Harnessing neural networks to accelerate Edge AI

Humans excel at cognitive processing, for example, recognizing faces, vehicle lane tracking, or separating human speech from background noise. This happens because the brain’s neural networks learn how to analyze and interpret important visual and audio cues.

Creating artificially intelligent machines with the same abilities is challenging but important in applications such as automotive safety, surveillance, and security. Accelerating edge AI deployment in neural network-based designs is critical to addressing this challenge.

One solution lies in supplying a dedicated low power AI processor for Deep Learning at the edge, combined with a deep neural network (DNN) graph compiler that:

  • Automatically quantizes and converts models for use in real-time Edge AI devices, offering significant reduction in time-to-market
  • Ensures operation with the minimal power and memory bandwidth overheads in embedded systems
  • Delivers superior performance while retaining the flexibility to stay up-to-date with the latest technology in the constantly evolving domain of embedded machine learning

target markets


Deep neural networks provide the sophisticated image processing that advanced driver assistance systems (ADAS) need to recognize signs, pedestrians, and vehicles.

Security and Surveillance

Embedded systems that offer face recognition based on neural networks are increasingly employed in camera-based surveillance. Coupled with audio sensors, neural networks can identify sounds, such as breaking glass or dogs barking, and trigger a planned response.

Augmented Reality

Real-time augmented reality applications on battery-powered mobile devices rely on deep learning and energy-efficient operation.

Smart Home

Sophisticated interpretation and response to voice commands and audio inputs by smart appliances and personal assistants depend on deep learning.

Retail Automation

Facial age and gender profiling enabling retail kiosks to match offers to customers, and natural language processing allowing them to interact, both require deep learning processing.


Deep learning supports audio keyword detection and natural language processing in patient diagnostic systems.