Bioengineers create circuit board modeled on the human brain

Stanford bioengineers have developed faster, more energy-efficient microchips based on the human brain – 9,000 times faster and using significantly less power than a typical PC. This offers greater possibilities for advances in robotics and a new way of understanding the brain. For instance, a chip as fast and efficient as the human brain could drive prosthetic limbs with the speed and complexity of our own actions.

Stanford bioengineers have developed a new circuit board modeled on the human brain, possibly opening up new frontiers in robotics and computing.

For all their sophistication, computers pale in comparison to the brain. The modest cortex of the mouse, for instance, operates 9,000 times faster than a personal computer simulation of its functions.

Not only is the PC slower, it takes 40,000 times more power to run, writes Kwabena Boahen, associate professor of bioengineering at Stanford, in an article for the Proceedings of the IEEE.

"From a pure energy perspective, the brain is hard to match," says Boahen, whose article surveys how "neuromorphic" researchers in the United States and Europe are using silicon and software to build electronic systems that mimic neurons and synapses.

Boahen and his team have developed Neurogrid, a circuit board consisting of 16 custom-designed “Neurocore” chips. Together these 16 chips can simulate 1 million neurons and billions of synaptic connections. The team designed these chips with power efficiency in mind. Their strategy was to enable certain synapses to share hardware circuits. The result was Neurogrid – a device about the size of an iPad that can simulate orders of magnitude more neurons and synapses than other brain mimics on the power it takes to run a tablet computer.

The National Institutes of Health funded development of this million-neuron prototype with a five-year Pioneer Award. Now Boahen stands ready for the next steps – lowering costs and creating compiler software that would enable engineers and computer scientists with no knowledge of neuroscience to solve problems – such as controlling a humanoid robot – using Neurogrid.

Its speed and low power characteristics make Neurogrid ideal for more than just modeling the human brain. Boahen is working with other Stanford scientists to develop prosthetic limbs for paralyzed people that would be controlled by a Neurocore-like chip.

"Right now, you have to know how the brain works to program one of these," said Boahen, gesturing at the $40,000 prototype board on the desk of his Stanford office. "We want to create a neurocompiler so that you would not need to know anything about synapses and neurons to able to use one of these."

Brain ferment

In his article, Boahen notes the larger context of neuromorphic research, including the European Union’s Human Brain Project, which aims to simulate a human brain on a supercomputer. By contrast, the U.S. BRAIN Project – short for Brain Research through Advancing Innovative Neurotechnologies – has taken a tool-building approach by challenging scientists, including many at Stanford, to develop new kinds of tools that can read out the activity of thousands or even millions of neurons in the brain as well as write in complex patterns of activity.

Zooming from the big picture, Boahen’s article focuses on two projects comparable to Neurogrid that attempt to model brain functions in silicon and/or software.

One of these efforts is IBM’s SyNAPSE Project – short for Systems of Neuromorphic Adaptive Plastic Scalable Electronics. As the name implies, SyNAPSE involves a bid to redesign chips, code-named Golden Gate, to emulate the ability of neurons to make a great many synaptic connections – a feature that helps the brain solve problems on the fly. At present a Golden Gate chip consists of 256 digital neurons each equipped with 1,024 digital synaptic circuits, with IBM on track to greatly increase the numbers of neurons in the system.

Heidelberg University’s BrainScales project has the ambitious goal of developing analog chips to mimic the behaviors of neurons and synapses. Their HICANN chip – short for High Input Count Analog Neural Network – would be the core of a system designed to accelerate brain simulations, to enable researchers to model drug interactions that might take months to play out in a compressed time frame. At present, the HICANN system can emulate 512 neurons each equipped with 224 synaptic circuits, with a roadmap to greatly expand that hardware base.

Each of these research teams has made different technical choices, such as whether to dedicate each hardware circuit to modeling a single neural element (e.g., a single synapse) or several (e.g., by activating the hardware circuit twice to model the effect of two active synapses). These choices have resulted in different trade-offs in terms of capability and performance.

In his analysis, Boahen creates a single metric to account for total system cost – including the size of the chip, how many neurons it simulates and the power it consumes.

Neurogrid was by far the most cost-effective way to simulate neurons, in keeping with Boahen’s goal of creating a system affordable enough to be widely used in research.

Speed and efficiency

But much work lies ahead. Each of the current million-neuron Neurogrid circuit boards cost about $40,000. Boahen believes dramatic cost reductions are possible. Neurogrid is based on 16 Neurocores, each of which supports 65,536 neurons. Those chips were made using 15-year-old fabrication technologies.

