WO2010106305A1

WO2010106305A1 - Phase change devices and methods for their manufacture

Info

Publication number: WO2010106305A1
Application number: PCT/GB2010/000437
Authority: WO
Inventors: Stephen Richard Elliott; Jozsef Hegedus
Original assignee: Cambridge Enterprise Limited
Priority date: 2009-03-16
Filing date: 2010-03-11
Publication date: 2010-09-23
Also published as: GB0905359D0

Abstract

A phase change memory element can be temperature-controlled in order to change the crystal structure of at least a part of the memory element from a first structure (amorphous) to a second structure (crystalline), these structures having different electrical resistivity. A templating interface is provided in contact with the phase change memory element to promote, on phase change of the element, the crystallization of at least a part of the phase change element into the second structure. The second structure is different to the preferred crystal structure of a bulk material of the same composition under the same temperature control as the phase change element. The phase change element comprises Sb with one or more of Ge, As, Te, Sr, S, Sn, In, Ga, Pb, Se, e.g. Sb₈₅Ge₁₅. The material of the templating interface can be selected from Sr-Se materials, Sr-S materials, SrSe, and SrS.

Description

PHASE CHANGE DEVICES AND METHODS FOR THEIR MANUFACTURE

BACKGROUND TO THE INVENTION

Field of the invention

The present invention relates to phase change devices and to methods for the manufacture of phase change devices . Such devices have particular, but not exclusive, application in the area of information storage, such as memory cells.

Related art

The content of various published references is briefly discussed below. The full contents of each of these references are hereby incorporated by reference in their entirety.

The size scaling of Si-based integrated-circuit components

(CPU and memory), has followed the well-known Moore's law for the last few decades. However, this doubling of capacity every two years, due to a reduction in feature size, is predicted to come to an end in 2-3 years for the case of non- volatile (flash) memory: the thickness of the oxide insulator layer in such field-effect transistor devices will be insufficient to prevent electrons trapped at the gate from tunnelling away, thereby causing volatility. There is a need, therefore, for a new scalable, non-volatile memory technology to replace flash.

Phase-change (PC) materials are of interest for non-volatile electronic-memory technology to replace flash memory. Suitable phase change materials have amorphous and crystalline states with very different values of electrical resistivity, thereby enabling the encoding of bits of information. Furthermore, suitable phase change materials allow fast, reversible transformations between the amorphous and crystalline states by suitable control of the temperature and heating/cooling rate of the material. Typically, heat is applied to the material to control the temperature by passing electrical current through the material and relying on joule heating. Thus, the ability of phase-change (PC) materials to be reversibly and rapidly transformed between high-resistance amorphous and low-resistance crystalline states by joule heating via imposed current pulses provides the potential for recording binary bits of information.

US-A-2005/265072 shows that direct joule heating may not be essential for phase change memory elements. Instead, a separate heater may be provided. Similarly, US-A-2006/279978 discloses an external heater for controlling the phase of a layer of GST phase change material. The reason for the external heater is to speed up the write/erase operation - heating the phase change material by direct joule heating is said to have speed disadvantages, requiring a time of about 100 nanoseconds.

It is considered that PC memory may provide the following advantages: sub-25nm size scalability, 10ns read/write speeds, 10⁸ cyclability, multilevel (e.g. 4-bit) memory-cell programming capability, low-power operation and radiation hardness .

At present, it is not clearly understood why the crystallization rate of some suitable materials is so fast. Additionally, the origin of the resistivity contrast between amorphous and crystalline states is not clearly understood.

Most PC memory materials investigated to date have been Ge-Sb- Te (GST) compounds, of which Ge₂Sb2Te₅ (GST 225) is probably the most studied. A useful discussion of suitable materials is set out in a review article by Raoux et al 2008 [S. Raoux, G. W. Burr, M. J. Breitwisch, C. T. Rettner, Y. -C. Chen, R. M. Shelby, M. Salinga, D. Krebs, S. -H. Chen, H. -L. Lung, C. H. Lam, "Phase-change random access memory: A scalable technology" IBM J. RES. & DEV. VOL. 52 NO. 4/5 JULY/SEPTEMBER 2008, pp.465-479] .

US-A-2006/279978 further discloses the decreasing of read/write times by ensuring that a phase boundary is present in the GST material when there is amorphous phase present. Then crystallization can propagate not by nucleation but by movement of the phase boundary. Modelling in US-A-2006/279978 suggests switching times of 2 ns. The thickness of the phase change material is 60 nm is total, but the thickness of the amorphous region, when formed, is less than this, at about 2- 10 nm. This small thickness limits the difference between resistance of the two states, and so limits the number of bits that can be reliably stored by the cell.

US-A-2005/227177 discusses suitable materials for contact with a phase change material element in a memory cell. This document discloses that a suitable material for contact with GST is a Bi-Sr-Te solid solution. US-A-2007/264504 discloses methods for the formation of memory cell layers. Suitable substrates are metal foils.

US-A-2006/226409 discloses the formation of a phase change element (of GST) with a narrow neck. This means that only a channel in the phase change material is switched, leading to increases read/write speeds. US-A-2009/014704 discloses forming a current constricting layer in a GST phase change memory element using nanoparticles. This reduces the current required to heat the material to the required temperatures.

