· sensAI™ 3.0 doubles the performance and halved power consumption of AI applications at the network edge.
sensAI™ 3.0 will target smart vision applications in surveillance/security, robotics, automotive and computing.
Shanghai, China – May 21, 2020 – Leading supplier of low-power programmable devicesLatticesemiconductor Corporation (NASDAQ: LSCC) today announced the latest release of its complete set of solutions for device-side AI processing at the network edge, Lattice sensAI™ 3.0. This release now supports CrossLink-NX™ series FPGAs for low-power smart vision applications. It also features custom Convolutional Neural Network (CNN) IP, a flexible accelerator IP that simplifies the implementation of common CNN networks and is optimized to take full advantage of the parallel processing capabilities of FPGAs. With new support for CrossLink-NX FPGAs, Lattice sensAI will bring another breakthrough in power and performance to smart vision applications in surveillance/security, robotics, automotive and computing.
To address issues such as data security, latency and privacy, developers want to move the AI processing tasks that intelligent vision and other AI applications rely on from the cloud to the network edge. Most network edge devices are battery powered or very power sensitive, so developers need hardware and software solutions that not only provide the processing power required for AI applications, but also minimize power consumption. With the enhanced version of sensAI, Lattice will continue to provide customers with more optimized solutions for power consumption and performance. For network edge applications with high AI performance requirements, such as intelligent vision, the CrossLink-NX FPGA running sensAI software reduces power consumption by half compared to the previous version, while doubling the performance.
“The latest version of Lattice sensAI is optimized for the new low-power CrossLink-NX FPGA family, delivering the perfect blend of high performance and low power,” said Hussein Osman, Lattice Semiconductor Marketing Manager. Continue to consolidate the achievements made so far, continue to launch new easy-to-use application demos and reference designs for mainstream AI applications, and continuously improve the functionality and ease of use of sensAI.”
The new and updated features of the sensAI solution collection are as follows:
New CNN engine IP and compiler now support CrossLink-NX – sensAI supports running custom CNN accelerator IP on CrossLink-NX FPGA, taking full advantage of the FPGA’s parallel processing architecture. The Neural Network Compiler software tool has been updated to allow developers to easily compile and download trained neural network models to CrossLink-NX FPGAs.
CrossLink-NX-based Object Counting Demo – The VGG-based object counting demo running on a CrossLink-NX FPGA has 10 frames/sec performance and consumes only 200 mW. Object counting is a common smart vision application in surveillance/security, industrial, automotive and robotics markets.
CrossLink-NX Optimized FPGA Architecture – When running solutions on CrossLink-NX FPGAs, sensAI provides up to 2.5 Mb of distributed memory and RAM blocks and additional DSP resources to efficiently implement AI operations on-chip, reducing The need for cloud-based analytics.
Up to 75% lower power consumption—CrossLink-NX FPGAs are fabricated on a 28 nm FD-SOI process and consume up to 75% less power than competing FPGAs.
High-performance I/O – Many components used in intelligent vision systems (image sensors, application processors, etc.) need to support the MIPI I/O standard. One of sensAI’s target applications is intelligent vision, and CrossLink-NX devices are currently the only low-power FPGAs that can provide MIPI I/O at rates up to 2.5 Gbps. CrossLink-NX FPGA is an ideal hardware platform for sensAI applications that require MIPI support. The I/O of the CrossLink-NX FPGA provides instant-on performance that can complete self-configuration in less than 3 ms, while the entire device configuration takes only 8 ms.
Support for more neural network architectures – previous versions of sensAI supported VGG and MobileNet v1 neural network models. The latest version adds support for MobileNet v2, SSD and ResNet models on the Lattice ECP5TM family of general-purpose FPGAs.
About Lattice Semiconductor
Lattice Semiconductor (NASDAQ: LSCC) is a leading supplier of low-power, programmable devices. We provide solutions from the network edge to the cloud for customers in the growing communications, computing, industrial, automotive and consumer markets. Our technology, longstanding partnerships and world-class technical support enable our customers to quickly and easily start their innovation journey to create a smart, secure and connected world.