With neurons being grown on silicon chips, Michael Gross investigates the possibility of direct communication between man and machine.

With neurons being grown on silicon chips, Michael Gross investigates the possibility of direct communication between man and machine.

This summer, the ubiquitous Microsoft corporation announced that it had secured a US patent covering the use of the human body as a conductor in connection with electronic appliances. Newspaper readers were quick to respond with suggestions of possible consequences (’A fatal error occurred. Your body will be shut down. All unsaved blood may be lost’), but in reality it is far from clear what device exactly is to be plugged directly into its buyer, as the corporation acknowledges that it doesn’t have a specific product relating to that patent.

One researcher who has been studying possible connections between silicon electronics and biological cells for over two decades is Peter Fromherz, a director at the Max Planck institute for biochemistry at Martinsried near Munich, Germany. In 1985 he drew a cartoon showing an unhappy computer user interfacing with the machine via a keyboard on one side, and a much happier user with the wires from the computer plugged directly into his head on the other. The question inspiring this cartoon and Fromherz’s research is simple, almost naive: as both computers and brains communicate with electrical signals, why should it not be possible to create a direct interface between them, without the need for eyes and monitors, ears and speakers, hands and keyboards? If that computer is so clever, why can’t it just read my mind?

To address this challenge, Fromherz, who was then at the University of Ulm, set out to grow neurons from the medicinal leech ( Hirudo medicinalis) on silicon chips and persuade the two parties to talk to each other. Transmission of a signal from the neuron to the chip first succeeded in 1991, the reverse process four years later. Essentially, the recording of the neuron signal by the chip relies on a field effect transistor, while the electronic stimulation of the neuron arises from a voltage pulse applied to a capacitor, so both processes are absolutely non-invasive and don’t affect the survival of the cell in any way.

At the Max Planck institute, which Fromherz joined in 1994, he built on that pioneering research with the goals to establish the precise nature of the chip/neuron interface, expand the work to cells from other sources, and to build more complex systems consisting of neurons and semiconductors.

Regarding the mechanisms of the contact between chip and cell, Fromherz and his coworkers established that an ordinary silicon chip, with the outermost 15 nm oxidised, is an ideal substrate to cultivate neurons on. The silicon oxide layer insulates the two sides and stops any electrochemical charge transfer, which might damage the chip or the cell. Instead, there is only a capacitative connection, established by a so-called planar core-coat conductor. Proteins sticking out of the lipid membrane ensure that there is a thin (50-100 nm) conducting layer between lipid and silicon oxide, which constitutes the core of the conductor.

The whole arrangement can be represented by a simplified electrical circuit, where both the membrane and the silicon oxide have a defined capacitance, and the electrolyte layer between them (which is part of the medium surrounding the whole cell) has a given ohm resistance. The dynamic properties of the system are dominated by the ion channels within the cell membrane, which determine the ohmic conductance of the membrane and thus the propagation of the electrical action potential, which are the typical neuronal signals.

In the neuron-to-chip experiment, the current generated by the neuron has to flow through the thin electrolyte layer between cell and chip. This layer’s resistance creates a voltage, which a transistor inside the chip can pick up as a gate voltage that will modify the transistor current. In the reverse signal transfer, a capacitative current pulse is transmitted from the semiconductor through to the cell membrane, where it decays quickly, but activates voltage-gated ion channels that create an action potential.

Investigating the neuron-chip interface in more detail, Fromherz and his team established that the gap of up to 100 nm between the two is a natural consequence of the cell adhesion mechanisms mediated by membrane proteins. Using the interference patterns of light reflected by the silicon oxide/silicon layer, they showed that a naked lipid membrane glued onto a chip leaves only about 1 nm space, but living cells grown on this substrate will always leave a gap of at least 50 nm, so a more direct contact is not possible if the biological function of the membrane is to stay intact. On the other hand, neurons growing on chips may favour the researchers’ interest in placing their ion channels preferentially in the contact area. In a set of experiments involving the highly conductive maxi-K channel, they found that the density of channels per membrane area was one order of magnitude higher in the contact area than in the free membrane.

The next challenge was to move upwards from one neuron communicating with one stimulator or sensor to more complex neuro-electronic architectures, with the distant goal of getting entire neuronal networks plugged into electronics in a way that would allow their function to be studied in detail or use them for computational devices. As a first, elementary step from one neuron to networks, Fromherz and his team implemented a simple signalling pathway including information transfer from a chip to a neuron, and then onwards to a second neuron and back to the chip.

