There can be few more potent ideas in futurology and science fiction than that of the brain chip – a direct interface between the biological information processing systems of the brain and nervous system and the artificial information processing systems of microprocessors and silicon electronics. It’s an idea that underlies science fiction notions of “jacking in” to cyberspace, or uploading ones brain, but it also provides hope to the severely disabled that lost functions and senses might be restored. It’s one of the central notions in the idea of human enhancement. Perhaps through a brain chip one might increase ones cognitive power in some way, or have direct access to massive banks of data. Because of the potency of the idea, even the crudest scientific developments tend to be reported in the most breathless terms. Stripping away some of the wishful thinking, what are the real prospects for this kind of technology?
The basic operations of the nervous system are pretty well understood, even if the way the complexities of higher level information processing work remain obscure, and the problem of consciousness is a truly deep mystery. The basic units of the nervous system are the highly specialised, excitable cells called neurons. Information is carried long distances by the propagation of pulses of voltage along long extensions of the cell called axons, and transferred between different neurons at junctions called synapses. Although the pulses carrying information are electrical in character, they are very different from the electrical signals carried in wires or through semiconductor devices. They arise from the fact that the contents of the cell are kept out of equilibrium with their surroundings by pumps which selectively transport charged ions across the cell membrane, resulting in a voltage across the membrane. This voltage can be relaxed when channels in the membrane, which are triggered by changes in voltage, open up. The information carrying impulse is actually a shock wave of reduced membrane potential, enabled by transport of ions through the membrane.
To find out what is going on inside a neuron, one needs to be able to measure the electrochemical potential across the membrane. Classically, this is done by inserting an electrochemical electrode into the interior of the nerve cell. The original work, carried out by Hodgkin, Huxley and oters in the 50’s, used squid neurons, because they are particularly large and easy to handle. So, in principle one could get a readout of the state of a human brain by measuring the potential at a representative series of points in each of its neurons. The problem, of course, is that there are a phenomenal number of neurons to be studied – around 20 billion in a human brain. Current technology has managed to miniaturise electrodes and pack them in quite dense arrays, allowing the simultaneous study of many neurons. A recent paper (Custom-designed high-density conformal planar multielectrode arrays for brain slice electrophysiology, PDF)) from Ted Berger’s group at the University of Southern California shows a good example of the state of the art – this has electrodes with 28 µm diameter, separated by 50 µm, in an array of 64 electrodes. These electrodes can both read the state of the neuron, and stimulate it. This kind of electrode array forms the basis of brain interfaces that are close to clinical trials – for example the BrainGate product.
In a rather different class from these direct, but invasive probes of nervous system activity at the single neutron level, there are some powerful, but indirect measures of brain activity, such as functional magnetic resonance imaging or positron emission tomography. These don’t directly measure the electrical activity of neurons, either individually or in groups; instead they rely on the fact that thinking is hard work (literally) and locally raises the rate of metabolism. Functional MRI and PET allow one to localise nervous activity to within a few cubic millimeters, which is hugely revealing in terms of identifying which parts of the brain are involved in which kind of mental activity, but which remains a long way away from the goal of unpicking the brain’s activity at the level of neurons.
There is another approach does probe activity at the single neuron level, but doesn’t feature the invasive procedure of inserting an electrode into the nerve itself. These are the neuron-silicon transistors developed in particular by Peter Fromherz at the Max Planck Institute for Biochemistry. These really are nerve chips, in that there is a direct interface between neurons and silicon microelectronics of the sort that can be highly miniaturised and integrated. On the other hand, these methods are currently restricted to operate in two dimensions, and require careful control of the growing medium that seems to rule out, or at least present big problems for, in-vivo use.
