integrating our algorithms to run on Intel NPU brings key advantages, including the following:

  • Energy-efficient performance: One key benefit of NPUs is their ability to handle AI tasks at lower power consumption than CPUs. In our lab benchmarks, an Intel NPU demonstrated 3 times lower power consumption and 46% faster processing versus the execution on the CPU. This efficiency is crucial for battery-powered devices like laptops and portable medical devices, where battery life is extended without sacrificing performance.
  • Optimized CPU availability: Offloading AI tasks to specialized NPUs frees up the CPU for other critical functions, enhancing performance efficiency and battery life.
NPU vs CPU illustration

It’s important to note that while NPUs are increasingly integrated into more laptops and tablets, not all include it yet. If available, the Tobii algorithm automatically executes on Intel NPU and falls back to the CPU when not. This allows developers and OEMs to benefit from hardware acceleration for eye tracking without drastically changing their existing architectures.

Edge AI: Low power NPUs near the camera

As AI-powered devices move toward edge computing — where low-power, small-footprint devices handle real-time processing directly — Edge AI NPUs become increasingly important. These units are designed for ultra-efficient processing on edge devices, making them ideal for products such as laptops, monitors, IoT devices, and mobile applications.

NPU computer illustration

Here’s how running the Tobii eye tracking algorithms on Edge NPUs offers great advantages:

  • Enhanced data privacy: Edge AI NPUs allow AI algorithms to process camera images and produce gaze data without those images reaching the main compute board. This is particularly important for devices or displays where data privacy is critical.
  • Minimal power consumption: Edge AI NPUs are optimized for low power, making them a perfect fit for devices where battery life is crucial. This allows eye tracking technology to be integrated into smaller, more portable devices like phones and laptops without draining the battery.
  • Real-time performance: Edge AI solutions manage the image pipeline from sensor to processing, cutting out image processing that happens today on an OS level. This setup is crucial for low-latency, high-refresh-rate eye tracking. For instance, stereoscopic monitors or phones could provide a flawless 3D experience and gaze-based interaction.

The benefits for OEMs and ISVs: flexibility and futureproofing

By supporting AI algorithms that can run across CPU, Intel NPU, and Edge AI NPUs, Tobii eye tracking offers OEMs and ISVs several key benefits:

  • Performance optimization: OEMs and ISVs can select the processing architecture that best fits their product’s performance and energy requirements—whether it's a high-end device using an Intel NPU for ultra-fast, high-fidelity tracking or a mobile device running on Edge AI NPU for edge processing.
  • Scalability across devices: With support for multiple hardware architectures, developers can create vertically scalable solutions that run efficiently on everything from high-end desktops to lightweight IoT devices. This versatility reduces development time by enabling a single AI model to run across various products.
  • Costs and energy efficiency: Running AI tasks on specialized NPUs like Intel’s or Edge AI NPUs allows OEMs to deliver high-performance, AI-driven features without needing to over-spec the CPU, keeping costs down and power consumption in check.

A new frontier for AI-powered eye tracking

We’re excited to see how our integration with NPUs is reshaping the future of eye tracking technology. By enabling our AI algorithms to run seamlessly across CPU, Intel NPU, and Edge AI NPU, we empower OEMs and ISVs to build smarter, faster, and more energy-efficient products.

Whether you’re looking to enhance accessibility, improve healthcare assessment, or create more intuitive user experiences, our eye tracking algorithms are designed to deliver precise, real-time insights on the hardware platform that best suits your needs. We’re committed to providing the flexibility and performance required to meet the industry's evolving demands.

Want to see it in action? Download a demo. When run on Intel NPU-enabled devices, our webcam eye tracking algorithms automatically leverage the neural processing unit.

This is just the beginning, and we’re excited to partner with you to drive the next wave of innovation!

Written by

  • Tobii employee

    Erland George-Svahn

    Product portfolio manager screen-based integrations

    Erland George-Svahn manages the Tobii screen-based integration portfolio that consists of both hardware and software integration offerings for B2B. He has over 20 years of experience in designing products, leading cross-functional teams and helping customers succeed!

Deepen your knowledge on webcam eye tracking for devices

Swoosh Top

Subscribe to our blog

Subscribe to our stories about how people are using eye tracking and attention computing.