Samuel Greengard

Author / Journalist / Speaker

Neuroscientist Theodore Berger is working real hard to be the first person to implant microchips between your ears.

Clad in slacks and a dress shirt, Theodore Berger sits in his cluttered office, surrounded by the flotsam and jetsam of his work: modems, Zip drives, electrostatic bags, manuals, floppy disks, white cotton gloves, books. The intense neurobiologist is talking neurons and silicon chips, dendrite tissue and advanced photonics. Beneath the wall-mounted photographs of brain tissue and schematic diagrams of VLSI chips sit a pair of Gateway 2000 Pentium PCs. Their hard plastic outer cases have been removed, the inside electronics ready for quick exploration. It's an unmistakable metaphor for a guy who finds the bony cranium an incidental obstacle to studying gray matter.

During the past four years, Berger and a seven-member team of scientists - experts in everything from semiconductors to the neurophysiology of tissue cultures - have worked ceaselessly at the University of Southern California in Los Angeles to unlock the brain's complex mathematical model so they can bridge the gap between silicon and cerebrum. "If we can speak to the brain in its own language then we begin to understand the biological basis of thought and learning," says the 46-year-old scientist. "It's a matter of understanding the way neurons act and react in every conceivable situation, then building a device that can duplicate their basic functions and transmit the appropriate electrical impulses."

If Berger can link the two - computer and brain - he may create a parallel- processing network that could function as a brain implant. Such a device would restore physical and mental functions lost to stroke, head trauma, Alzheimer's, epilepsy, and an array of other maladies. His plan to create a bionic brain is bold, brash, and just a bit, well, mind-blowing. Berger - scientist, tinkerer, dreamer - wants to be the man who implants microchips between your ears. And the amazing thing is that he just might succeed.

Simple problem, complex solution

Understanding the human brain is a formidable task. Parked inside the average cranium are between 1 and 10 billion neurons. Extensions called dendrites relay electrical impulses to other neurons, the brain cells that serve as the basic building blocks of thought and activity. The electrical impulses are carried on fibers called axons. The result is a neural network of remarkable complexity.

The power of neurons is that they can work together in different ways to produce speech, vision, hearing, and thought. Only when something goes terribly wrong are we aware of how serious it is to lose these connections. Only then are we able to recognize how difficult it is to regain what's lost and how much therapy may be required.

Neuroscience, a field that didn't exist until the mid-1960s, has struggled to map the human brain and comprehend how all the pieces fit together. It's analogous to a space alien dissecting computers, robots, and other devices stuffed with semiconductors. Without a road map, it would be virtually impossible to know what each chip does and how it interacts with other microprocessors.

A decade ago, when Berger began studying "wet" brain tissue taken from animals, he saw how little was known about the way neurons process information, or how populations of neurons can work together. Although scientists know a great deal about the functioning of the individual nerve cells that compose the hippocampus - the part of the brain that oversees memory - there's still no clear idea of how their interaction produces the cognitive function known as learning. "A great deal of classical neurophysiology is devoted to trying to understand how the neurons process the signal and what the output represents," he explains.

That's not a nagging concern for Berger. Instead, he spends 12 to 14 hours a day identifying and cataloging predictable and repeatable neural electrical patterns, continuing a pattern of neural/technical intervention that has been explored for decades. In the 1950s, for instance, a stir was caused by a famous experiment on rats by J. Olds and P. Milner. Electrodes were implanted in each rat's lateral hypothalamus and connected to a stimulator that could be activated by pressing a small button. Since the electrical stimulation caused feelings of pleasure, the rats soon learned to stimulate themselves, choosing that button rather than another that would provide them with food, even when they were hungry.

A decade later, American neurophysiologist J. M. R. Delgado used stereotactic instruments to implant microelectrodes in precisely targeted areas within the brain. Their signals were transmitted wirelessly, making it possible to send high-voltage current. That led to drastic changes in emotional, sexual, and social behavior.

