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

Photo: University of Michigan and TSMCOne of several varieties of University of Michigan micro motes. This one incorporates 1 megabyte of flash memory.

Computer scientist David Blaauw pulls a small plastic box from his bag. He carefully uses his fingernail to pick up the tiny black speck inside and place it on the hotel café table. At one cubic millimeter, this is one of a line of the world’s smallest computers. I had to be careful not to cough or sneeze lest it blow away and be swept into the trash.

Blaauw and his colleague Dennis Sylvester, both IEEE Fellows and computer scientists at the University of Michigan, were in San Francisco this week to present ten papers related to these “micro mote” computers at the IEEE International Solid-State Circuits Conference (ISSCC). They’ve been presenting different variations on the tiny devices for a few years.

Their broader goal is to make smarter, smaller sensors for medical devices and the internet of things—sensors that can do more with less energy. Many of the microphones, cameras, and other sensors that make up 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. Some have predicted that by 2035, there will be 1 trillion such devices. “If you’ve got a trillion devices producing readings constantly, we’re going to drown in data,” says Blaauw. 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.

Photo: University of Michigan/TSMCMade of multiple layers of computing.

At the conference, they described micro mote designs that use only a few nanowatts of power to perform tasks such as distinguish 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 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.

Blaauw and Sylvester say they take a holistic approach to adding new features without ramping up power consumption. “There’s no one answer” to how the group does it, says Sylvester. If anything, it’s “smart circuit design,” Blaauw adds. (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 tradeoffs can improve performance, says Sylvester. Previous versions of the micro motes used 8 kilobytes of SRAM, which makes for a pretty low-performance computer. To record video and sound, the tiny computers need more memory. So the group worked with TSMC to bring flash memory on board. Now they can make tiny computers with 1 megabyte of storage.

Flash can store more data in a smaller footprint than SRAM, but it takes a big burst of power to write to the memory. With TSMC, the group designed a new memory array that uses a more efficient charge pump for the writing process. The memory arrays are a bit less dense than TSMC’s commercial products, for example, but still much better than SRAM. “We were able to get huge gains with small trade-offs,” says Sylvester.

Another micro mote they presented at the ISSCC incorporates a deep-learning processor that can operate a neural network while using just 288 microwatts. Neural networks are artificial intelligence algorithms that perform well at tasks such as face and voice recognition. 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. But even those processors still use over 50 milliwatts of power—far too much for a micro mote. 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.

The idea is to bring neural networks to the internet of things. “A lot of motion detection cameras take pictures of branches moving in the wind—that’s not very helpful,” says Blaauw. 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. On-board deep-learning processors could make better decisions, but only if they don’t use too much power. The Michigan group imagine deep-learning processors could be integrated into many other internet-connected things besides security systems. For example, an HVAC systems could decide to turn the air conditioning down if they see multiple people putting on their coats.

After demonstrating many variations on these micro motes in an academic setting, the Michigan group hopes they will be ready for market in a few years. Blaauw and Sylvester say their start-up company CubeWorks is currently prototyping devices and researching markets. The company was quietly incorporated in late 2013. Last October, Intel Capital announced they had invested an undisclosed amount in the tiny computer company. 

ORIGINAL: IEEE Spectrum

By Katherine Bourzac

Posted 10 Feb 2017

Electric eel inspires bio-friendly power source, what happens next may shock you

Could a device inspired by the electric eel offer a safer way to power medical implants?

Scientists are always on the lookout for safer, more natural ways to power devices that go into our bodies. After all, who really needs toxic battery elements and replacement surgery?

One organism that is pretty good at generating biocompatible power (for itself, at least) is the electric eel, and scientists have now used the high-voltage species as a blueprint for a promising new self-charging device that could one day power things like pacemakers, prosthetics and even augmented reality contact lenses.

