News & Views | Published:

Electronics

Inside story of ferroelectric memories

Nature volume 483, pages 279281 (15 March 2012) | Download Citation

The drive to improve digital memory through ever-shrinking electronic circuitry will ultimately face a bottleneck. Researchers propose exploiting the room 'inside' memory elements as a solution.

The amount of stored information in the world is reaching an astronomical1 5×1021 bits — roughly one million times the number of cereal grains produced on Earth in a year2. And the number of bits doubles every three years1. This has motivated efforts to produce improved digital memory solutions that have better reliability, lower power consumption, lower cost and, crucially, higher storage density. Writing in Advanced Materials, Lee et al.3 propose a method to increase the storage density of an existing class of commercial memory. The technique involves controlling the physical process through which information is written in memory cells — the building blocks of data-storage devices — instead of the usual approach of reducing the cells' size.

The data-storage market is dominated by hard-disk drives (HDDs) and Flash memories. HDDs provide the highest-capacity solutions, with an unbeatable cost per bit. But they are slow and prone to failure when shocked, causing the loss of documents, data or holiday pictures. Solid-state memories such as Flash, which are ubiquitous in memory sticks and memory cards for cameras, are faster and more robust than HDDs. However, they cannot match the cost per bit and maximum storage capacity of HDDs. But they are well suited to mobile electronic devices, and the consensus is that most future data-storage solutions will be based on solid-state memories.

The question is how to increase storage density and reduce the cost of solid-state memories while maintaining their good performance (fast data-access and write times, low power consumption and high endurance). Following the advice of physicist Richard Feynman, researchers have for decades exploited the “room at the bottom” — that is, they have tried to reduce the size of memory cells. Impressive progress in nanometre-scale lithography has allowed a marked reduction in the size of a memory cell. So today, the smallest commercial Flash memory cell4 has a lateral size of about 80 nanometres.

To further reduce the size of memory cells, problems related to charge leakage, which causes undesired data erasure, and general reliability issues must be solved. This is proving more challenging than ever as cells reduce in size. One solution is to increase storage density not by using fine-pitch lithography but by piling up two-dimensional arrays of memory cells and creating three-dimensional structures, thereby exploiting the 'room at the top'. The practical implementation is, however, extremely difficult in terms of circuitry, which prevents the pile-up of more than a few layers.

In their study, Lee et al.3 propose another method of data storage that exploits 'the room inside' the memory cell itself to store more than just one bit of information. Rather than Flash, the authors address ferroelectric random access memories (FERAMs), which, along with phase-change and magnetic memories, are an alternative type of non-volatile random-access-memory technology5. In a FERAM cell, a bit of information (a '1' or '0' logic state) is usually stored in the direction (up or down) of the polarization of a ferroelectric material — one that can retain a permanent electric dipole in the absence of an electric field (Fig. 1a). The polarization of a ferroelectric is the sum of the electric dipoles present in each unit cell of the material. Its direction can be switched by applying a voltage, just as the magnetization in a ferromagnet (a material that maintains a magnetic dipole in the absence of a magnetic field) can be reversed with a magnetic field.

Figure 1: Binary and multilevel data storage.
Figure 1

a, In a classical binary ferroelectric random access memory (FERAM), a bit of information, with a logic value of '1' or '0', is encoded in the up (red block) or down (blue block) polarization direction of the device's constituent cells. In this 36-bit example, the bits are stored in 6 × 6 cells, each of which stores one bit. Electrical wires (grey) are used to apply voltages across the cells to write and read the cells' polarization. b, Lee et al.3 describe a multilevel FERAM that can store three bits in a single cell. In this way, three-bit logic states (000, 001, 010, 011, 100, 101, 110 and 111) can be stored in each of 3 × 4 cells of a smaller 36-bit memory array. Each state corresponds to a specific configuration of domains of opposite polarization in the cells.

Rather than just relying on the direction of the polarization to store information in a FERAM cell, Lee and colleagues make the best of the physical mechanism at the heart of a memory operation by taking advantage of the structure of domains of different polarization orientation in the ferroelectric material. When a voltage is applied to reverse a ferroelectric's polarization, it usually does not reverse all of the material's constituent electric dipoles homogeneously. Instead, small regions (the domains), each with their own polarization, nucleate and then progressively expand. Depending on the voltage value, or the time during which it is applied, configurations — in which up and down domains coexist — can be stabilized. The net polarization is the sum of the up and down domains, and thus intermediate states of net polarization may be used to store information. These intermediate states define a multilevel memory element.

However, owing to the stochastic nature of ferroelectric polarization switching, controlling the up- and down-domain fractions by this voltage-based procedure creates poorly defined final intermediate polarization states. Lee et al.3 show that this poor definition can be greatly diminished by limiting the electrical current that appears in the ferroelectric when its polarization starts to be reversed by the applied voltage. Thus, complete polarization reversal is hampered, and the ferroelectric is left in a multidomain configuration. In this way, the authors were able to write eight well-defined logic states (000, 001, 010, 011, 100, 101, 110 and 111; Fig. 1b), converting a one-bit conventional FERAM memory element into a three-bit one. Although the main results were obtained on single-crystalline ferroelectric materials, they also show that the approach is applicable to polycrystalline films, which are typically used in commercial FERAMs.

The storage density of today's FERAMs is tenfold lower than that of Flash6, but exploiting ferroelectric domains could greatly improve it. Along with their excellent power consumption, endurance and write times, such improved storage density would put FERAMs back in the race towards a universal memory. In addition, the use of multidomain structures to encode intermediate polarization states could be applied to other devices based on ferroelectric layers, such as ferroelectric transistors7, tunnel junctions8 or switchable photovoltaics9. Lee and colleagues' work also emphasizes how a deep understanding of the physical processes underlying information storage allows even the highest technological hurdles to be overcome.

References

  1. 1.

    & Science 332, 60–65 (2011).

  2. 2.

  3. 3.

    et al. Adv. Mater. 24, 402–406 (2012).

  4. 4.

    & 3rd IEEE Int. Memory Workshop (2011).

  5. 5.

    Jpn. J. Appl. Phys. 49, 100001-1–100001-14 (2010).

  6. 6.

    et al. IEEE J. Solid-state Circuits 45, 142–152 (2010).

  7. 7.

    & Nano Lett. 11, 4003–4007 (2011).

  8. 8.

    et al. Nature Nanotechnol. 7, 101–104 (2011).

  9. 9.

    , , , & Science 324, 63–66 (2009).

Download references

Author information

Affiliations

  1. Vincent Garcia and Manuel Bibes are at Unité Mixte de Physique CNRS/Thales, Campus de l'Ecole Polytechnique, Palaiseau 91767, France.

    • Vincent Garcia
    •  & Manuel Bibes

Authors

  1. Search for Vincent Garcia in:

  2. Search for Manuel Bibes in:

Corresponding authors

Correspondence to Vincent Garcia or Manuel Bibes.

About this article

Publication history

Published

DOI

https://doi.org/10.1038/483279a

Further reading

Comments

By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.

Newsletter Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing