AI solution delivers performance boost for low power at the Edge

For reasons including latency, network bandwidth limitations, and data privacy, OEMs designing always-on Edge devices want to minimise sending data to the Cloud for analytics.

Lattice sensAI enables such OEMs to seamlessly update their existing designs with low power AI inferencing optimised for their application requirements. Incorporating such local intelligence also lowers expenses related to Cloud-based analytics by only sending relevant data for further processing.

“At Pixcellence, our expertise is in developing image processing and computer vision solutions with industry-first features like colour night vision. Interest in the IoT is fueling demand for smart cameras that support AI applications like presence detection or facial recognition,” said Seton P. Kasmir, CEO, Pixcellence.

“However, smart cameras have strict power consumption and cost requirements that make it a challenge to use off-the-shelf ASSPs. The Lattice sensAI solutions stack lets us easily add low power, flexible AI inference support to our existing and new camera designs, and get our value-added products to market faster.”

“Our sensAI stack was developed to address the growing need for Edge devices with AI support, and we’re encouraged by the market’s response to our solution,” said Deepak Boppana, Senior Director of Segment and Solutions Marketing, Lattice Semiconductor.

“In addition to the multiple industry awards received by sensAI, it is exciting to see the broader partner ecosystem and customers developing low power sensAI-based solutions for always-on IoT devices such as smart doorbells and security cameras.”

New enhancements to the Lattice sensAI solution stack include:

• Ten times performance boost over previous version - driven by an updated CNN IP and neural network compiler with features like 8-bit activation quantisation, smart layer merging and a dual-DSP engine
• Seamless user experience - accelerates the design cycle with several new features including:
  • Expanding neural network and ML frameworks support including Keras
  • Support for quantisation and fraction setting schemes for neural network training eliminates iterative post-processing
  • Simple neural network debugging via USB
  • New customisable reference designs accelerate time to market for popular use cases like object counting and presence detection
• A growing design service partner ecosystem, including full product design capability from partners such as Pixcellence, makes it easier for customers to get to market faster.