Communications of the ACM
News
Memristors: Pass or Fail?
An image of a circuit with 17 memristors captured by an atomic force microscope at Hewlett-Packard's Information and Quantum Systems Lab.
Credit: R. Stanley Williams / HP Labs
A fundamental electronic device, whose existence was postulated five decades ago but which proved hard to understand, let alone build, is ready to emerge from the lab, corporate and university researchers say. If so, the memristor (or memory resistor), as it is called, may arrive just in time to save the information storage industry from the transistor's collision with the scaling wall at the end of Moore's Law.
Hewlett-Packard announced last August that it would team with the South Korean computer memory maker Hynix Semiconductor to develop memristor-based memory chips, called resistive RAM (ReRAM), which they say will be on the market in about three years. The companies say their titanium-based chips could replace flash memorywhich has become nearly ubiquitous in mobile applicationsand would be 10 times faster and 10 times more energy efficient. Meanwhile, Rice University has joined with Austin, TX-based PrivaTran, a semiconductor design company specializing in custom integrated systems, to develop an all-silicon ReRAM chip that could be a substitute for flash memory. But a senior research official at Intel says it is far from certain that either effort will succeed.
A memristor is a tiny two-terminal electronic component that can be made from a variety of materialsincluding polymers, metal oxides, and conventional semiconductors like siliconwhose resistance varies with the voltage applied across it and with the length of time the voltage is applied. Its initial applications are likely to be as binary memory devices, but it could work in an analog fashion and could eventually become the basis for certain types of logic circuits. "That [logic ability] could change the standard paradigm of computing, by enabling computation to one day be performed in chips where data is stored, rather than on a specialized CPU," says Gilberto Medeiros Ribeiro, a senior scientist at HP Labs.
The memristor has several qualities that make it attractive for memory chips. First, it is nonvolatile, so that it remembers its state after electrical current is switched off. Second, it can be scaled to a single nanometer (nm) in size, researchers believe, whereas the one-bit flash memory cell is expected to reach its scaling limit at about 20 nm. And Leon Chua, a professor of electrical engineering and computer science at the University of California, Berkeley, says the memristor's size advantage isn't its sole advantage. "You can not only build them smaller, but use fewer of them," he says. "Ten memristors might do the same thing as 50 transistors, so it's a new ball game."
In 1971, Chua published a paper, "MemristorThe Missing Circuit Element," in IEEE Transactions on Circuit Theory, which outlined the mathematical underpinnings of memristors, which he called the fourth fundamental building block of electronics (along with resistors, capacitors, and inductors). The existence of memristance had been reported earlierin 1960 by Bernard Widrow at Stanford University, for examplebut it was not well understood.
Earlier researchers had erroneously interpreted memristance as a hysteresis relationship (one in which effect lags cause) between voltage and current, when in fact it is based on flux and charge, the time integrals of voltage and current, says Chua. He likens the pre-1971 view of memristance to Aristotle's belief that force is proportional to velocity and not, as Newton correctly demonstrated 2,000 years later, as proportional to the change in velocity, or acceleration.
In 2006, HP designed and built a titanium memristor that worked predictably and retained its state when powered off, based on the mathematical framework proposed by Chua. "For years people built [memristance] devices almost by accident. It's to the great credit of HP that they finally figured it out," he says. Figuring it out, according to HP's Stanley Williams, the chief architect of the company's memristor, meant "understanding the mathematical framework for memristors."
Almost 40 years seems a long time between the emergence of Chau's framework and the ability to reliably produce memristors, but enormous engineering hurdles had to be overcome. It required methods and tools, such as scanning tunneling microscopy, that could work at atomic scales. HP says it experimented with an enormous number of device types, many based on exotic materials and structures, but the results were often inconsistent and unexplainable. It was not until 2006 that HP developed equations that explained just what was occurring in its titanium memristors.
More Speed, Less Power
The breakthrough achieved by HP in 2006 could revolutionize memory technology, the company says. "Memristor memory chips promise to run at least 10 times faster and use 10 times less power than an equivalent flash memory chip," according to Williams, director of HP's Information and Quantum Systems Lab. "Experiments in our lab also suggest that memristor memory can be erased and written over many more times than flash memory. We believe we can create memristor ReRAM products that, at any price point, will have twice the capacity of flash memory."
The memristor could enable computation to be performed in chips where data is stored, rather than on a specialized CPU, says Gilberto Medeiros Ribeiro.
However, not everyone has been impressed by the recent announcements from HP and Rice. "The memristor is only one of several interesting [recent] flash technologies, and by no means the most interesting," says Justin Rattner, Intel's chief technology officer. "Any time someone hypes a particular memory technology before building a large memory chip, you should be suspicious, very suspicious. It's one thing to demonstrate a storage device in the lab, but it's an entirely different thing to demonstrate it can be built in high volume at low cost and with exceptional reliability."
