AI stack doubles performance for edge AI
For on-device AI processing at the edge, Lattice Semiconductor has introduced the sensAI 3.0, available on low power, 28nm FD-SOI-based Lattice CrossLink-NX FPGAs.
This latest version of the company’s stack features customised convolutional neural network (CNN) IP. This flexible accelerator IP simplifies the implementation of common CNN networks, explained Lattice, and is optimised to leverage the parallel processing capabilities of FPGAs. Adding support for CrossLink-NX FPGAs brings new levels of power and performance to smart vision applications in the surveillance/security, robotics, automotive and computing markets, said the company.
To address data security, latency and privacy issues, developers want to move the AI processing that powers their smart vision and other AI applications from the cloud to the edge. Most edge devices are battery-powered or sensitive to power consumption, so developers need hardware and software that deliver the processing capabilities needed for AI applications, while keeping power consumption as low as possible.
For applications like smart vision that require higher edge AI performance, CrossLink-NX FPGAs running sensAI software deliver twice the performance at half the power when compared to prior releases of the solutions stack, confirmed Lattice.
Updates to the NN compiler software tool let developers easily compile a trained NN model and download it to a CrossLink-NX FPGA.
A new feature for this version is a VGG-based object counting demo operating on a CrossLink-NX FPGA. It delivers 10 frames per second while consuming only 200mW. Object counting is used in smart vision applications that are used in the surveillance, security, industrial, automotive and robotics markets.
When running on a CrossLink-NX FPGA, the sensAI solutions stack offers up to 2.5Mbit of distributed memory and block RAM with additional DSP resources for efficient on-chip implementation of AI workloads, this reduces the need for cloud-based analytics.
Being manufactured in a 28nm FD-SOI process means that the CrossLink-NX FPGAs deliver a 75 per cent reduction in power, in comparison to similar, competing FPGAs, claims Lattice.
A target application for sensAI is smart vision, and CrossLink-NX devices are currently the only low-power FPGAs to deliver MIPI I/O speeds of up to 2.5Gbits per second to support components such as image sensors. This makes CrossLink-NX FPGAs a suitable hardware platform for sensAI applications requiring MIPI support. CrossLink-NX FPGA’s I/Os offer instant-on performance and are able to configure themselves in less than three milliseconds, with full-device configuration in as little as eight milliseconds.
There is also increased neural network architecture support in the form of support for the MobileNet v2, SSD, and ResNet models on the Lattice ECP5 family of general-purpose FPGAs.
Lattice Semiconductor specialises in low power programmable devices, working with customers across the network, from the edge to the cloud, in the communications, computing, industrial, automotive and consumer markets.