Although GST materials have been much studied for PCRAM applications, nevertheless there are problems with these materials, notably the desire to keep Te out of Si-CMOS fabrication lines used to make the supporting electronics or CPU circuitry. Thus, some workers have been investigating extensively the possibility of using Ge-Sb (GS) alloys for PCRAM applications. See, for example, Chen et al [Y. C. Chen, C. T. Rettner, S. Raoux, G. W. Burr, S. H. Chen, R. M. Shelby, M. Salinga, W. P. Risk, T. D. Happ, G. M. McClelland, M. Breitwisch, A. Schrott, J. B. Philipp, M. H. Lee, R. Cheek, T. Nirschl, M. Lamorey, C. F. Chen, E. Joseph, S. Zaidi, B. Yee, H. L. Lung, R. Bergmann, and C. Lam, "Ultra-Thin Phase-Change Bridge Memory Device Using GeSb" International Electron Devices Meeting 2006, IEDM'06, vols. 1,2, 531-534 (IEEE, NY: 2006) ] , in which GeSb is investigated as a material for a bridge-type memory cell, and Cabral et al [C. Cabral, Jr., L. Krusin-Elbaum, J. Bruley, S. Raoux, V. Deline, A. Madan, and T. Pinto "Direct evidence for abrupt postcrystallization germanium precipitation in thin phase-change films of Sb-15 at.% Ge" APPLIED PHYSICS LETTERS 93, 071906 2008].

US-B-7,491, 573 discloses the use of a solid solution of Sb with 1-12 at% Ge. A typical time of crystallization is 10 nanoseconds. US-B-7, 491, 573 suggests that Sb alone would in theory provide a fast phase change. However, this document explains that for practical reasons a doping element such as Ge is required. SUMMARY OF THE INVENTION

Ge₂Sb₂Te₅ (GST-225) crystallizes into a metastable rocksalt structure (containing vacancies on the (Ge, Sb) sub-lattice) under the switching conditions pertaining to phase change random access memory (PCRAM) operation [T. Matsunaga and N. Yamada "Structural investigation of GeSb₂Te₄ : A high-speed phase-change material" PHYSICAL REVIEW B 69, 104111 2004; T. Matsunaga, N. Yamada and Y. Kubotac "Structures of stable and metastable Ge2Sb2Te5, an intermetallic compound in GeTe-Sb₂Te₃ pseudo-binary systems" Acta Cryst. (2004) . B60, 685-691] . It is considered possible that the structural simplicity of this crystal phase is the reason for the very rapid (nanosecond- scale) crystallization rate observed in these materials. Moreover, the (defective) octahedral local coordination of the metastable rocksalt structure has been associated with the resonant bonding (presumed to be absent in the amorphous phase) that is considered to be likely to be responsible for the low-resistance (metallic) state of the crystal [K. Shportko, S. Kremers. M Woda, D. Lencer, J. Robertson and M. Wuttig "Resonant bonding in crystalline phase-change materials", Nature Materials Vol. 7 August 2008] .

The present inventors have realized that it would be advantageous to provide phase change materials using compositions different to GST. However, suitable alternative materials, whilst providing phase changes of interest, tend to behave differently to GST materials. In particular, some suitable materials tend to have heterogeneous crystal nucleation (i.e. crystallization is growth-limited) [L. van Pieterson, M. H. R. Lankhorst, M. van Schijndel, A. E. T. Kuiper, and J. H. J. Roosen "Phase-change recording materials with a growth-dominated crystallization mechanism: A materials overview" JOURNAL OF APPLIED PHYSICS 97, 083520 2005], whereas nucleation in GST is considered to be homogeneous [J. H. Coombs, A. P. J. M. Jongenelis, W. van Es-Sρiekman, and B. A. J. Jacobs "Laser-induced crystallization phenomena in GeTe- based alloys. II. Composition dependence of nucleation and growth" J. Appl . Phys . 78 (8), 15 October 1995] . Moreover, some suitable materials crystallize to a structure that is not the metastable rocksalt phase characteristic of GST. The GST crystal structure is considered to be particularly advantageous due to its very low electrical resistivity. Still further, some suitable materials tend to show phase separation upon repeated phase change cycling. These points apply in particular to Ge-Sb materials, but it is to be noted that the present invention is not necessarily limited to the use of such materials, although they are preferred in some embodiments. The Ge-Sb crystallization product, formed even under PCRAM device switiching conditions, is based on the stable rhombohedral A7 Sb structure. Phase separation into Sb and Ge crystalline products is considered to be likely to occur particularly in eutectic Ge-Sb. Accordingly, the present invention has been made in order to address at least one of these problems. Preferably, the present invention reduces, overcomes, avoids, or ameliorates at least one of these problems. It should be noted that the present invention need not necessarily solve all of these problems, in certain preferred embodiments, this may be achieved.

In a general aspect, the present invention provides a templating interface in contact with a phase change memory material to promote, on phase change of the material, the crystallization of the phase change material into a crystal structure different to the preferred crystal structure of the bulk phase change material.

