As capacitors, the ubiquitous components of electronic circuitry, get smaller, keeping them insulating is a challenge. But that's not necessarily bad news — some conductivity might be just the thing for data storage.
A general problem in the electronics industry is that the insulating materials used in the continually shrinking capacitors and transistors start to leak charge when they become too thin. This leads to large power consumption and, in the case of memory, to difficulties in storing and retrieving information. But on page 81 of this issue, Garcia et al.1 show that this generally undesirable leakage current can in fact be very useful. They find that the leakage current flowing through ultrathin (1–3 nanometres) ferroelectric films of barium titanate (BaTiO3) is strongly dependent on their electric polarization states — that is, on whether the net electric dipole of the material is in one or the other of the two possible orientations. The authors' result, which allows direct reading of the polarization state through a simple measurement of the material's electrical resistance, may be just what is needed to put ferroelectric random access memories (FeRAMs) — those based on storing information in the polarization states of ferroelectric materials — back on track in the race for faster and better memory.
The ability of ferroelectrics to retain a permanent dipole in the absence of an electric field, and the possibility of reversing its direction with a modest voltage, has been a driving force behind decades of intense research in ferroelectric memory, where the 'up' and 'down' polarization states are used to code the 'ones' and 'zeros' of binary information2. Offering the non-volatility — the ability to retain information even when power is switched off — of hard disks, combined with speeds at which data are read and written comparable to those of 'dynamic random access memories' (DRAMs), FeRAMs were touted as the potential replacement for the flash memories found in today's mobile phones and digital cameras.
But despite huge technological advances and the successful commercialization of FeRAMs by several leading electronics manufacturers, the dream of the ultimate memory is at present still beyond reach, and FeRAMs remain competitive only in a number of niche applications. Industrial forecasts for the role of FeRAMs in the memory market have become more mixed. Whereas Samsung has recently presented its new vision of a FeRAM as part of a fusion memory3, rather than as a stand-alone solution, and subsequently shelved its FeRAM programme altogether, other manufacturers remain optimistic. For example, Toshiba has just announced a new 128-megabit prototype with writing speeds of 1.6 gigabytes per second (ref. 4).
The obstacles encountered by FeRAMs in the memory race are as much financial as technical. One of the main disadvantages of current FeRAMs is that they are charge-sensing devices. The information is stored in the dipole orientation of the ferroelectric, the insulating layer that is sandwiched between two metallic electrodes to make a tiny capacitor. To determine this orientation, a voltage is applied that, depending on the dipole's original direction, either reverses it or leaves it unchanged. A reversal of the polarization is accompanied by a current pulse that can be detected and so allows the dipole's orientation to be determined. The magnitude of this current pulse depends on the charge stored on the capacitor plates, and therefore on the area of the capacitor. With lateral dimensions approaching 100 nm, the charge available for sensing during the read operation is reduced. A concomitant increase in parasitic conduction (leakage) currents associated with downscaling of the capacitors further complicates the memory readout. What's more, the read process is destructive, in that each bit must be rewritten after being read. Achieving non-destructive readout is a major quest, and NASA's Jet Propulsion Laboratory in the 1990s5, and more recently Tonouchi's group6, have investigated various optical routes.
In their experiment, Garcia et al.1 explore another promising non-destructive readout technique. They use the conducting tip of an atomic force microscope (AFM) to create and then image ferroelectric domains (tiny ferroelectric regions with polarization 'up' or 'down' that can constitute the memory bits) on ultrathin ferroelectric films of BaTiO3, a well-known ferroelectric from the perovskite family of compounds. They show that films as thin as 1 nm are still ferroelectric at room temperature. With ferroelectricity expected to disappear below a certain film thickness, the value of which is still a matter of debate among both theoreticians and experimentalists7, this by itself is a major result, setting a new lower-size limit for ferroelectricity in BaTiO3.
Next, the conducting AFM tip was used to apply a small voltage and measure the leakage current through the film while scanning the tip across the sample surface, thus mapping the resistance of the ferroelectric domains. The resistance map showed remarkable homogeneity within each domain, and most importantly, a dramatic change (up to a factor of 750) between oppositely polarized domains. Hence, by measuring this resistance (or the leakage current for a given voltage), the direction of the polarization could be easily determined without altering it, allowing the highly desirable non-destructive readout of the ferroelectric memory to be obtained by exploiting the leakage currents.