Electric eels generate voltage through long stacks of thin cells that run end-on-end through their bodies. Called electrocytes, these cells create electricity by allowing sodium ions to rush into one end and potassium ions out the other, all at the same time. The voltage created by each cell is small, but together, the stacks within a single eel can generate as many as 600 V.

To recreate this effect, researchers from the University of Fribourg, the University of Michigan and the University of California San Diego turned to the difference in salinity between fresh and saltwater. They deposited hydrogel, ion-conducting blobs onto clear plastic sheets and separated them with ion-selective membranes.

Hundreds of blobs containing salt and freshwater were arranged in an alternating pattern. When the team had all these gel compartments make contact with one another, they were able to generate 100 V through what is known as reverse electrodialysis, where energy is generated through differing salt concentrations in the water.

While the eel triggers the simultaneous contact of its electrocytes using a neurotransmitter called acetylcholine as the command signal, the team achieved this by carefully working a special origami pattern – called a Miura-ori fold – into the plastic sheet. This meant that when pressure was applied to the sheet, it quickly snapped together and the cells shifted into exactly the right positions to create the electricity.

The device, which the team calls an artificial electric organ, isn't in the same ball park as an eel in terms of output, but the researchers do have some ideas around how to boost its efficiency. It points to the metabolic energy created by ion differences in the eel's stomach, or the mechanical muscle energy, as some of the possibilities, but does note that recreating these would be a major challenge.

"The electric organs in eels are incredibly sophisticated, they're far better at generating power than we are," Mayer said. "But the important thing for us was to replicate the basics of what's happening."

The research was published in the journal Nature. You can hear from Mayer in the video below.

 

Source: University of Fribourg, University of Michigan

ORIGINAL: NewAtlas
Nick Lavars
December 14th, 2017

Meet the First Artificial Animal

Scientists genetically engineered and 3-D-printed a biohybrid being, opening the door further for lifelike robots and artificial intelligence.

CREDIT: Getty Images

If you met this lab-created critter over your beach vacation, you'd swear you saw a baby ray. In fact, the tiny, flexible swimmer is the product of a team of diverse scientists. They have built the most successful artificial animal yet. This disruptive technology opens the door much wider for lifelike robots and artificial intelligence.

Like most disruption, it started with a simple idea. Kit Kevin Parker, PhD, a Harvard professor researching how to build a human heart, saw his daughter entranced by watching stingrays at the New England Aquarium in Boston. He wondered if he could engineer a muscle that could move in the same sinuous, undulating fashion. The quest for a material led to creating an artificial ray with a 3-D-printed rubber body at the School of Engineering and Applied Sciences at Harvard. Scientists from the University of Illinois at Urbana-Champaign, the University of Michigan, and Stanford University's Medical Center joined the team.

They reinforced the soft rubber body with a 3-D-printed gold skeleton so thin it functions like cartilage. Geneticists adapted rat heart cells so they could respond to light by contracting. Then, they were grown in a carefully arranged pattern on the rubber and around the gold skeleton.

The muscular circuitry is one of the most interesting parts of the research, and there's more about it in this video:

The birth of biohybrid beings

The new engineered animal responds to light so well scientists were able to guide it through an obstacle course 15 times its length using strong and weak light pulses.

The study authors write, "Our ray outperformed existing locomotive biohybrid systems in terms of speed, distance traveled, and durability (six days), demonstrating the potential of self-propelled, phototactically activated tissue-engineered robots."

What biohybrid mean for robots and artificial intelligence

Science of this type is fundamental for engineering special-purpose creations such as artificial worms that sniff out and eat cancer. Or bionic body parts for those who have suffered accidents or disease. Imagine having little swimmers in your system that rush to the site of a medical emergency such as a stroke. The promise of sensor-rich soft tissue frees robots to move more easily and yet not be cut off from needed input. Sensitized robot soft tissue could perform without the energy-sucking heaviness of metal or the artificial barrier of hard-plastic exoskeletons.