In a first preferred aspect, the present invention provides a phase change memory element, adapted to be temperature- controlled in order to change the crystal structure of at least a part of the memory element from a first, optionally substantially amorphous, structure, to a crystalline second structure, a templating interface being provided in contact with the phase change memory element to promote, on phase change of the element, the crystallization of at least a part of the phase change element into the second structure, the second structure being different to the preferred crystal structure of a bulk material of the same composition under the same temperature control as the phase change element. In a second preferred aspect, the present invention provides a method of operating a phase change memory element, including controlling the temperature of the phase change memory element to change the crystal structure of at least a part of the memory element from a first, optionally substantially amorphous, structure, to a crystalline second structure, wherein a templating interface is provided in contact with the phase change memory element which promotes, on phase change of the element, the crystallization of at least a part of the phase change element into the second structure, the second structure being different to the preferred crystal structure of a bulk material of the same composition under the same temperature control as the phase change element.

In a third preferred aspect, the present invention provides a phase change memory element including a material formed of Sb with one or more of Ge, As, Te, Sr, S, Sn, In, Ga, Pb, Se, the preferred bulk crystal structure of the material being other than cubic-based, the element being adapted to be temperature- controlled in order to change the crystal structure of at least a part of the memory element from a first, optionally substantially amorphous, structure, to a crystalline second structure, a templating interface being provided in contact with the phase change memory element to promote, on phase change of the element, the crystallization of at least a part of the phase change element into the second structure, wherein the second structure is a cubic-based crystal structure.

In a fourth preferred aspect, the present invention provides a method of operating a phase change memory element according to the third aspect, including controlling the temperature of the phase change memory element to change the crystal structure of at least a part of the memory element from the first structure to the second structure.

In a fifth preferred aspect, the present invention provides a phase change memory cell including a phase change memory element according to the first aspect or according to the third aspect, including resistance determining means for determining the resistance of the phase change memory element. The phase change memory cell may further include heating means for heating the phase change memory element.

In a sixth preferred aspect, the present invention provides a data storage module including an array of phase change memory cells according to the fifth aspect.

In a seventh preferred aspect, the present invention provides a composition of Sb_xGe_y with a cubic-based crystal structure, wherein (x + y) is equal to one or about one and wherein y is less than or equal to 0.5. Preferably, the crystal structure is a rocksalt crystal structure. Preferred and/or optional features of the invention will now be set out. These are applicable singly or in any combination with any of the aspects of the invention, unless the context demands otherwise. Any of the aspects of the invention may be combined with any other aspects of the invention.

Preferably, the melting temperature of the phase change memory element is less than 1000⁰C. The melting temperature may be 900⁰C or less, or 800⁰C or less. A relatively low melting temperature reduces the power required to heat the phase change memory element in order to form the first structure, e.g. by cooling from the melt. However, a very low melting point may cause unwanted data volatility for the memory element. Thus, the melting temperature of the phase change memory element may be more than 200⁰C, preferably more than 300⁰C or more than 400⁰C.

Preferably, the resistivity of the phase change memory element in the first structure is at least 10 times greater (more preferably at least 100 times greater) than the resistivity of the phase change memory element in the second structure. In less preferred embodiments, the resistivity of the phase change memory element in the first structure may be at least 10 times smaller (e.g. at least 100 times smaller) than the resistivity of the phase change memory element in the second structure. Preferably, the first structure is substantially amorphous. Preferably, the second structure is cubic-based. More preferably, the second structure is a rocksalt (NaCl) crystal structure. The second structure may be a metastable crystal structure.

Typically, the preferred, bulk crystal structure is a rhombohedral crystal structure.

Preferably, the phase change of the phase change element between the first and second structures is reversible and/or repeatable. In use, the phase change element may be intended for cycling between the first and second structures at least 10 times. More preferably, the phase change element may be intended for cycling between the first and second structure at least 10³ times. It would be advantageous , at least 10⁵ times, at least 10⁷ times, at least 10⁹ times, at least 10ⁿ times or at least 10¹³ times.

The phase change element may be adapted to store one bit of information, based on the difference in electrical resistivity (or other physical properties, such as optical properties (e.g. reflectivity) between the first and second structures. Preferably, phase change element is adapted to store more than one bit (e.g. two bits, three bits, four bits or more) of information. This capacity may be provided by control of the proportion of the total volume of the phase change element that is allowed to form the second structure, thereby providing identifiable, intermediate positions between the extremes of the phase change element being either all first structure or all second structure.

Preferably, the phase change element may be formed from a chalcogenide material. However, preferably the phase change element is not formed from a chalcogenide material. Furthermore, preferably the phase change element does not include Te, except possibly as a trace impurity.

Preferably, the phase change element comprises Sb with one or more of Ge, As, Te, Sr, S, Sn, In, Ga, Pb, Se. More preferably, the phase change element consists of Sb with one or more of Ge, As, Sr, S, Sn, In, Ga, Pb, Se and optionally including incidental and/or trace impurities. Ge-Sb materials (e.g. Sbo.85Geo.15) are particularly preferred, optionally with a dopant and optionally including incidental and/or trace impurities. It is preferred to use compositions with relatively low melting point, in order to reduce the power required in order to form the first structure in the phase memory element (e.g. by melting and quenching) . For specific composition spaces of interest, it is therefore preferred to use compositions which provide the lowest melting point, e.g. eutectic compositions.

