Self-programming logic circuits
It’s been a long time — too long? — since I’ve really taken the opportunity to mess with circuit design and the associated calculations, but I definitely fall into the category of people that is described at the beginning of this article: “if you’ve spent any time in an electrical engineering classroom, you probably only have experience with three [basic circuit components]: capacitor, inductor, and resistor.”
There is, of course, a fourth basic circuit element, as the sentence above implies — the memristor. Memristors only really existed as theoretical elements since being proposed in 1971 by Berkeley professor Leon Chua, until May of last year, when a handful of techs at HP managed to fabricate one.
The theoretical understanding of a memristor predicted that the current moving through a memristor would be proportional to the flux of magnetic field that had flowed through the material. This caused scientists to look for memristor behavior in magnetic materials, but it turned out they were looking in all the wrong places. HP found resistance in TiO2 thin-film multilayers could be controlled by varying an applied voltage and, critically, that the resistance change was non-volatile–the hallmark of a memristor.
The key to this behavior is creating alternating layers of high-quality (high resistance) and defect-rich (low resitance) TiO2 thin films. Charge transport in TiO2 is dominated by O2- conduction, instead of the electrons of most semiconductor devices. Applying a voltage between the layers causes the O2- defects to diffuse into the low defect region, reducing the device’s overall resistance. By applying a reverse bias, the vacancies diffuse back into the defective layer and the resistance returns to its original state.
It’s all very exciting, of course…but equally, most people who heard news of the discovery probably dismissed it just as quickly, wondering “so what?” Scientists and engineers who work with electronics would have cause to be fascinated, but what did the fabrication of memristors actually mean? What could be done with a memristor?
Well, it turns out there’s an answer to that question now, too: memristors can be — and, now, have been — used to fabricate logic circuits which can self-program. Okay, nothing quite on the order of a full-fledged AI (indeed, nothing even close), but still pretty amazing:
…[a] memristor grid was programmed to perform a Boolean sum-of-product operation. The resistance of each memristor junction was mapped using a probe station and a logic circuit was designed based on these measurements. ON memristors were programmed by applying a 4.5 V bias using the contacting nanowires, while OFF memristors were programmed with a 2.2 V bias, and logic operations were performed between 0 and 1 V to prevent accidental de-programming of the circuits. The circuits successfully performed the sum-of-product operation at 2.8 kHz.
The full potential of memristors was demonstrated when the devices were made to actively re-program themselves. Dynamic reprogramming was achieved by linking the output of the sum-of-products circuit described above to another memristor inside the device. Instead of simply returning a value of 1 or 0 (voltages of 0.3V or 1.45V, respectively) the voltage was applied to another memristor in the system. This voltage reprogrammed the target memristor to the ON position. In this way, calculations carried out in the device can reprogram circuits in other areas, effectively allowing the device to reprogram itself and adapt to different situations. While the memristor logic circuit may not have become self aware or searched out Sarah Connor, the result was extremely impressive, and it opens doors to exciting new systems.
The ability to introduce in-built error correction into digital systems is one of those things that, if one considers it, is simply staggering for its scale. Self-correcting, fault-tolerant hardware and flash memory that can detect and correct data errors as they happen — to name just two possibilities — would alone be worth perhaps unquantifiable sums of money given their applicability to industry and sensitive data.
If, that is, the solution that HP came up with can be improved upon. The current test model, as noted above, only operates at 2.8 kHz, which isn’t exactly speedy.
Still: very cool. At least to a guy like me.
No Comments Comments Feed
Sorry, the comment form is closed at this time.
The comments are closed.