Thanks to disruptive, cross-disciplinary applied science like this, entrepreneurs in the next few years will be able to play on the border of what life is, what alive means, and what life can be. Expect to see companies use biohybrid beings to commercialize applications that solve some of the largest, and most lucrative, challenges we face today.

ORIGINAL: INC
BY LISA CALHOUN General partner, Valor Ventures@Lisa_Calhoun

Scientists coax stem cells to form 3-D mini lungs

University of Michigan Health System

Scientists have coaxed stem cells to grow the first three-dimensional mini lungs. Previous research has focused on deriving lung tissue from flat cell systems or growing cells onto scaffolds made from donated organs.

In a study published in the online journal eLife the multi-institution team defined the system for generating the self-organizing human lung organoids, 3D structures that mimic the structure and complexity of human lungs.

Figure 1. Generation of three-dimensional ventral anterior foregut spheroids from endoderm monolayers.

(A) hESCs were differentiated into foregut endoderm by treating cells with 4 days of Activin A (ACTA) followed by 4 days of NOG+SB. (B) Foregut endoderm (NOG+SB) had high expression of the foregut marker SOX2 while the hindgut marker CDX2 was significantly reduced compared to untreated endoderm controls (End). NOG+SB monolayers had high expression of ventral anterior foregut genes NKX2.1 and PAX8 while the posterior foregut marker PDX1 was reduced. The foregut marker HHEX is expressed in the developing liver, biliary system, and thyroid and remained unchanged. (C) The majority of cells in NOG+SB treated cultures were SOX2 positive (green) compared to the control, in which only scattered clusters of cells were SOX2 positive. The scale bar represents 200 µm. (D) hESCs were differentiated into foregut spheroids by treating cells with 4 days of ACTA and then additional 4–6 days of NOG+SB+FGF4+Ch. Representative images of a spheroid in a matrigel droplet are shown as a whole mount image. Scale bar represents 100 µm. (E) Foregut spheroids (NOG+SB+FGF4+Ch) had high expression of the foregut marker SOX2 while the hindgut marker CDX2 was significantly reduced compared to untreated endoderm control (End) (top panel). Spheroids had high expression of anterior foregut genes NKX2.1 and PAX8 while the posterior foregut marker PDX1 was reduced and HHEX was unchanged (bottom panel). *p < 0.05, error bars represent SEM. (F) The majority of cells in foregut spheroids are FOXA2+ (green, left panel) and SOX2+ (white, right panel) and ECAD+ (red, right panel). Scale bar represent 50 µm.

DOI: http://dx.doi.org/10.7554/eLife.05098.003

Biochip quickly tests results of cancer therapy

The inside structure of the biochip is comprised of 3 layers, each testing a key element of the therapy: drug, oxygen, and light. (Credit: Xia Lou/University of Michigan)


University of Michigan right Original Study ("A high-throughput photodynamic therapy screening platform with on-chip control of multiple microenvironmental factors")

A new type of lab-on-a-chip could make it easier to determine the effectiveness of a promising cancer treatment that combines photosensitive drugs, light, and oxygen.

The treatment, known as photodynamic therapy (PDT), uses a unique type of drug, called a photosensitizer, that generates high-energy oxygen when activated by light, in this case an LED laser.

The oxygen reaction destroys cells locally only in the immediate surrounding area, which in this case is the tumor, without damaging other healthy cells in the rest of the body. PDT may also prompt the patient’s immune system to attack the tumor, whereas without treatment it will likely ignore it.

“PDT is a fairly complicated therapy because it requires light, oxygen, and the drug, all three of which need to be carefully controlled,” says Xia Lou, a postdoctoral fellow at the University of Michigan. “With chemotherapy, for example, you only control how much of the drug is used.”

The biochip can test the interaction of the drug, light, and oxygen simultaneously, generating results in a fraction of the time of current testing practices.

“In cancer research doctors are always looking for better drugs,” adds Lou. “But there has always been a lack in the ability to efficiently test new drugs.