Preferably, the phase change element is formed as a layer over a substrate. The thickness of the phase change element is preferably less than 100 μm, e.g. less than 10 μm, more preferably less than 1 μm and still more preferably 100 nm or less. The phase change element may have a thickness of 1 nm or more.

Preferably, the templating interface is provided by a surface of a templating material of different composition to the material of the phase change element. The templating material preferably has a melting point higher than that of the phase change element. This is preferred in order that the crystal structure of the templating material is substantially unaffected by operation (typically heating and cooling) of the phase change element to cycle between the first and second crystal structures. Preferably, the melting point of the templating material is at least 200⁰C higher (and preferably at least 500⁰C higher) than that of the phase change element.

The templating interface preferably presents to the phase change material a lattice structure substantially matching at least one crystal plane of the second structure of the phase change element. By "substantially matching" here it is intended that the lattice structure presented to the phase change material at the templating interface has a maximum difference in lattice parameter (e.g. lattice constant along at least one axis) of 10% (but preferably significantly less, e.g. 9%, 8%, 7%, 6%, 5% or less) compared with the at least one crystal plane of the second structure of the phase change element .

Preferably, the second structure of the phase change element is a cubic-based crystal structure. It is preferably not a rhombohedral-base crystal structure. Preferably, the templating interface presents to the phase change element a lattice structure which is compatible with at least one crystal plane of a cubic-based crystal structure. For example, where the at least one crystal plane of the cubic- based crystal structure is a {100} plane, the templating interface preferably presents to the phase change element a substantially square lattice structure with a lattice constant substantially matching the {100} lattice constant of the second structure. Such a templating interface may, for example, be provided by a cubic-based or tetragonal-based templating material.

As another example, where the at least one crystal plane of the cubic-based crystal structure is a {110} plane, the templating interface preferably presents to the phase change element a substantially rectangular (non-square) lattice structure with a lattice constant substantially matching a {110} lattice parameter of the second structure. Such a templating interface may, for example, be provided by a cubic- based or tetragonal-based templating material. The templating material may be provided as a layer in contact with the phase change element. For example, the templating material may be formed as a layer between a substrate and the phase change element layer. Additionally or alternatively, the phase change element may be provided as a layer formed over the phase change element layer. Further still, the templating material may be provided as one or more side walls for the phase change element. In these situations, the templating interface may be substantially flat. This is advantageous in that a wider choice of suitable templating material compositions becomes available, since more materials will satisfy the requirements of substantial lattice matching with the second structure along a single interface plane only. The templating material may be formed as an epitaxial layer on a substrate. For example, the substrate may be a Si substrate or a Si-based substrate. The substrate may be a single crystal substrate.

The templating interface may, however, have a non-flat shape. For example, it may be curved and/or irregular. In this case, a narrower range of suitable templating materials will be available that satisfy the requirements of substantial lattice matching with the second structure. For example, the templating material may be provided as one or more inclusions in the phase change element. In particular, the templating material may be provided as one or more nanoparticles in the phase change element. Providing a significant number of nanoparticles in the phase change element increases the available area of templating interface, and thus can be expected to increase the speed of crystallization to the second structure.

The templating material may, for example, be a Sr-based material, with a melting point of at least 1000⁰C, preferably at least 1500⁰C. Particularly preferred templating materials are Sr-Se (e.g. SrSe) and Sr-S (e.g. SrS) .

Preferably, the speed of crystallization of the phase change element to the second structure is 10 ns or less, more preferably 5 ns or less, still more preferably 4 ns or less, 3 ns or less, 2 ns or less, 1 ns or less or 0.1 ns or less. Typically, this time corresponds to a "write" or "reset" operation of the memory element.

Further preferred and/or optional features are set out below.

BRIEF DESCRIPTION OF THE DRAWINGS

Preferred embodiments of the present invention are described in detail below, with reference to the accompanying drawings. A brief description of the drawings is set out here:

Figs. l(a)-(c) . Simulation of the templated heterogeneous nucleation and growth of the A7 rhombohedral Sb crystal structure in 144-atom models of pure Sb (Figs. l(a) and (b) ) and the eutectic GS alloy, SbβsGeis (Fig. 1 (c) ) . In all cases, the left-hand panel is a projection of the model, onto the (x- z) plane, of the atomic coordinates of the initial (random) starting configuration - a liquid at high temperature

(T > 2500K) . At time t = 1 ps, the temperature of the system is lowered instantaneously to 600K, and the system is left to evolve isothermally. The middle panel shows the time evolution of the z-coordinate of the atomic positions (each represented as a dot for each time step) showing the appearance of crystal planes with time. The right-hand panel shows the final (crystalline) structure at the end of the simulation. The position of the epitaxy template is shown as a continuous line in the time-evolution plot. The template configurations are: Fig. 1 (a) Sb templated with one A7 layer; Fig. l(b) Sb templated with two A7 layers; Fig. l(c) Sbg₅Gei5 templated with 2 Sb A7 layers (Sb atoms are open circles, Ge atoms are filled circles) . After crystallization completed in Fig. l(a) and (b) , the characteristic lattice spacing of the A7 structure can be seen in the middle panel; the alternating lattice-plane distances correspond to the 3+3, shorter and longer bond distances in the rhombohedral (A7) Sb structure. The GS melt (c) did not crystallize completely on this time scale.