By switching to modern manufacturing processes and fabricating the chips in large volumes, he could cut a Neurocore’s cost 100-fold – suggesting a million-neuron board for $400 a copy. With that cheaper hardware and compiler software to make it easy to configure, these neuromorphic systems could find numerous applications.

For instance, a chip as fast and efficient as the human brain could drive prosthetic limbs with the speed and complexity of our own actions – but without being tethered to a power source. Krishna Shenoy, an electrical engineering professor at Stanford and Boahen’s neighbor at the interdisciplinary Bio-X center, is developing ways of reading brain signals to understand movement. Boahen envisions a Neurocore-like chip that could be implanted in a paralyzed person’s brain, interpreting those intended movements and translating them to commands for prosthetic limbs without overheating the brain.

A small prosthetic arm in Boahen’s lab is currently controlled by Neurogrid to execute movement commands in real time. For now it doesn’t look like much, but its simple levers and joints hold hope for robotic limbs of the future.

Of course, all of these neuromorphic efforts are beggared by the complexity and efficiency of the human brain.

In his article, Boahen notes that Neurogrid is about 100,000 times more energy efficient than a personal computer simulation of 1 million neurons. Yet it is an energy hog compared to our biological CPU.

"The human brain, with 80,000 times more neurons than Neurogrid, consumes only three times as much power," Boahen writes. "Achieving this level of energy efficiency while offering greater configurability and scale is the ultimate challenge neuromorphic engineers face."

Fruit flies, fighter jets use similar nimble tactics when under attack

When startled by predators, tiny fruit flies respond like fighter jets – employing screaming-fast banked turns to evade attacks.

Researchers at the University of Washington used an array of high-speed video cameras operating at 7,500 frames a second to capture the wing and body motion of flies after they encountered a looming image of an approaching predator.

“Although they have been described as swimming through the air, tiny flies actually roll their bodies just like aircraft in a banked turn to maneuver away from impending threats,” said Michael Dickinson, UW professor of biology and co-author of a paper on the findings in the April 11 issue of Science. “We discovered that fruit flies alter course in less than one one-hundredth of a second, 50 times faster than we blink our eyes, and which is faster than we ever imagined.”

In the midst of a banked turn, the flies can roll on their sides 90 degrees or more, almost flying upside down at times, said Florian Muijres, a UW postdoctoral researcher and lead author of the paper.

“These flies normally flap their wings 200 times a second and, in almost a single wing beat, the animal can reorient its body to generate a force away from the threatening stimulus and then continues to accelerate,” he said.

The fruit flies, a species called Drosophila hydei that are about the size of a sesame seed, rely on a fast visual system to detect approaching predators.

“The brain of the fly performs a very sophisticated calculation, in a very short amount of time, to determine where the danger lies and exactly how to bank for the best escape, doing something different if the threat is to the side, straight ahead or behind,” Dickinson said.

“How can such a small brain generate so many remarkable behaviors? A fly with a brain the size of a salt grain has the behavioral repertoire nearly as complex as a much larger animal such as a mouse. That’s a super interesting problem from an engineering perspective,” Dickinson said.

The researchers synchronized three high-speed cameras each able to capture 7,500 frames per second, or 40 frames per wing beat. The cameras were focused on a small region in the middle of a cylindrical flight arena where 40 to 50 fruit flies flitted about. When a fly passed through the intersection of two laser beams at the exact center of the arena, it triggered an expanding shadow that caused the fly to take evasive action to avoid a collision or being eaten.

With the camera shutters opening and closing every one thirty-thousandth of a second, the researchers needed to flood the space with very bright light, Muijres said. Because flies rely on their vision and would be blinded by regular light, the arena was ringed with very bright infrared lights to overcome the problem. Neither humans nor fruit flies register infrared light.

How the fly’s brain and muscles control these remarkably fast and accurate evasive maneuvers is the next thing researchers would like to investigate, Dickinson said.

Brain process takes paper shape

A paper-based device that mimics the electrochemical signalling in the human brain has been created by a group of researchers from China.

The thin-film transistor (TFT) has been designed to replicate the junction between two neurons, known as a biological synapse, and could become a key component in the development of artificial neural networks, which could be utilised in a range of fields from robotics to computer processing.

The TFT, which has been presented today, 13 February, in IOP Publishing’s journal Nanotechnology, is the latest device to be fabricated on paper, making the electronics more flexible, cheaper to produce and environmentally friendly.

The artificial synaptic TFT consisted of indium zinc oxide (IZO), as both a channel and a gate electrode, separated by a 550-nanometre-thick film of nanogranular silicon dioxide electrolyte, which was fabricated using a process known as chemical vapour deposition.