In one dramatic demonstration, Delgado implanted an electronic receiver in the medial region of the hypothalamus of a bull and then challenged the animal to do battle in a bullfighting ring. Delgado stopped the charging bull at the last minute by activating electrodes with a radio transmitter.

More recent experiments by researchers like Caltech's Carver Mead have shown that the elementary operations found in the nervous system can be realized by analog circuits created with standard silicon fabrication technology. Many neural areas are organized as thin sheets and carry two-dimensional representations of their computational space. These structures map well onto the two-dimensional silicon surface. Coincidentally, in both neural and silicon technologies, the active devices (synapses and transistors) occupy no more than a small percentage of the space - "wire" fills the remaining area.

Berger and fellow team member Vasilis Marmarelis, a professor of biomedical engineering, have built on those insights through an elaborate game of copycat, developing a complicated mathematical model that precisely replicates the input and output signals that a natural neural system uses. Just as a basketball player doesn't require a degree in advanced trigonometry to toss a jump shot through a hoop, Berger doesn't need to unlock the vast secrets of the brain to make an implant work.

Wander into Berger's lab and you begin to see where he's going with all this. Alongside a workbench covered with beakers and tubes, a researcher (in the finest tradition of a Carnegie Deli chef) slices human brain tissue into ultrathin sections. These samples are then inserted into test tubes filled with oxygen and artificial cerebral-spinal fluid, which keeps the fresh tissue alive for several hours. The sample is transported to another part of the lab, where a series of electrodes and probes are inserted. This leads to a smorgasbord of electronic amplifiers and oscilloscopes.

For the next few hours, the tissue is subjected to a series of electronic impulses designed to simulate every possible combination that the brain can throw at it. A Pentium computer charts the data, creates a unique profile for each piece of tissue, and then slots the information into an enormous database.

Berger & Co. repeat this process thousands of times. It's not enough to know how a piece of brain tissue reacts to different circumstances. They must understand the range of responses to the same stimulus at different times.

Hardwiring the head

Constructing a brain implant is more than a meat-and-test-tube operation, however. Even with a growing understanding of the mathematics involved in driving such a system, today's digital microprocessors are not fast enough to execute the complex instructions in real time. Minutes, hours, even days could elapse before a software-driven brain implant could process even the most basic task, such as moving a finger that has been paralyzed by stroke or restoring a fragment of memory lost to Alzheimer's. Imagine trying to surf the Web on a 300-bps modem or commute to work in a go-cart. Possible? Yes. Practical? Definitely not.

That's where silicon and hybrid analog/digital microchips enter the picture. By hardwiring the algorithms that represent, say, 100 neurons onto a microchip, it is possible to rapidly accelerate the data processing. A calculation that might have taken 10 seconds or more in the past is now completed in only about 100 nanoseconds. What's more, these neural microchips can operate as parallel-processing devices. But there's still a long way to go. Putting the function of perhaps 100 or 200 neurons on a single chip and then hardwiring tens of thousands of chips together would replicate only a fraction of the 4 or 5 million neurons in the hippocampus. You'd need an implant device the size of a pickup to get anywhere near the processing power of the brain.

Armand R. Tanguay Jr., director of the Center for Neural Engineering at USC and a key member of the team, hopes to solve the scaling and bandwidth problems through a combination of holography and laser optics. By using light signals to replace physical wiring, it's possible to tightly stack 1-by-1-centimeter chips, which can then communicate with each other. This tiny parallel-processing network potentially allows for real-time response, and might eventually fit into a space the size of a peanut or, more important, the human skull.

Such an implant could be tailored to replace a lost ability in humans. "Because the brain is such an adaptive organ," Berger says, "the device wouldn't have to precisely mimic the lost functions. If the brain is provided with a basic set of instructions, it would fill in the blanks and regain the lost functions. Over days and weeks, the person might make a full recovery."