So what is the source of leakage currents? Various imperfections in the material's crystal structure, and even the boundaries between the different ferroelectric domains, may lead to localized conducting channels through an insulator, sometimes with rather interesting properties8,9. Alternatively, electrons can surmount the insulating barrier by jumping over it, if given enough thermal energy, or passing through it via the quantum-mechanical process of tunnelling. The key to understanding the results of Garcia et al. is to note that the height and shape of the barrier that the insulator presents to the flow of electrons might be modified by changing its polarization state10, hence altering the probabilities of electrons being thermally excited over the barrier or tunnelling through it.
Both thermally activated and tunnelling conduction can lead to the desired giant changes in resistance11,12, which allow the two polarization states to be distinguished with ease. But discriminating between the different conduction mechanisms is not trivial. Measurements of resistance as a function of temperature may be crucial, as tunnelling conduction generally has a weaker dependence on temperature, making it the preferred mechanism from the point of view of applications where temperature stability of the devices is essential. More detailed studies of the (nonlinear) dependence of leakage currents on applied voltage would be worthwhile to clarify the complexities at the interface with the AFM tip and the physical mechanism that underlies the resistance changes reported by Garcia and colleagues.
Although highly motivating for the FeRAM community, within the broader field of resistive memory, the discovery of Garcia et al. will encounter some serious rivalry with the various contenders in the non-volatile-memory category13. The most mature technology is phase-change memory, which exploits switching upon heating between conducting crystalline and highly resistive amorphous phases. In 2006, Samsung announced a prototype 512-Mbit phase-change memory featuring a cell size of 46.7 nm. Whether devices based on Garcia and colleagues' development can, with the help of current FeRAM or DRAM processing techniques, be made more compact and make their way into the memory market, is an open question. Although a multi-AFM-tip approach, such as that developed at IBM for the Millipede project (Fig. 1), is a possible path to a ferroelectric memory device and bears the closest resemblance to the current work, conventional capacitor arrays with standard row-and-column memory addressing may be the preferred choice.
The work of Garcia et al.1 may profit from recent advances in oxide deposition methods, which allow the growth of 'perfectly' ordered (epitaxial) strontium titanate (SrTiO3) films on silicon wafers14. These SrTiO3 layers can be used as templates for further growth of perovskite oxides that include not only BaTiO3 and other technologically important ferroelectrics, but a whole spectrum of materials displaying a huge diversity of functional properties — from superconductivity and colossal magnetoresistance (the material's ability to change its electrical resistance when placed in a magnetic field) to multiferroicity (the coexistence in a material of both electric and magnetic ordering, which offers innovative means for memory storage15). The quality of many perovskite thin films is now comparable to that of group III–V semiconductors, and thus the successful integration of epitaxial oxide films onto silicon opens the door to a new era in oxide electronics and possibly to the next generation of non-volatile memory devices.
Garcia, V. et al. Nature 460, 81–84 (2009).
Scott, J. F. & Araujo, C. A. Science 246, 1400–1405 (1989).
Kim, K. & Jung, D. J. Integr. Ferroelectrics 96, 100–111 (2008).
Thakoor, S. Appl. Phys. Lett. 60, 3319–3321 (1992).
Takahashi, K., Kida, N. & Tonouchi, M. Phys. Rev. Lett. 96, 117402 (2006).
Rabe, K. M., Ahn, C. H. & Triscone, J.-M. (eds) Physics of Ferroelectrics (Springer, 2007).
Szot, K., Speier, W., Bihlmayer, G. & Waser, R. Nature Mater. 5, 312–320 (2006).
Seidel, J. et al. Nature Mater. 8, 229–234 (2009).
Tsymbal, E. Y. & Kohlstedt, H. Science 313, 181–183 (2006).
Blom, P. W. M., Wolf, R. M., Cillessen, J. F. M. & Krijn, M. P. C. M. Phys. Rev. Lett. 73, 2107–2110 (1994).
Maksymovych, P. et al. Science 324, 1421–1425 (2009).
Meijer, G. I. Science 319, 1625–1626 (2008).
McKee, R. A., Walker, F. J. & Chisholm, M. F. Phys. Rev. Lett. 81, 3014–3017 (1998).
Gajek, M. et al. Nature Mater. 6, 296–302 (2007).