Simply stated, memristors are resistors with memory. The measured resistance of the device is proportional to the charge that was last imparted on it. Leon Chau first described the concept in 1971, and a semiconductor device was actually fabricated by Stanley Williams at HP Labs in 2008. Advocates of memristors suggest that the device represents a fourth type of passive component, in addition to inductors, capacitors and ordinary resistors.
So, how do memristors actually work? The way the device operates is that if current follows in one direction for a finite amount of time, resistance decreases, and if it flows in the other direction, resistance increases. Most importantly, when no more current flows through the device, the device retains the resistance value it had at that point, displaying that value when voltage is again applied.
Because the resistance that the memristor records will not change when the voltage is turned off, the device can, in effect, serve as the basis of non-volatile memory systems. And, unlike today’s memory devices, which can only hold one of two values – on or off,
Memristors can hold a multitude of resistance values. This property can, at the very least, serve as the basis for a new type of device memory. It may also have profound implications for future computer architectures.
Building Memristors
The original memristors device that was built at Hewlett Packard was fabricated with titanium dioxide as its “memristance” material, but many groups today are building the devices using other metal oxides. The good news is that this type of work can be done with exactly the same equipment using the same standard photolithography processes now being used to build ICs around the world with very few modifications. It is also worth noting that memristors have already been implemented alongside CMOS devices, on the same silicon substrate.
The picture below illustrates HP’s TiO2 memristor, its equivalent circuit and its symbol. There is a very resistive undoped region next to a doped region, which is highly conductive. As charge flows through the device, the regions shift borders, changing the overall resistance between the two electrodes on either edge. When the voltage switches off and charge is no longer imparted, the overall resistance stays at the same value. In this manner the device serves as an analog memory, or it can be used to store one resistance for “1,” and another for “0.”
Plans for Deployment
Indeed, the memristor is touted as a replacement for both dynamic-random-access memory (DRAM) and for NAND-based flash memory, which form the basis of SSDs (solid-state drives). The weakness of RAM is that once the power is turned off, the memory (1s or 0s) is lost. The DRAM memory must be refreshed from hard drives or SSDs at the start of operation. DRAM is fast, but SSDs and hard drives are, in a computer’s time scale, quite slow, so time is lost. Memristors are faster, though not yet as fast as DRAM, and it is envisioned that the eventual commercial devices will have greater packing densities than DRAM.
Because of this fudging the line between volatile and nonvolatile memory, designers of computer systems are expected to rethink the entire question of memory in relation to other elements of the system. Some say that there will no longer be a need for a central hard drive or SSD, but that executable programs, data and even the various portions of the operating system itself will be stored at disparate locations within the computing system, closest to those other portions with which they communicate most intensively.
Of course, the ability to store analog data and the distribution of memory suggest a living brain to scientists who hope to someday develop a neuromorphic computer – otherwise known as a computer that “thinks” like a living brain. These advocates are heartened by the development at Northwestern University of a “three-ended” memristor. The resistance of this device can be affected by more than one source. As such, it is said that this memristor mimics the action of neurons, which can be fired from more than one other neuron.
Others have more modest ambitions for the near future. Moore’s Law has essentially been set aside by the realities of quantum mechanics. It simply will not be possible to store a 1 or a 0 in much less space then the 10 nanometers or so that it now takes. Some limited improvements are not only possible, but expected. But each battle from now on will be tougher than the last, with fewer rewards. These planners see a future in which memristors only gradually replace NAND-based memory, with basic architecture changes coming later.
Hewlett Packard Goes All-In
Hewlett Packard is developing a major new computer around the memristor. Aptly named, at least for the time being, “The Machine,” it will also substitute fiber optics for much of the copper cabling that connects the parts of this groundbreaking computer to further speed up data transfer. For now, no mimicking brains, but plenty of brain power will be devoted to its development; an astonishing 75 percent of the personnel of the famed HP Labs will devote themselves to the task.
HP is developing its own open-source language for “The Machine,” and also one based on UNIX. Both will, along with the system’s architecture, take advantage of the fact that data will always be available immediately to subsystems because there will be no waiting for SSDs, let alone hard drives. It will also take advantage of the shorter, simpler read-and-write cycles that memristors enjoy, instead of the time-consuming, complex steps that older memory devices must go through to read data in or out. And yes, although the first implementation of this project will most likely be a server, there are also plans for an Android version because HP expects eventually to deploy memristors to mobile devices as well.
According to HP, one of the problems with present-day computer architecture is the constant shuffling of information to and from memory that stores data need right now, shortly, or at some indefinite time in the future. “The Machine” will eliminate that distinction, so there will be no need for separate hard drives or SSD, static RAM and dynamic RAM. Everything that “The Machine” will need to remember will be held in only one type of memory, and that will be memristors.
All told, “The Machine” represents a radical departure from the computing architectures that we are used to. Its success will have far-reaching effects on the future of computing, and it remains to be seen if others will follow HP’s lead.