Lattice Saves PC Power Using Edge AI

Article By : Sally Ward-Foxton

Implementing edge AI-enabled power saving features can extend laptop battery life by as much as 28 percent.

As part of the latest version of its SensAI stack for edge AI applications, FPGA maker Lattice has debuted reference designs for power saving in laptops to improve battery life. One technique, attention tracking, could potentially extend battery life by an estimated 28 percent.

SensAI version 4.1 includes hardware and software reference designs for features such as camera-based, user-presence detection that powers down a PC when idle. Also included is attention tracking, a feature that dims the screen’s brightness when the user is not looking at it.

“Users now expect to show up to the machine and the machine turns on again,” said Lattice marketing director Hussein Osman. “Your phone does that, why shouldn’t your laptop do it, and do it very easily, without the user having to move a mouse or press a power button?”

Lattice said attention tracking implemented via edge AI technology on an FPGA saved more power than was consumed if a user was distracted more than 2 percent of the time. When a user was distracted (looking away from the screen, for instance) for 45 percent of the time, which Osman said is typical, battery life jumped 28 percent.

Such functionality is readily available on smartphones, prompting Lattice to apply it to laptops. Osman said Lattice sees a place for implementing the technology on GPUs, CPUs and FPGAs in client-side PCs, similar to the heterogenous computing on the server side.

“Some manufacturers do have those use cases running on an SoC using software that they developed or that they acquired from other companies,” said Osman. “Based on the testing we’ve done, adding the software does not save battery life – it takes away from the battery because now you’re using the GPU and CPU to do inference, to understand what’s going on. Offloading that use case [from] the main CPU is an important piece.”

Lattice’s attention-tracking reference design “breaks even” on power saving if the user is distracted or looking away from the screen more than 2 percent of the time. (Source: Lattice Semiconductor) (Click to enlarge)

Ecosystem complexity has not helped. According to Osman, OEMs typically choose as many as three different sensors from different vendors as a hedge against supply constraints. That means OEMs have to support different AI ASICs alongside the sensor along with different operating systems or sensor software.

“All this adds up to a lot of complexities on the OEM side, and the OEM’s [approach] is to merge some of these together, to have something that works under all operating systems,” Osman said. “This has been a really important piece of our offering; we bring differentiation and an experience that feels the same across different platforms, across different operating systems and across different SoCs.”

The latest version of the Lattice SensAI stack also includes reference designs to handle face-framing during video conferencing and onlooker detection to help maintain user privacy. Other new capabilities include an updated neural network compiler. New AutoML features are intended to help designers select an AI model from Lattice’s models based on specified speed, power consumption and accuracy requirements.

As part of the stack, a new reference hardware platform for voice and vision machine learning includes an onboard image sensor, two microphones and expansion connectors for additional sensors.

This article was originally published on EE Times.

Sally Ward-Foxton covers AI technology and related issues for and all aspects of the European industry for EE Times Europe magazine. Sally has spent more than 15 years writing about the electronics industry from London, UK. She has written for Electronic Design, ECN, Electronic Specifier: Design, Components in Electronics, and many more. She holds a Masters’ degree in Electrical and Electronic Engineering from the University of Cambridge.


Leave a comment