For this first hybrid circuit, they followed the lead of neurology pioneers such as Eric Kandel and used neurons from snails. These unappealing invertebrates are popular among neuroscientists, because their neurons are an order of magnitude larger than ours, and because circuits consisting of only a few cells can already display a measurable biological function. As a substrate to grow the cells on, the researchers designed a specific chip with 14 two-way junctions ( ie areas that can both send signals to neurons and receive signals back) arranged in a circle of about 200 μm diameter. Typically, they planted five to seven snail neurons onto such junctions and cultivated them for a few days, hoping that at least some would form electrical synapses with others.

The experiment succeeded in producing a few such pairs of linked neurons that could build a bridge between a signal emitter and receiver in the silicon chip. Earlier this year, the equivalent achievement was also reported with a chemical instead of an electrical synapse. However, the process was much too inefficient and random to enable the construction of well-defined larger networks.

If a complex and well-defined neuronal network cannot be generated on the chip directly, maybe the chip can be interfaced with a pre-existing network, for instance a brain? Following this line of research, Fromherz and Michael Hutzler have recently presented the first successful connection between a chip of the kind described above, containing capacitors to stimulate and transistors to sense nerve action, and a brain slice containing well-characterised neuronal connections.

Specifically, the researchers turned their attention to the rat hippocampus, a brain region associated with long-term memory. It is known that in this part of the rat brain, a region known as CA3 stimulates the CA1 to which it is connected by extensive wiring. Brain slices can be prepared such that the cut runs alongside the CA3 to CA1 connection and makes this entire communications channel accessible to experiments. Using such slices, Hutzler and Fromherz demonstrated that their chip can (via its capacitor) stimulate the CA3 region such that these brain cells pass on the signal to CA1, where it can be recorded with the chip’s transistors.

While similar stimulation and recording is possible with metal electrodes, the silicon chip method is the least invasive method available. In the first experiments with a relatively simple chip device, the spatial resolution remained low, but in principle, it can be improved to the size of features on commercial microchips, currently standing somewhere near 100 nm.

A significant step in this direction is the recent development of a CMOS (complementary metal-oxide-semiconductor) chip with an array of 128 × 128 sensors for neural recording packed into one square millimetre. This was achieved by researchers at the Munich-based company Infineon Technologies, in collaboration with Fromherz’s group. The chip can practically generate a movie of neurons in action: it delivers 16 kilopixels at 2000 frames per second. The pitch ( ie the distance from one sensor to the corresponding part of the next one) is 7.8 μm, which is very close to the typical width of a vertebrate neuron.

The involvement of chip maker Infineon with Fromherz’s sensor work shows that there are hopes for commercially valuable spin-off products, although at this point it is far from clear what they will look like. Probably, says Fromherz, ’the first applications will be in brain research and diagnostics.’ Sensors like the 16 kilopixel CMOS chip will enable researchers to fill the gap between studies involving only a few cells and those operating at larger scales like magnetic resonance imaging. Processes like associative memory, which have been broadly localised, could be studied in detail using similar non-invasive devices.

Prosthetic devices to restore vision, hearing or limb control might be the next step. On a crude level, artificial retinae with just a few pixels have already been demonstrated to create visible images. Further in the future, Fromherz sees that ’the real dreams would be the realisation of the brain-in-computer and chip-in-brain’ arrangements. Will that be the point where the boundaries between biology and technology disappear? ’It’s not a matter of technology or biology,’ says Fromherz. ’You take everything on your shelf to create something technologically useful, be it minerals, polymers, colloids, proteins, cells, or even tissues. It is a kind of super-chemistry.’

Computer makers may be the first who want a piece of this super-chemistry. ’The microelectronics community becomes interested in our work because they hope neurons may be a solution to the end of Moore’s law that is visible in about 10 years,’ says Fromherz, referring to the exponential growth of chip performance upheld over the last three decades. At the moment, it is far from clear what will happen when this trend hits the physical limits of what is possible with silicon chips. Alternative methods including quantum, molecular, and biological computers may allow computer development to go beyond this limit, so at the moment all these options have to be explored.

Recognising that Fromherz’s work may be the basis for crucial technologies of the future, the Philip Morris Foundation bestowed upon him its prestigious Research award for 2004, which also resulted in widespread press coverage. With that much attention, it can only be a matter of time before Microsoft discovers his bio-electronic hybrid systems and starts developing operating systems for them.


Michael Gross is science writer in residence at the school of crystallography, Birkbeck College, University of London. He can be contacted via "the prose and the passion" website.