“Researchers are also hoping to get more reliable test results than is the norm. We are providing more precise drug test conditions to insure that the results we are getting will more closely match the actual results from cancer treatment.”

Test all three components There are two primary challenges in PDT today:

• One is determining the best treatment plan for the patient, and 
• the other is testing the efficacy of new and existing drugs.

For example, the location of a tumor in the body will determine the amount of light needed to excite the drug, and will impact the amount of drug that is administered. Also, when testing new drugs, it is again important to know the amount of light needed to excite the oxygen reaction, and the strength of the reaction.

This adds a high level of complexity compared to conventional drug testing that only tests one component, the drug itself.

Current PDT testing methods might take more than 24 hours to acquire just 10 data points.

“With our device,” says Lou, “we can finish the testing in 1 hour and have 1,000 data points. The result is a very comprehensive understanding of how the 3 elements need to be manipulated for maximum effect in an individual patient.”

The researchers reported the details of their findings in the journal Lab on a Chip. The NSF Engineering Research Center for Wireless Integrated Microsystems, Thermo Fisher Scientific, and the National Institute of Health supported the project.

Source: University of Michigan

ORIGINAL: Futurity

by Catharine June-U. Michigan 

February 28, 2014 

Graphene Gives You Infrared Vision in a Contact Lens

Illustration: University of Michigan

It sounds like something from a spy thriller movie: putting on a contact lens that gives you infrared vision without the need for a bulky contraption that covers your face. But now, thanks to research at the University of Michigan, such a contact lens is a real possibility.


The Michigan researchers turned to the optical capabilities of graphene to create their infrared contact lens. IBM last year demonstrated some of the photoconductivity mechanisms of graphene that make it an attractive infrared detector.


Graphene is capable of detecting the entire infrared spectrum, with visible and ultraviolet light thrown in. But where graphene giveth, it also taketh away. Because graphene is only one-atom thick, it can absorb only 2.3 percent of the light that hits it. This is not enough to generate an electrical signal, and without a signal it can’t operate as a infrared sensor.


"The challenge for the current generation of graphene-based detectors is that their sensitivity is typically very poor," said Zhaohui Zhong, assistant professor at the University of Michigan, in a press release. "It's a hundred to a thousand times lower than what a commercial device would require."


In research that was published in the journal Nature Nanotechnology ("Graphene photodetectors with ultra-broadband and high responsivity at room temperature"), the Michigan researchers devised a new method for generating the electrical signal. Instead of trying to measure the electrons that are released when the light strikes the material, they amplified an electrical current that is near the electrical signals generated by the incoming light.


To achieve this amplification, the researchers started by sandwiching an insulator between two sheets of graphene. The bottom sheet has an electrical current running through it. When light hits the top sheet, electrons are freed and positively charged electron holes are generated. The electrons are able to perform a quantum tunneling effect through the insulator layer, which would be impenetrable in classical physics.


The electron holes that are left behind in the top layer generate an electric field that impacts the way electricity flows through the bottom layer. By measuring this change in the flow of current in the bottom layer, the researchers could derive just how much light hit the top layer.


This device has very nearly the same sensitivity as cooled mid-infrared detectors, but achieves it at room temperature. The researchers have already been able to produce infrared sensors the size of a pinky nail, or a standard contact lens.


"If we integrate it with a contact lens or other wearable electronics, it expands your vision," Zhong said in the release. "It provides you another way of interacting with your environment."


Most of us are familiar with the military applications of infrared vision, which allows the soldiers to see in the dark. But the technology also has medical applications such as letting doctors monitor blood flow.


Whether the ability to see in the infrared is an attractive feature for the rest of us remains to be seen. But that may become a possibility since the fundamental mechanism underlying the technology could become a mechanism for other material and device platforms. Is infrared vision mode for Google Glass in the offing?



ORIGINAL: IEEE Spectrum

By Dexter Johnson

17 Mar 2014