Computer circuits that require fewer components than our existing technology employs, combine several functions and are capable of 'self-learning' have been fabricated for the first time.
Researchers at Hewlett-Packard Laboratories in Palo Alto, California combined memristors with transistors in a hybrid circuit array to demonstrate conditional self-programming and show that just a few such device elements can be configured to act as logic, switching and memory components simultaneously. The use of fewer circuit elements offers the benefits of smaller circuit size and lower power consumption.
The term 'memristor' means 'memory resistor', the fourth type of passive circuit element in addition to the (fixed) resistor, the capacitor and the inductor. Having been predicted by theory in 1971, the first memristor device wasn't fabricated until 2008. The memristor is a two-terminal circuit element that changes its resistance in response to the positive or negative polarity of the voltage applied to it or the amount of current flowing through it.
As passive devices, memristors must be combined with active elecronic devices such as transistors, which can amplify or switch electronic signals, in order to become useful in analog or digital circuits.
Due to their memory capability, memristors are also programmable. By routing the output of a logic circuit containing memristors back into those same memristors, the circuit can even be made to reconfigure itself, thus in effect self-programming itself.
Stan Williams of HP told Physorg.com, 'It actually takes at least a dozen transistors to mimic the electrical properties of a single memristor.' He explained that building logic circuits incorporating fewer memristors in place of larger numbers of transistors might offer logic designers opportunities to develop circuit capabilities while decreasing the total number of devices required, concluding, 'Thus, it may be possible to continue the equivalent of Moore's law for a couple of generations not by making transistors smaller, but by replacing some subset of them with memristors.'
In their demonstration, the HP research team fabricated a crossbar array consisting of two sets of 21 parallel 40nm-wide wires arranged perpendicular to each other, between which a 20nm-thick film of the semiconductor titanium dioxide (TiO2) was layered to fabricate a memristor at the intersection of each crossed pair of wires. To complete the combined memristor-transistor circuit, the researchers connected a peripheral array of field effect transistors to the crossbar array of memristors with metal traces.
The basic nature of a memristor is that the resistance of the device can be changed and then persist. In the memristors that the HP researchers fabricated, the memory 'switch' is physically manifested by the movement of positively charged oxygen ions, which are the dopants in a semiconducting TiO2 layer. A positive bias voltage pushes these 'holes' away from an electrode and increases the resistance, whilst a negative bias attracts the ions toward the electrode and decreases the resistance. If not subjected to either a positive or a negative voltage, the programmed state of the memristor will remain in force for at least a year.
The researchers exercised the hybrid circuit they created by performing a basic logic function (AB + CD) from four different voltage inputs representing the four logic values, applied across two different row and column pairs of the memristor crossbar array. They routed the output results through the transistors to amplify the signals and directed the outputs back into the memristor array to change their resistance values, thus using the results of the logic function to reprogram the memristors.
'Self-programming is a form of learning,' HP's Williams said. 'Thus, circuits with memristors may have the capacity to learn how to perform a task, rather than have to be programmed to do it.' To put it another way, the modification of subsequent behaviour in response to feedback received might be viewed as the essence of learning. At a very basic level, this is how nerve synapses in a brain develop.
HP's researchers hope that their demonstration of a hybrid memristor-transistor device will spur progress in integrating memristors with conventional circuits. They also believe that their demonstration of a prototype electronic circuit that can perform self-programming might lead to various new design endeavors, such as efforts to develop adaptive synaptic circuits.
Such advances in the basic technologies underlying electronic logic devices and design patterns could conceivably advance not only computing and information technology, but also the state of the art in artificial intelligence.
Thus self-programming hybrid memristor-transistor circuitry may hold great promise to help us extend Moore's Law. Even more importantly perhaps, it might also help light the path toward creating truly intelligent machines.
We look forward to greeting our intelligent machine overlords. X
Check out
Physorg.com