Figs. 2 (a) and 2 (b) . Simulation of the templated heterogeneous nucleation and growth of the metastable cubic (rocksalt) structure in 150-atom models of: Fig. 2 (a) SbβsGeis (Sb atoms are open circles, Ge atoms are filled circles) and Fig. 2 (b) pure Sb. In both cases, the three panels correspond to the same situations as in Figs. l(a)-(c), and a single cubic Sb templating layer has been used (positioned, in these cases, at the edge of the simulation box, at Z = 0 A) . In Fig. 2 (b) , two crystal grains have grown, rotated with respect to each other about the z-axis. In the first three picoseconds of the simulations, a high temperature liquid phase is equilibrated (> 2000 K) to randomize the atomic positions, then the temperature was lowered instantaneously to T=600 K at t=3 ps for crystallization.

Fig. 3. Simulation of the heterogeneous nucleation and growth of the metastable cubic (rocksalt) structure of SbβsGeis (Sb atoms are open circles, Ge atoms are filled circles) using a single epitaxy template of SrSe (at the edge of the simulation box, at Z = 0 A) . In this case, the starting configuration was the crystal phase (equilibrated at 5OK for 2 ps, from t = 0 ps to t = 2 ps) . This was then heated to 600K for the next 1 ps (from t = 2 ps to t = 3 ps) , during which time it remained stable. The crystal was then melted at 3600K for 1.5 ps (from t = 3 ps to t = 4.5 ps) prior to the temperature being reduced to 600K for the isothermal crystallization run, starting at t = 4.5 ps. The left panel shows the atomic configuration just after annealing at 3600 K (t = 4.5 ps) . Figs. 4A-F. Time evolution of the maximal intensity of the 3D Fourier transform (FT) of the atomic positions of crystallizing models of Sb and Sb_SsGe₁₅ : the intensity of the normalized FT is unity when all atoms in the model lie on nodes of a plane wave. Figs. 4A-F show results for templated crystallization, showing that crystallization to the metastable cubic form is considerably faster than to the stable A7 rhombohedral form.

Figs. 5A-D. Time evolution of the maximal intensity of the 3D Fourier transform (FT) of the atomic positions of crystallizing models of Sb and SbssGeis : the intensity of the normalized FT is unity when all atoms in the model lie on nodes of a plane wave. Figs. 5A-D are simulations showing that untemplated (u) runs did not crystallize on time scales considerably longer than the times taken for cubic-templated (t) models to crystallize. In addition, it can be seen that structural ordering in the templated crystallization process of Sb8sGei5 to the cubic phase occurs more rapidly in the plane parallel to the template layer than in the planes perpendicular to the template.

Figs. 6A and 6B. Electronic density of states for cubic (A) and rhombohedral (B) A7 GS.

Figs. 7A-7E. The total radial distribution functions for

125-atom models of a-Sb and a-GS . The partial RDFs for a-GS are also shown. The amorphous models were quenched using untemplated simulation boxes: melts were cooled to 600K and equilibriated at that temperature for several tens of picoseconds before quenching to 300K in 6ps (i.e. at a cooling rate of 50K/ps) .

Figs. 8A-D. Bond-angle distributions in the a-GeSb model

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS, FURTHER PREFERRED AND/OR OPTIONAL FEATURES OF THE INVENTION

In order to improve the PCRAM behaviour, e.g. of GS alloys (although other materials are also contemplated) , we have aimed to design new materials in silico, by directing the metastable crystallization product to a favourable form (e.g. the rocksalt structure) , not normally produced under existing experimental conditions, by means of an epitaxial templating procedure. This demonstration has been achieved by means of ab initio molecular-dynamics (AIMD) simulations using the VASP code (explained in more detail below) , in which the simulated crystallization of Sb and of the eutectic GS composition, Sb₈5Gei₅, has been controlled by means of different crystal templates placed in the simulation box.

Constant-volume molecular-dynamics simulations were carried out using VASP (Vienna ab initio simulation package) [G.

Kresse and J. Hafner, "Ab initio molecular dynamics for liquid metals", Phys. Rev. B47, 558-561 (1993)] . We used the projector augmented-wave method with the Perdew-Burke- Ernzerhof exchange-correlation functional [Kresse, G. & Joubert, D. "From ultrasoft pseudopotentials to the projector augmented-wave method" . Phys. Rev. B, 59, 1758-1775 (1999);

Perdew, J. P., Burke, K. & Ernzerhof, M. "Generalized Gradient Approximation Made Simple" Phys. Rev. Lett., 77, 3865-3868 (1996)]. The plane-wave energy cutoff was set at 130 eV. In cubic and rhombohedral crystallization simulations 2x2x2 and 2x2x1 k-point mesh was used respectively and a Gaussian smearing of 0.1 eV was applied. The outer s and p electrons were treated as valence electrons. The temperature was controlled by Nose-Hoover thermostat. The integration time step was 4 fs. The densities used for rhombohedral and cubic simulations were 0.0312 atoms per cubic angstrom, and 0.0336 atoms per cubic angstroms respectively, these values were determined by VASP relaxing the cell volume at T = 0 K. DOS calculations were carried out on two atom unit cells of Sb using the Tetrahedron method with Bloechl corrections.