It's tough, time-consuming work, but engineering an implant from the silicon up is a bit easier when you've got your own photonic fabrication facility on the premises. Just a few hundred feet from Berger's lab is a US$10 million setup that has the ability to make the photonic elements that "wire" the silicon chips together, as well as the multielectrode probes that act as an interface between living and man-made neural circuits.

Surrounded by an array of equipment that would seem more fitting in Silicon Valley than downtown LA - including a bonding machine and clean room that can filter down to 10 particles per cubic foot - assistant professor Chris Kyriakakis shows off a hybrid optical electronic device that contains a set of stacked chips. "It has the built-in electronics to power the entire device," he explains. "The first chip drives another optical chip that can modulate light. Ultimately, that's what runs the tiny computer network."

The final and most difficult step is placing such a device inside the skull and connecting it to a living brain. Crosswiring such wildly divergent universes - one brains, the other bytes - isn't for the faint of heart or the intellectually timid. It's certainly not difficult to let your imagination run amok and conjure up an image of some crazed zombie straight out of a Boris Karloff flick. But Berger is no Boris. Although Berger eschews much of the conventional wisdom that permeates the rarefied field of neuroscience - particularly traditional reductionist thinking that says every piece of the puzzle must be fully understood in order to build a scientific model - he also knows that it's crucial to retain credibility within the discipline by adhering to accepted standards and practices.

You might call his attitude skeptically reverent. "History is littered with the debris of firmly established explanations that have suddenly been overturned by new information," he says. "There are countless cases where humanity believed that an explanation was absolutely right, and then someone else came along with a radically different explanation that washed the old theory down the drain. So, it is very difficult to accept the truth of the day. It's crucial to look at various perspectives and explanations, to always keep the big picture in mind."

Berger seems to do that pretty well, though his approach perturbs some in the scientific community. Having grown up in a household that placed a high premium on introspection and accomplishment ("Not necessarily the accomplishment that comes with dollars, but the accomplishment of having an impact on the people around you," he explains), and with a father who helped pioneer transistor research at IBM, he realized early on that he wanted to make a real difference in the world.

Berger studied mathematics and psychology at college but shied away from a career in psychology, opting instead for the hands-on world of neuroscience. "I realized that psychology doesn't have the tools to fully understand the cause and effect of behavior," he says. "It is a very incomplete part of the puzzle."

Fast-forward a couple of decades, and Berger's life puzzle is mostly assembled. He lives in a pleasant split-level home overlooking the bluffs and beaches of Palos Verdes, commuting the 40 minutes to work in a 1992 Honda Accord. He spends spare moments gardening or tinkering with computers; both allow him to further explore elaborate relationships as well as evolution. Then there's his wife and his 7-year-old daughter, the latter serving as his most prized experiment. "Watching her brain develop is fascinating," he observes. "You realize that there are some things you can influence and some things you can't. You can try to teach something to someone, but if the brain isn't ready or willing to accept it, it ain't goin' in."

Which is precisely the challenge of Berger's odyssey into the mind. Dropping a functioning brain implant into a person's head and expecting it to work requires a voyage beyond the flat earth of today's neuroscience. For starters, the shape of the hippocampus doesn't remotely resemble a computer board, so inserting a conventional microchip into the brain and trying to send electrical waves back and forth would be like speaking into a telephone that's not connected to anything. Berger has circumvented this problem with a digital interface chip that will serve as the gateway between the brain's cells and the stacked set of chips that creates a group of artificial neurons.

Since transistors and neurons both send and receive electrical signals, it all boils down to arranging the chip's conductors in a pattern that matches the layout of neurons in a particular part of the brain. The brain then believes it's receiving its own signals, when in fact they're originating from the computer network. "A transistor can send electricity and so can a neuron," Berger says. "So, instead of using a computer chip that has transistors on it, you build a chip with a piece of exposed metal. You place the metal right next to the neuron, so that it can sense the electrical signals sent out by the neuron and send signals to the neuron." By slightly overbuilding the device - designing a layer of electrodes that are perhaps 30 microns thick compared with the cell's layer of 20 microns - Berger increases the odds of a successful contact.

