40 billion devices operating at the network edge by 2025
| sensAI Adds Low-Power AI Inference Capabilities to OEM Designs
| 10x Improvement in Performance with CNN IP and Neural Network Compiler

Lattice Semiconductor announced on the 22nd that it is enhancing the key performance and design flow of its sensAI solution stack.

The sensAI stack provides a comprehensive hardware and software solution for implementing low-power (1mW to 1W) characteristics and always-on AI capabilities in smart devices operating at the network edge.

IHS projected that 40 billion devices will operate at the network edge by 2025.

OEMs designing always-on edge devices want to minimize the number of times data is sent to the cloud for analysis due to various reasons, such as latency, network bandwidth limitations, and data privacy.

sensAI enables OEMs to seamlessly add low-power AI inference capabilities optimized for their application requirements to their existing designs.

By integrating local intelligence, OEMs can also reduce costs associated with cloud-based analytics by transmitting relevant data only when further data processing is required.

Pixcellence, Inc.Seton P. Kasmir of .) CEO Kasmir stated, “Pixelence’s expertise lies in the development of image processing as well as computer vision solutions that support industry-first features such as color night vision.”

He added, “With the recent rise in interest in IoT, there is increasing demand for smart cameras that support AI applications such as motion detection and facial recognition; however, smart cameras face limitations in using off-the-shelf ASSPs due to strict power consumption and cost constraints.”

He also explained, “The Lattice sensAI solution stack enables us to easily add low-power and flexible AI inference capabilities to our existing or new camera designs, allowing us to bring our value-added products to market faster.”

The Lattice sensAI solution stack has improved performance by 10 times compared to the previous version thanks to improved convolutional neural network IP (CNN IP) and a neural network compiler that support features such as 8-bit activation quantization, smart layer integration, and dual-DSP engines.

You can also accelerate the design cycle through new features such as support for extended neural network and machine learning frameworks like Keras, the elimination of repetitive post-processing tasks through support for quantization and fractional setting schemes for neural network training, and simple neural network debugging via USB.

Deepak Boppana, Senior Director of Segment and Solution Marketing at Lattice Semiconductor, said, “The Lattice sensAI stack was developed to meet the growing demand for edge devices that support AI capabilities.”

He added, “sensAI not only already has a track record of multiple awards, but also low-power sensAI for always-on IoT devices such as smart doorbells and security cameras "The customer and partner ecosystem developing foundational solutions is also growing," they stated.

"Lattice solves network-related challenges for customers from the edge to the cloud in the growing telecommunications, computing, industrial, automotive, and consumer markets," said Ying Jen Chen, Director of Business Development for Asia Pacific at Lattice Semiconductor. "We will do our best to help customers more easily and quickly achieve the innovations needed to realize a smarter, safer, and more connected world through Lattice's technology, long-term trusted relationships, and world-class technical support."