In a simulation box with normal periodic boundary conditions (and no templates), simulated melts of Sb and the GS alloy were found not to crystallize at 600 K in 90 ps-long annealing runs. In simulations mimicking the growth-limited, heterogeneously-nucleated crystallization observed experimentally in GS alloys, template layers, having the A7 rhombohedral structure, were placed in the simulation box. Melts of pure Sb, with one or two A7 template layers, crystallized to the A7 structure in 30-50 ps (Figs. Ia and Ib) . A melt of eutectic GS, templated with 2 pure Sb A7 layers, also (partially) crystallized to the A7 structure on the same time-scale (Fig. Ic) . There is some evidence from the time-evolution plots (Figs. Ia and Ib) of layer formation (in the (x-y) plane parallel to the template layers) that another, possibly cubic, layer spacing very rapidly begins to become established before the model finally crystallizes into the A7 structure.

However, directed crystallization to a new metastable phase, not previously reported experimentally for GS, was achieved in a simulation in which a cubic template of Sb was placed in the simulation box. In this case, the GS melt crystallized to the rocksalt structure, in epitaxy with the template, with the Ge atoms apparently distributed randomly rather than being clustered (Fig.2a). A cubic-templated melt of pure Sb similarly crystallized to the metastable rocksalt phase (Fig. 2b) . In both cases, the time-scale of crystallization to the cubic phase was significantly shorter than to the equilibrium A7 phase, taking place in only about 10 ps for the eutectic GS alloy (Fig. 2a) . Under normal experimental conditions, the cubic phase of Sb is only stable at high pressure, although there have been reports that metastable thin films of Sb can grow in the cubic phase. Although this new rocksalt GS phase has very advantageous properties relative to the A7 phase, as will be discussed later, it would not be feasible experimentally to use this phase itself as a templating layer in real PCRAM devices: the template layer of GS would invariably be the stable A7 form. However, materials other than GS, but with the rocksalt structure, could in principle also act as epitaxial crystallization templates. We observed, for example, that the stable crystalline phases of SrS and SrSe have the rocksalt structure with almost the same lattice parameter (a = 6.024 A and 6.236 A respectively [R. Khenata, H. Baltache, M. Rlerat, M. Driz, M. Sahnoun, B. Bouhafs, B. Abbar "First-principle study of structural, electronic and elastic properties of SrS, SrSe and SrTe under pressure" Physica B 339 (2003) 208-215]) as twice that of the metastable rocksalt phase of pure Sb ( a = 3.07A [J.A. Graves, J. H. Perepezko "Undercooling and crystallization behaviour of antimony droplets" JOURNAL OF MATERIALS SCIENCE 21 (1986) 4215 4220], 3.16A [D. Akhtar, V. D. Vankar, T. C. Goel, K. L. Chopra "Stabilization and transformation kinetics of the metastable phases of liquid- quenched antimony" JOURNAL OF MATERIALS SCIENCE 14 (1979) 2422-2426] ) . These materials also have the advantage that they are extremely refractory, having melting points (>2000oC and I6OO0C, respectively) greatly in excess of the eutectic temperature (about 600 ⁰C [B.C. Giessen and C. Borromee- Gautier "Structure and Alloy Chemistry of Metastable GeSb"

JOURNAL OF SOLID STATE CHEMISTRY 4, 447-452 (1972)]) for Sb₈sGei5; therefore epitaxy templates of them should survive repeated phase change cycling of the GS material, even if via the GS melt.

It was found that, indeed, a template of SrSe in the simulation box also caused crystallization of the eutectic GS melt to the rocksalt phase in a comparable time to that when cubic GS is used as a template (Fig. 3) . However, in this case, the crystallized GS rocksalt product does not grow in exact epitaxy with the SrSe template but, instead, the atoms in the first crystallized GS layer are displaced by a small constant lateral amount from the Sr and Se atoms in the template (Fig.3) . Further simulation of the SrSe-GeSb crystal-liquid interface showed that, while GeSb is melting, the SrSe crystal template remains intact, even at 1800 K.

Estimates for the time scale for (templated) crystallization in these simulated models can be obtained more accurately by calculating the fraction of atoms that are closer to final crystalline lattice planes than 0.4 A; this quantity is plotted in Figs. 4A-F as a function of time. This analysis clearly shows that, all other things being equal (viz number of template layers, size of models, annealing temperature = 600K) , cubic GS [or Sb) crystallizes much faster (in about 10 ps) than the rhombohedral A7 structure. Moreover, there is a subtle difference in crystallization-ordering behaviour in layers parallel and perpendicular to the templating layers: Fourier analysis of the atomic positions shows that structural ordering in layers parallel to the template layer takes place more quickly than for layers perpendicular to the template for crystallization to the cubic phase (Figs. 5A-D) .

This new metastable, templated rocksalt structure for the GS alloy exhibits three significant improvements in PC behaviour with respect to the normal A7 phase. First, its crystallization speed from the melt is much faster. Simulated crystallization of GS in a supercell of width about 2θA (Figs. 2a and 3), templated on both sides, occurs in about 10 ps . Extrapolated to the case of crystallization of a 25 run-width actual PCRAM cell, (templated on both sides) , the rocksalt crystallization time is predicted to be 0.125 ns, very significantly faster than current switching times of the order of tens of nanoseconds. Such an ultrafast crystallization speed would permit the replacement of DRAM memory by PC-memory technology.

Another important operational characteristic of PCRAM materials is the contrast in electrical resistance between amorphous and crystal phases: a large contrast in electrical conductivity allows for multilevel-memory operation. Figs. 6A and 6B show the calculated electronic densities of states (EDOS) for Sb₈₅Ge_I5 in the three phases, rocksalt, A7 rhombohedral and amorphous. It can be seen that the rocksalt phase exhibits the largest EDOS at the Fermi level, g(E_F), implying that the cubic phase would exhibit an even larger electronic conductivity than the A7 phase currently being studied experimentally, since the electronic conductivity of metallic materials depends on g(E_F) . Such behaviour is commensurate with an expected increase in resonant bonding associated with the octahedral coordination characteristic of the rocksalt structure, compared with its reduction due to the Peierls distortion associated with the A7 structure and its assumed absence in the amorphous phase.

Another improvement in PCRAM behaviour of the templated rocksalt form of GS concerns the potential problem of phase separation, into rhombohedral Sb and diamond-cubic Ge phases, on repeated PC cycling. On the basis of the present AIMD simulations, we consider that PC transformations involving the rocksalt structure of GS should be more resistant to such deleterious phase separation. There is little evidence for the onset of Ge clustering (at least on the time-scale of these simulations) , and it is well-known that Ge atoms can be readily accommodated in octahedral sites in the rocksalt structure, as in GST compounds. Hence, it is likely that Ge in Sb-Ge alloys would be less liable to phase separate during templated rocksalt crystallization, compared with normal crystallization to the A7 structure.

Little is known of the structure and properties of the amorphous GS phase (a-GS phase) , which also plays a key role in PCRAM behaviour. Thus, the findings of this AIMD simulation for the rapid quenching to the glassy phase is of interest. The radial distribution functions (RDFs) for 125- atom models of a-Sb and a-GS are shown in Figs. 7A-7E. It can be seen that the RDFs are rather similar, but with one exception. The first peak in the case of a-GS is broader than for a-Sb as a result of the nearest-neighbour Sb-Ge bond length being appreciably shorter than the Sb-Sb bond length, as seen in the partial RDFs for a-GS. In addition, there are also Sb-Sb contributions on the high-r side of the first peak, at r of about 3.3A, in the region of the minimum between first and second coordination shells in the RDF. Such correlations are also evident, but less pronounced in the RDF of pure a-Sb (Fig. 7A) .

Sb atoms in these models are mainly 3-fold coordinated, in accord with the value expected from the 8-N rule, but in addition 4-fold coordinated sites are also present, the number of these increasing with an increase in the cut-off distance in the region of the minimum between first and second peaks in the RDF. Such 4-fold sites are thus associated with the correlations at about 3.3A on the high-r side of the first peak in the RDF.

Bond-angle distributions (BADs) for the a-Sb model are shown in Figs. 8A-D, plotted respectively for 3-fold, 4-fold and 5- fold coordinated sites. It can be seen that the dominant bond angle is θ about equal to 90°, indicative of local octahedral- like geometry, even for the case of 4-fold coordinated Sb atoms, although there is perhaps a hint of some tetrahedral local geometry evident from the shoulder in the BAD at θ about equal to 109°. There is also evidence for (defective) octahedral local coordination for 4-fold sites in the contribution to the BAD at very large angles: each 4-fold coordinated site with ideal octahedral geometry would have five angles of 90° and one of 180°. Distorted sites with some bond angles somewhat greater than 90° would correspondingly have a large bond angle somewhat less than 180°. The 5-fold Ge is mainly 90° plus 180° but the 4-fold Ge has significant 109° components (tetrahedral) , so the 4-fold Ge has tetrahedral and octahedral characteristics while the 5-fold Ge is mainly octahedral. We consider that Ge "likes" to be in octahedral positions if it is more than 4-fold coordinated, so this indicates the possible prevention of phase separation of Ge if GeSb is in cubic form.

It can be inferred, therefore, that the reason why crystallization to the metastable cubic form is so much faster than to the A7 form in Sb or GS (Figs. 4A-F) is because the octahedral local bonding geometry, characteristic of the simple cubic structure, is already present in the amorphous phase. In conclusion, we have demonstrated the first in silico design of a new, improved phase-change memory material, namely Sbβ5Gei5 having the metastable rocksalt structure. Ab initio molecular-dynamics simulations have shown that this material crystallizes to this new phase (having improved phase-change properties) when grown in contact with a suitable epitaxy template material, in this embodiment SrSe.

As the skilled person will appreciate, the present invention may be applied to other materials systems of interest to phase-change elements for non-volatile memory applications, and in particular the various materials set out above.