"The ever present scaling issue"

In almost all cases, electrical signals trigger the release of chemical reactions in the brain, which in turn stimulate other cells to produce electrical reactions. These result in additional chemical activity. In fact, Berger points out that cells cannot talk toeach other without the release of chemicals that effectively bridge the gap between neurons. So, the brain implant would need only to stimulate electrical activity to work. Many chemical imbalances would still be treated by medication, which allows the brain to restore the proper electrical activity.

With all the parts in place, Berger will have essentially created a circuit between the device and the gray matter that operates much like a dedicated phone line. Brain cells could actually grow on top of the silicon or metal surfaces, further blurring their distinction. A pair of the team's scientists - Michel Baudry, a professor of biological sciences, and Roberta Brinton, an associate professor of molecular pharmacology - have begun developing such cultures with animal tissue. When their research is finalized, says Berger, "We will have created a physical connection, or synapses, between the brain cells and computer chips. We have the basic ingredients for a brain replacement."

The most remarkable aspect of all, says fellow team member Tanguay, is that such a device represents the sheer oddity of today's science. Increasingly, researchers are able to find solutions without fully understanding the underlying process. "We may never really know how the brain works, but we might be able to understand enough about how neurons correlate and work together to run the mathematical model," he explains. "By connecting a silicon neural circuit to a probe chip and allowing it to 'learn' and then feed data back, it's possible to diagnose and fix the problem without having a clue how the system achieved the results. The brain would use its natural adaptive abilities to reestablish all the correct connections."

In other words, the device could learn the brain's native language and speak it, but not interpret it to the outside world.

The research is also yielding data that can be plugged into an array of other, seemingly divergent, scientific disciplines. Says Joel Davis, program manager at the Office of Naval Research: "Once you understand the algorithms that drive the brain, you can apply them to all sorts of real-world problems that require computing. You can use reverse engineering to solve problems in communications, sonar, and neural networks."

One spin-off has been a new type of parallel-processing network that could dramatically improve the quality of wireless and cellular communication. By embedding an existing algorithm in a microchip that mimics a biological model, it is possible to drastically reduce the effects of noise and distortion. Yet another prototype design, which used the same biological processing model, promises to improve speech recognition.

Of course, connecting all the dots on the brain implant remains the team's central focus - and their enormous challenge. For one thing, there's a dizzying array of experimental technologies that must intertwine and complement one another perfectly. It's one thing to create computer simulations onscreen in a highly controlled environment, but it's quite another to make them work in the real world of chemistry, biology, and neurophysiology. Stimulating a neuron or two in the brain is a pretty good achievement but mere child's play compared with stimulating thousands of neurons and running electrical channels through each.

Then there's the profound question of how many artificial neurons scientists can fit on a microchip, and how many chips they can pack into a cranium. "The ever present scaling issue," as Tanguay puts it. Even the simplest prototype versions of the device will require far more exotic and sophisticated chips and interconnections than the group has so far managed to design and build. Alas, there's also the humbling reality that if any member of the highly skilled research team drops out or dies, the project could quickly crumble. So far, the team has steered remarkably clear of ego and turf battles, but the irony remains inescapable - the development of what could be the most advanced parallel-processing network on earth hinges on the most basic and tenuous: fellow humans.

Berger believes the team is about five years away from designing a brain implant for animals and about 10 to 15 years away from the first device for humans. With custom microchip designs taking weeks or months and other technical hurdles at every turn, it's certainly not a project for anyone with less than the patience of Buddha and the persistence of someone who sells insurance.

That suits Berger just fine. Nobody ever said that building an electronic brain would be easy, and it's clear that he's just as infatuated with the process as with the thought of changing people's minds. "You build it neuron by neuron and chip by chip," he says. "You enjoy each experiment and piece of the puzzle while keeping a focus on the bigger picture." Berger won't rest until he has built a bionic brain. He definitely wants to get inside your head.