The design was specific to that of a biological synapse—a small gap that exists between adjoining neurons over which chemical and electrical signals are passed. It is through these synapses that neurons are able to pass signals and messages around the brain.

All neurons are electrically excitable, and can generate a ‘spike’ when the neuron’s voltage changes by large enough amounts. These spikes cause signals to flow through the neurons which cause the first neuron to release chemicals, known as neurotransmitters, across the synapse, which are then received by the second neuron, passing the signal on.

Similar to these output spikes, the researchers applied a small voltage to the first electrode in their device which caused protons—acting as a neurotransmitter—from the silicon dioxide films to migrate towards the IZO channel opposite it.

As protons are positively charged, this caused negatively charged electrons to be attracted towards them in the IZO channel which subsequently allowed a current to flow through the channel, mimicking the passing on of a signal in a normal neuron.

As more and more neurotransmitters are passed across a synapse between two neurons in the brain, the connection between the two neurons becomes stronger and this forms the basis of how we learn and memorise things.

This phenomenon, known as synaptic plasticity, was demonstrated by the researchers in their own device. They found that when two short voltages were applied to the device in a short space of time, the second voltage was able to trigger a larger current in the IZO channel compared to the first applied voltage, as if it had ‘remembered’ the response from the first voltage.

Corresponding author of the study, Qing Wan, from the School of Electronic Science and Engineering, Nanjing University, said: ‘A paper-based synapse could be used to build lightweight and biologically friendly artificial neural networks, and, at the same time, with the advantages of flexibility and biocompatibility, could be used to create the perfect organism–machine interface for many biological applications.’

MIT robot may accelerate trials for stroke medications

The development of drugs to treat acute stroke or aid in stroke recovery is a multibillion-dollar endeavor that only rarely pays off in the form of government-approved pharmaceuticals. Drug companies spend years testing safety and dosage in the clinic, only to find in Phase III clinical efficacy trials that target compounds have little to no benefit. The lengthy process is inefficient, costly, and discouraging, says Hermano Igo Krebs, a principal research scientist in MIT’s Department of Mechanical Engineering.

“Most drug studies failed and some companies are getting discouraged,” Krebs says. “Many have recently abandoned the neuro area [because] they have spent so much money on developing drugs that don’t work. They end up focusing somewhere else.”

Now a robot developed by Krebs and his colleagues may help speed up drug development, letting pharmaceutical companies know much earlier in the process whether a drug will ultimately work in stroke patients.

To receive approval from the Food and Drug Administration, a company typically has to enroll 800 patients to demonstrate that a drug is effective during a Phase III clinical trial; this sample size is determined, in part, by the accuracy of standard outcome measurements, which quantify a patient’s ability over time to, say, lift her arm past a certain point. A clinical trial can take several years to enroll appropriate patients, run tests, and perform analyses.

The study’s authors found that by using a robot’s measurements to gauge patient performance, companies might only have to test 240 patients to determine whether a drug works — a reduction of 70 percent that Krebs says would translate to a similar reduction in time and cost.

While pharmaceutical companies would still have to adhere to the FDA’s established guidelines and outcome measurements to receive final drug approval, Krebs says they could use the robot measurements to guide early decisions on whether to further pursue or abandon a certain drug. If, after 240 patients, a drug has no measurable effect, the company can pursue other therapeutic avenues. If, however, a drug improves performance in 240 robot-measured patients, the pharmaceutical company can continue investing in the trial with confidence that the drug will ultimately pass muster.

The researchers have published their results in the journal Stroke.

Creating a translator for stroke recovery

In their study, Krebs and his colleagues explored the robot MIT-Manus as a tool for evaluating patient improvement over time. The robot, developed by the team at MIT’s Newman Laboratory for Biomechanics and Human Rehabilitation, has mainly been used as a rehabilitation tool: Patients play a video game by maneuvering the robot’s arm, with the robot assisting as needed.

While the robot has mainly been used as a form of physical therapy, Krebs says it can also be employed as a measurement tool. As a patient moves the robot’s arm, the robot collects motion data, including the patient’s arm speed, movement smoothness, and aim. For the current study, the researchers collected such data from 208 patients who worked with the robot seven days after suffering a stroke, and continued to do so for three months.

The researchers created an artificial neural network map that relates a patient’s motion data to a score that correlates with a standard clinical outcome measurement.

The authors then selected a separate group of nearly 3,000 stroke patients who did not use the robot, but who went through standard clinical tests. In particular, the researchers calculated the “effect size” — the difference in patient performance from the beginning to the end of a trial, divided by the standard deviation, or variability, of improvement among these patients. To determine whether a drug works, the FDA will often look to a study’s effect size.