A memory element may be formed using the materials described above, including the templating layer. The memory element may be employed in any suitable memory cell architecture, for example as described in Raoux et al 2008 [S. Raoux, G. W. Burr, M. J. Breitwisch, C. T. Rettner, Y. -C. Chen, R. M. Shelby, M. Salinga, D. Krebs, S. -H. Chen, H. -L. Lung, C. H. Lam, "Phase-change random access memory: A scalable technology" IBM J. RES. & DEV. VOL. 52 NO. 4/5 JULY/SEPTEMBER 2008, pp.465-479], the content of which is hereby incorporated by reference in its entirety. Each memory cell may include electrodes in order to determine the resistivity of the memory element (i.e. in order to "read" the state of the memory element) . In some device architectures, the same electrodes may be used in order to heat the memory element, using current/voltage pulses and relying on direct joule heating. These heating steps correspond to "reset" (normally heating to above the melting point and rapidly quenching to form the amorphous phase) and "set" (normally heating to a temperature below the melting point and allowing slower cooling in order to form the second crystal structure (preferably cubic-based) ) operations. Alternatively, separate heating means may be provided.

Claims

1. A phase change memory element, adapted to be temperature- controlled in order to change the crystal structure of at least a part of the memory element from a first, optionally substantially amorphous, structure, to a crystalline second structure, wherein a templating interface is provided in contact with the phase change memory element to promote, on phase change of the element, the crystallization of at least a part of the phase change element into the second structure, the second structure being different to the preferred crystal structure of a bulk material of the same composition under the same temperature control as the phase change element.

2. A phase change memory element according to claim 1 wherein the melting point of the phase change memory element is less than 1000⁰C but more than 200⁰C.

3. A phase change memory element according to claim 1 or claim 2 wherein the resistivity of the phase change memory element in the first structure is at least 10 times greater than the resistivity of the phase change element in the second structure.

4. A phase change memory element according to any one of claims 1 to 3 wherein the first structure is substantially amorphous .

5. A phase change memory element according to any one of claims 1 to 4 wherein the resistivity of the phase change memory element in the second structure is at least 100 times smaller than the resistivity of the phase change memory element in the first structure.

6. A phase change memory element according to any one of claims 1 to 5 wherein the second structure is a cubic-based crystal structure.

7. A phase change memory element according to any one of claims 1 to 6 wherein the phase change of the phase change element between the first and second structures is reversible and/or repeatable.

8. A phase change memory element according to any one of claims 1 to 7 wherein the phase change element does not include Te, except optionally in trace amounts.

9. A phase change memory element according to any one of claims 1 to 8 wherein the phase change element comprises Sb with one or more of Ge, As, Te, Sr, S, Sn, In, Ga, Pb, Se.

10. A phase change memory element according to any one of claims 1 to 9, formed as a layer over a substrate, of thickness 100 nm or less.

11. A phase change memory element according to any one of claims 1 to 10 wherein the templating interface is provided by a surface of a templating material of different composition to the material of the phase change element.

12. A phase change memory element according to claim 11 wherein the templating material has a melting point at least 200⁰C higher than that of the phase change element.

13. A phase change memory element according to claim 11 or claim 12 wherein the templating interface presents to the phase change material a lattice structure substantially matching at least one crystal plane of the second structure of the phase change element.

14. A phase change memory element according to any one of claims 11 to 13 wherein the templating material is provided as a layer in contact with the phase change element.

15. A phase change memory element according to any one of claims 11 to 14 wherein the templating material is formed as an epitaxial layer on a substrate.

16. A phase change memory element according to any one of claims 11 to 15 wherein the templating material is a Sr-based material, with a melting point of at least 1000°C

17. A phase change memory element according to any one of claims 11 to 16 wherein the templating material is selected from Sr-Se materials, Sr-S materials, SrSe, and SrS.

18. A method of operating a phase change memory element, including controlling the temperature of the phase change memory element to change the crystal structure of at least a part of the memory element from a first, optionally substantially amorphous, structure, to a crystalline second structure, wherein a templating interface is provided in contact with the phase change memory element which promotes, on phase change of the element, the crystallization of at least a part of the phase change element into the second structure, the second structure being different to the preferred crystal structure of a bulk material of the same composition under the same temperature control as the phase change element.

19. A phase change memory element including a material formed of Sb with one or more of Ge, As, Te, Sr, S, Sn, In, Ga, Pb, Se, the preferred bulk crystal structure of the material being other than cubic-based, the element being adapted to be temperature-controlled in order to change the crystal structure of at least a part of the memory element from a first, optionally substantially amorphous, structure, to a crystalline second structure, a templating interface being provided in contact with the phase change memory element to promote, on phase change of the element, the crystallization of at least a part of the phase change element into the second structure, wherein the second structure is a cubic-based crystal structure.

20. A method of operating a phase change memory element according to claim 19, including controlling the temperature of the phase change memory element to change the crystal structure of at least a part of the memory element from the first structure to the second structure.

21. A phase change memory cell including a phase change memory element according to any one of claims 1 to 17 or claim 19 including resistance determining means for determining the resistance of the phase change memory element.

22. A phase change memory cell according to any one of claims 1 to 17, 19 and 21, further including heating means for heating the phase change memory element.

23. A data storage module including an array of phase change memory cells according to claim 21 or claim 22.