AI News, Millimeter-Scale Computers: Now With Deep-Learning Neural Networks on Board

Millimeter-Scale Computers: Now With Deep-Learning Neural Networks on Board

Many of the microphones, cameras, and other sensors that make up the eyes and ears of smart devices are always on alert, and frequently beam personal data into the cloud because they can’t analyze it themselves.

At the conference, they described micromote designs that use only a few nanowatts of power to perform tasks such as distinguishing the sound of a passing car and measuring temperature and light levels.

They showed off a compact radio that can send data from the small computers to receivers 20 meters away—a considerable boost compared to the 50-centimeter range they reported last year at ISSCC.

They also described their work with TSMC (Taiwan Semiconductor Manufacturing Company) on embedding flash memory into the devices, and a project to bring on board dedicated, low-power hardware for running artificial intelligence algorithms called deep neural networks.

(They pass ideas back and forth rapidly, not finishing each other’s sentences but something close to it.) The memory research is a good example of how the right trade-offs can improve performance, says Sylvester.

They typically demand both large memory banks and intense processing power, and so they’re usually run on banks of servers often powered by advanced GPUs.

Some researchers have been trying to lessen the size and power demands of deep-learning AI with dedicated hardware that’s specially designed to run these algorithms.

The Michigan group brought down the power requirements by redesigning the chip architecture, for example by situating four processing elements within the memory (in this case, SRAM) to minimize data movement.

Security cameras and other connected devices are not smart enough to tell the difference between a burglar and a tree, so they waste energy sending uninteresting footage to the cloud for analysis.

New Futuristic Computers Could Completely Eliminate the Need for the Cloud

Technology has a habit of shrinking in size while simultaneously growing in functionality year after year.

This idea, coupled with the prediction that almost 1 trillion devices will be in circulation within the Internet of Things (IoT)  by 2035, led to the creation of the tiny, energy-efficient computer sensors.

The pair also worked with Taiwan Semiconductor Manufacturing Company (TSMC) on making it possible to run artificial intelligence algorithms known as deep neural networks on these tiny computers. Older sensors utilize low-powered SRAM (static RAM), which makes it difficult to optimize the full potential of video and sound.

While we don’t know if that’s big money the company invested in these tiny computers, any backing from the tech giant is likely to help propel development on these powerful little devices.

Michigan Micro Mote (M3) Makes History

Michigan Micro Mote (M3), the world’s smallest computer, has taken its place among other revolutionary accomplishments in the history of computing at the Computer History Museum in Mountain View, CA.

It is the achievement of Michigan faculty members David Blaauw, Dennis Sylvester, David Wentzloff, Prabal Dutta and several key graduate students over the years, some of whom have already founded companies to exploit key aspects of the technology.

a computer system must have an input of data, the ability to process that data - meaning process and store it, make decisions about what to do next – and ultimately, the ability to output the data.”

The Michigan Micro Mote contains solar cells that power the battery with ambient light, including indoor rooms with no natural sunlight, allowing the computers to run perpetually.

devices includes computers equipped with imagers (with motion detection), temperature sensors, and pressure sensors.

The Phoenix processor is miniscule at 915 x 915µm2, and boasts ultra-low operating voltage and a unique standby mode that results in an average power consumption of only 500pW.

(Consider that 1pW is the average power consumption of a single human cell.) Blaauw explained why Phoenix’s extreme energy efficiency is so important: “As you shrink down in size, the percentage of the system tends to be dominated by the battery.

With the M3, engineers at Michigan are the first to accomplish energy neutrality via indoor energy harvesting in a wireless system of its size.

That’s about a million times less power than the average mobile phone consumes while on standby, or the comparative difference between the thickness of a sheet of paper and the length of a football field.

The current pinnacle of the project is an imaging system that packs visual imaging, ultra-low power motion detection, wireless communications, battery, power management, solar harvesting, processor and memory into a package measuring a mere 2 x 4 x 4mm3.

“Down the road we want these sensors to be able to talk to one another,” says Blaauw, “and we’re currently working to extend their range to about 20m.” A

stumbling block to extending the range of these devices is the antenna size and accompanying increase in power needed to communicate long distances.

The sensor system included a MEMS pressure sensor that could be deployed to monitor intraocular pressure in glaucoma patients and intracranial pressure in trauma victims.

In the meantime, the Michigan team continues to redefine computer technology ahead of the IoT curve as they collaborate with industry, the Kellogg Eye Center, the U-M Medical School, and colleagues here at Michigan and around the world.

Speck-Size Computers: Now With Deep Learning

Many of the microphones, cameras, and other sensors that make up the eyes and ears of smart devices are always on alert, and frequently they beam personal data into the cloud because they can’t analyze it themselves.

By developing tiny, energy-efficient computing sensors that can do analysis on board, Blaauw and Sylvester hope to make these devices more secure, while also saving energy and bandwidth.

In San Francisco, they described micromote designs that use only a few nanowatts of power to perform tasks such as distinguishing the sound of a passing car and measuring temperature and light levels.

They showed off a compact radio that can send data from the small computers to receivers 20meters away—a considerable boost compared with the 50-centimeter range they reported last year.

They also described their work with TSMC (Taiwan Semiconductor Manufacturing Co.) on embedding flash memory into the devices and a project to bring on board dedicated, low-power hardware for running artificial intelligence algorithms called deep neural networks.

(They pass ideas back and forth rapidly, not finishing each other’s sentences but something close to it.) The memory research is a good example of how the right trade-offs can improve performance, says Sylvester.

They typically demand both large memory banks and intense processing power, and so they’re usually run on banks of servers often powered by advanced GPUs.

The Michigan group brought down the power requirements by redesigning the chip architecture, for example by situating four processing elements within the memory (in this case, SRAM) to minimize data movement.

Security cameras and other connected devices are not smart enough to tell the difference between a burglar and a tree, so they waste energy sending uninteresting footage to the cloud for analysis.

Mod-01 Lec-35 Variation Tolerant Design

Low Power VLSI Circuits & Systems by Prof. Ajit Pal, Computer Science and Engineering, IIT Kharagpur. For more details on NPTEL visit