Using the robot-derived neural network map, the group calculated the effect size at twice the rate usually achieved with standard clinical outcome measurements, indicating that the robot scale demonstrated greater sensitivity in measuring patient recovery.

The study’s authors went one step further and performed a power analysis that determines the optimal sample size for a given technique, finding that the robot scale would require only 240 patients to determine a drug’s effectiveness — a reduction in sample size that would save a company up to 70 percent in time and cost.

“Such a savings would be fantastic,” says David Reinkensmeyer, a professor of physical medicine and rehabilitation at the University of California at Irvine. “Robotic measurements will help us identify promising treatments with smaller numbers of patients and provide better insight into the mechanisms of the treatments, so that we can target those mechanisms and improve the treatments.”

Currently, only a few stroke drugs are in the late stages of development. However, once a company reaches a Phase III clinical trial, Krebs says it may use the MIT-Manus robot as a more efficient way to evaluate the drug’s impact by employing the measurement techniques on a smaller group of patients.

Researchers reveal more about how our brains control our arms

Ready, set, go.

Sometimes that’s how our brains work. When we anticipate a physical act, such as reaching for the keys we noticed on the table, the neurons that control the task adopt a state of readiness, like sprinters bent into a crouch.

Other times, however, our neurons must simply react, such as if someone were to toss us the keys without gesturing first, to prepare us to catch.

How do the neurons in the brain control planned versus unplanned arm movements?

Krishna Shenoy, a Stanford professor of electrical engineering, neurobiology (by courtesy) and bioengineering (affiliate), wanted to answer that question as part of his group’s ongoing efforts to develop and improve brain-controlled prosthetic devices.

In a paper published today in the journal Neuron, Shenoy and first author Katherine Cora Ames, a doctoral student in the Neurosciences Graduate Program, present a mathematical analysis of the brain activity of monkeys as they make anticipated and unanticipated reaching motions.

Monitoring the neurons

The experimental data came from recording the electrical activity of neurons in the brain that control motor and premotor functions. The idea was to observe and understand the activity levels of these neurons during experiments in which the monkeys made planned or reactive arm movements. What the researchers found is that when the monkeys knew what arm movement they were supposed to make and were simply waiting for the cue to act, electrical readings showed that the neurons went into what scientists call the prepare-and-hold state – the brain’s equivalent of ready, set, waiting for the cue to go.

But when the monkeys made unplanned or unexpected movements, the neurons did not go through the expected prepare-and-hold state. “This was a surprise,” Ames said.

Before the experiment, the researchers had believed that a prepare-and-hold state had to precede movement. In short, they thought the neurons had to go into a “ready, set” crouch before acting on the “go” command. But they discovered otherwise in three variations of an experiment involving similar arm movements.

Experimental design

In all three cases, the monkeys were trained to touch a target that appeared on a display screen.

During each motion, the researchers measured the electrical activity of the neurons in control of arm movements.

In one set of experiments, the monkeys were shown the target but were trained not to touch it until they got the “go” signal. This is called a delayed reach experiment. It served as the planned action.

In a second set of experiments the monkeys were trained to touch the target as soon as it appeared. This served as the unplanned action.

In a third variant, the position of the target was changed. It briefly appeared in one location on the screen. The target then reappeared in a different location. This required the monkeys to revise their movement plan.

Monkey see, then monkey do

Ames said that, in all three instances, the first information to reach the neurons was awareness of the target.

“Perception always occurred first,” Ames said.

Then, about 50 milliseconds later, some differences appeared in the data. When the monkeys had to wait for the go command, the brain recordings showed that the neurons went into a discernable prepare-and-hold state. But in the other two cases, the neurons did not enter the prepare-and-hold state.

Instead, roughly 50 milliseconds after the electrical readings showed evidence of perception, a change in neuronal activity signaled the command to touch the target; it came with no apparent further preparation between perception and action. “Ready, set” was unnecessary. In these instances, the neurons just said, “Go!”

Applications

“This study changes our view of how movement is controlled,” Ames said. “First you get the information about where to move. Then comes the decision to move. There is no specific prepare-and-hold stage unless you are waiting for the signal to move.”

These nuanced understandings are important to Shenoy. His lab develops and improves electronic systems that can convert neural activity into electronic signals in order to control a prosthetic arm or move the cursor on a computer screen.

One example of such efforts is the BrainGate clinical trial here at Stanford, now being conducted under U.S. Food & Drug Administration supervision, to test the safety of brain-controlled, computer cursor systems – “think-and-click” communication for people who can’t move.

“In addition to advancing basic brain science, these new findings will lead to better brain-controlled prosthetic arms and communication systems for people with paralysis,” Shenoy said.