Rattner acknowledges that flash memory, which is a $20 billion-plus market, is rapidly approaching its scaling limit. But rather than memristors, Intel is concentrating on nonvolatile phase-change memory, by which certain types of glass can be made to switch between two states by the application of heat. In the amorphous state, the atomic structure of the glass is highly disordered and has high resistivity. But when switched to its crystalline state, the glass has a regular atomic structure and low resistivity. "We built commercial-grade, phase-change memories of sufficient size to fully understand the pros and cons of the technology in a high-volume environment," Rattner says. Intel is looking at additional novel approaches to nonvolatile memories such as the spin torque transfer memory, which exploits magnetic spin states to electrically change the magnetic orientation of a material.
The memristor prototype chip built at Rice is a 1-kilobyte ReRAM with sub-5 nm switches, according to Jim Tour, a synthetic organic chemist at Rice and a leading memristor researcher. Of Rattner's concern about manufacturing the devices, he says, "All so far looks goodmaterials cost, fabrication needs, scalability and switching timesexcept the switch voltage is a bit higher than we'd like, but we have some ideas to reduce that."
Rice claims to have an edge over HP's silicon-and-titanium memristor chip with its all-silicon model. "There are lots of engineering barriers to be overcome before this really takes off," says Doug Natelson, a professor of physics and astronomy at Rice. "But the use of all silicon makes the manufacturing very understandable."
Memristance comes from reduction-oxidation chemistry, in which atoms or molecules gain or lose their affinity for oxygen atoms, and in which the physical structure of materials can change. The Rice memristor chip, a thin layer of silicon oxide sandwiched between two electrodes, is made to convert back and forth between silicon (a conductor) and silicon oxide (an insulator.) A sufficiently large voltage (up to 13 volts) applied across the silicon oxide converts some of it into pure silicon nanocrystals that conduct current through the layer. The switch, according to Natelson, shows robust nonvolatile properties, a high ratio of current "on" to current "off" (>105), fast switching (sub-100 ns), and good endurance (104 write-erase cycles).
The HP version is conceptually similar, but works by the alternating oxidation and reduction of titanium. Titanium dioxide (TiO2) is a semiconductor and is highly resistive in its pure state. However, oxygen-deficient TiO2, which has oxygen "vacancies" where an oxygen atom would normally appear, is highly conductive. By applying a bias voltage across a thin film of semiconductor with oxygen-deficient TiO2 on one side, the oxygen vacancies move into the pure TiO2 on the other side of the semiconductor, thus lowering the resistance. Running current in the other direction will move the oxygen vacancies back to the other side, increasing the resistance of the TiO2 gain.
In the short term, memristors are most likely to be used in storage devices, but eventually may be used in artificial neural networks.
In the short term, memristors are most likely to be used in storage devices, but eventually may be used in artificial neural networks, in applications such as pattern recognition or real-time analysis of the signals from sensor arrays, in a way that mimics the human brain. A memristor works like a biological synapse, with its conductance varying with experience, or with the current flowing through it over time. Similarly, the brain learns and configures itself by varying the strength of synaptic connections between neurons. The ability of memristors to remember and to work as analog devices allows them to assume any of many values over a range, just as synapses do.
The memristor self-learns from experience, and the brain is made of memristors," Chua says. "That in the long run is much more interesting and important [than data storage], because it's how you can design intelligent machines. But it's in the next 50 years, not the next 10."
Further Reading
Chua, L.
Memristorthe missing circuit element, IEEE Transactions on Circuit Theory 18, 5, Sept. 1971.
Jo, S.H., Chang, T., Ebong, I., Bhadviya, B.B., Mazumder, P., and Lu, W.
Nanoscale memristor device as synapse in neuromorphic systems, Nano Letters 10, 4, March 1, 2010.
Strukov, D.B., Snider, G.S., Stewart, D.R., and Williams, R.S.
The missing memristor found, Nature 453, May 1, 2008.
Tour, J.M. and He, T.
Electronics: the fourth element, Nature 453, May 1, 2008.
Yao, J., Sun, Z., Zhong, L., Natelson, D., and Tour, J.M.
Resistive switches and memories from silicon oxide, Nano Letters 10, 10, Aug. 31, 2010.
Figures
Figure. An image of a circuit with 17 memristors captured by an atomic force microscope at Hewlett-Packard's Information and Quantum systems Lab.
©2011 ACM 0001-0782/11/0300 $10.00
Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and full citation on the first page. Copyright for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, to republish, to post on servers, or to redistribute to lists, requires prior specific permission and/or fee. Request permission to publish from [email protected] or fax (212) 869-0481.
The Digital Library is published by the Association for Computing Machinery. Copyright © 2011 ACM, Inc.
Comments
Adam Leventhal
The author claims that the scaling limit for flash is about 20 nm, but NAND vendor roadmaps include 1x nm technologies -- what's the source of the claim?
Displaying 1 comment