The central ingredient of this approach is a field effect transistor which is gated by the excitation of a nerve cell in contact with it (i.e., the current passed between the source and drain contacts of the transistor strongly depends on the voltage state of the membrane in proximity to the insulating gate dielectric layer). This provides a read-out of the state of a neuron; input to the neurons can also be made by capacitors, which can be made on the same chip. The basic idea was established 10 years ago – see for example Two-Way Silicon-Neuron Interface by Electrical Induction. The strength of this approach is that it is entirely compatible with the powerful methods of miniaturisation and integration of CMOS planar electronics. In more recent work, an individual mammalian cell can be probed “Signal Transmission from Individual Mammalian Nerve Cell to Field-Effect Transistor” (Small, 1 p 206 (2004), subscription required), and an integrated circuit with 16384 probes, capable of probing a neural network with a resolution of 7.8 µm has been built “Electrical imaging of neuronal activity by multi-transistor-array (MTA) recording at 7.8 µm resolution” (abstract, subscription required for full article).
Fromherz’s group have demonstrated two types of hybrid silicon/neuron circuits (see, for example, this review “Electrical Interfacing of Nerve Cells and Semiconductor Chips”, abstract, subscription required for full article). One circuit is a prototype for a neural prosthesis – an input from a neuron is read by the silicon electronics, which does some information processing and then outputs a signal to another neuron. Another, inverse, circuit is a prototype of a neural memory on a chip. Here there’s an input from silicon to a neuron, which is connected to another neuron by a synapse. This second neuron makes its output to silicon. This allows one to use the basic mechanism of neural memory – the fact that the strength of the connection at the synapse can be modified by the type of signals it has transmitted in the past – in conjunction with silicon electronics.
This is all very exciting, but Fromherz cautiously writes: “Of course, visionary dreams of bioelectronic neurocomputers and microelectronic neuroprostheses are unavoidable and exciting. However, they should not obscure the numerous practical problems.” Among the practical problems are the fact that it seems difficult to extend the method into in-vivo applications, it is restricted to two dimensions, and the spatial resolution is still quite large.
Pushing down to smaller sizes is, of course, the province of nanotechnology, and there are a couple of interesting and suggestive recent papers which suggest directions that this might go in the future.
Charles Lieber at Harvard has taken the basic idea of the neuron gated field effect transistor, and executed it using FETs made from silicon nanowires. A paper published last year in Science – Detection, Stimulation, and Inhibition of Neuronal Signals with High-Density Nanowire Transistor Arrays (abstract, subscription needed for full article) – demonstrated that this method permits the excitation and detection of signals from a single neuron with a resolution of 20 nm. This is enough to follow the progress of a nerve impulse along an axon. This gives a picture of what’s going on inside a living neuron with unprecendented resolution. But it’s still restricted to systems in two dimensions, and it only works when one has cultured the neurons one is studying.
Is there any prospect, then, of mapping out in a non-invasive way the activity of a living brain at the level of single neurons? This still looks a long way off. A paper from the group of Rodolfo Llinas at the NYU School of Medicine makes an ambitious proposal. The paper – Neuro-vascular central nervous recording/stimulating system: Using nanotechnology probes (Journal of Nanoparticle Research (2005) 7: 111–127, subscription only) – points out that if one could detect neural activity using probes within the capillaries that supply oxygen and nutrients to the brain’s neurons, one would be able to reach right into the brain with minimal disturbance. They have demonstrated the principle in-vitro using a 0.6 µm platinum electrode inserted into one of the capillaries supplying the neurons in the spinal cord. Their proposal is to further miniaturise the probe using 200 nm diameter polymer nanowires, and they further suggest making the probe steerable using electrically stimulated shape changes – “We are developing a steerable form of the conducting polymer nanowires. This would allow us to steer the nanowire-probe selectively into desired blood vessels, thus creating the first true steerable nano-endoscope.” Of course, even one steerable nano-endoscope is still a long way from sampling a significant fraction of the 25 km of capillaries that service the brain.
So, in some senses the brain chip is already with us. But there’s a continuum of complexity and sophisitication of such devices, and we’re still a long way from the science fiction vision of brain downloading. In the sense of creating an interface between the brain and the world, that is clearly possible now and has in some form been realised. Hybrid structures which combine the information processing capabilities of silicon electronics and nerve cells cultured outside the body are very close. But a full, two-way integration of the brain and artificial information processing systems remains a long way off.