Circuit City

This week something rather exceptional has occurred in the realm of microelectronics: a group of physicists working at HP Labs have built a new and revolutionary kind of circuit etched onto a microchip. This is highly unusual as most advances in this field tend to be fast-paced but rather incremental in nature. Here, we are talking about a completely different animal in the zoo of all those microscopic circuits that populate the chips in our PC's and the beast is called a memristor, condensed word for 'memory' and 'resistor'.

So far to build their microprocessors electrical engineers have had the choice of three different kinds of elementary circuits: resistors, capacitors and inductors, together with the now-ubiquitous transistor (this does not count as an elementary circuit since its behavior is non-linear.) Each of these tiny elementary circuits controls two different properties of the electrical current flowing through it-according to a linear differential relation.

For instance, middle-school students (in France at least) learn in physics class what is known as Ohm's law: dv = R di, where v is the voltage of the current and i it's intensity. The factor of R is called the resistance of any conducting material, and resistors are commonly used to control the ratio of voltage to intensity within a circuit. Similarly, capacitors relate the amount of charge q that passes through it with voltage, through the relation dq = C dv (C is called the capacitance of the capacitor), and inductors relate the magnetic flux to current intensity through dφ = L di (where the constant L is called-you guessed it-inductance).

About fourty years ago, Leon Chua, then a graduate student in electrical engineering at Berkeley figured that, for the sake of symmetry (see picture above), there ought to exist a fourth type of passive component, which would relate the magnetic flux φ to charge through the relation dφ = M dq, where the constant M would be called the memristance of the circuit. Yet the physical existence of the memristor proved elusive and no one had ever managed to produce one using simple materials until this past week.

The research team at HP Labs built their prototype device by sandwiching two layers of titanium oxide between two perpendicular electrodes (see picture above). The upper layer of titanium oxide is missing a few oxygen atoms, on the order of one percent, while the lower layer is regular titanium oxide. When a current is forced through the two electrodes, it pushes the charged voids down to the lower layer, thereby lowering the average resistance of the circuit. This can then be reversed by flowing current in the opposite direction.

But you may ask, this is all well but what is this new gizmo good for? well, my friends, this is good for many things. In particular, a memristor must have the interesting property that, in fact, its electrical resistance changes over time proportionally to the amount of charge that passes through it. So as its name indicates, a memristor acts like a memory component and precisely, electrical engineers are now thinking about using it for that purpose: memristors would provide a new kind of a very dense and also persistent memory (i.e. it can retain information even when the current is turned off, unlike our current PC's RAM chips.)

Even more interestingly, memristors may very well bring back into fashion analog circuits for all kinds of signal processing applications. In fact, scientists have already pointed out the similarities in the way memristors 'remember' their level of electrical activity and the way the human brain's neurons and synapses function. So memristors would make it possible to implement very dense neural networks in silicon, as opposed to simulating them in software on existing digital computers.

For decades, electrical engineers have been able to continuously shrink the size of the elementary circuits and transistors that they etch onto the surface of microprocessors, an evolution of technology now called 'Moore's law', after Gordon Moore the founder of chipmaker Intel who first came up with the observation that the number of transistors on our PC's microprocessors roughly doubles every two years. But now that the feature size of a transistor on a microchip is closing in on the size of a few hundred atoms (currently they're at 0.18 μm so about 1500 atoms), the performance of current transistor technology is running into its physical limits. Thankfully, memristors might eventually replace transistors completely as they are getting smaller, especially since the memristance effect gets better as the circuit gets tinier because it involves moving around oxygen atoms inside its structure-the less far they have to move the faster they get there.