MAGNETIC TUNNEL JUNCTION WITH PERPENDICULAR SHAPE ANISOTROPY AND MINIMISED VARIATION OF TEMPERATURE MEMORY POINT AND LOGIC ELEMENT INCLUDING THE MAGNETIC TUNNEL JUNCTION, METHOD OF MANUFACTURING THE MAGNETIC TUNNEL JUNCTION

Abstract

A magnetic tunnel junction with out-of-plane magnetisation includes a storage layer; a reference layer; and a tunnel barrier layer. The two magnetisation states of the storage layer are separated by an energy barrier including a contribution due to the shape anisotropy of the storage layer and a contribution of interfacial origin for each interface of the storage layer. The storage layer has a thickness comprised between 0.8 and 8 times a characteristic dimension of a planar section of the tunnel junction. The contribution to the energy barrier due to the shape anisotropy of the storage layer is at least two times greater and preferably at least 4 times greater than the contributions to the energy barrier of interfacial origin.

Claims

1. A magnetic tunnel junction with out-of-plane magnetisation including: a storage layer having a magnetisation switchable between two magnetisation states perpendicular to a plane of the layer; a reference layer having a fixed magnetisation and perpendicular to the plane of the layer; a tunnel barrier layer separating the storage layer and the reference layer; the two magnetisation states of the storage layer being separated by an energy barrier including a contribution due to the shape anisotropy of the storage layer and a contribution of interfacial origin for each interface of the storage layer, wherein: the storage layer has a thickness comprised between 0.8 and 8 times a characteristic dimension of a planar section of the tunnel junction; and the contribution to the energy barrier due to the shape anisotropy of the storage layer is at least two times greater than the contributions to the energy barrier of interfacial origin; the storage layer comprising an interfacial layer made of an alloy rich in cobalt in contact with the tunnel barrier.

2. The magnetic tunnel junction according to claim 1, wherein the storage layer comprises a volume layer having a Curie temperature above 400 C.

3. The magnetic tunnel junction according to claim 1, comprising a single tunnel barrier layer.

4. The magnetic tunnel junction according to claim 1, wherein the contribution to the energy barrier due to the shape anisotropy of the storage layer is at least four times greater than the contributions to the energy barrier of interfacial origin.

5. The magnetic tunnel junction according to claim 1, wherein the storage layer includes one or more magnetic materials having a Curie temperature above 400 C.

6. The magnetic tunnel junction according to claim 1, wherein the storage layer includes an alloy including cobalt and/or iron and an amorphising element, said alloy being in contact with the tunnel barrier layer.

7. The magnetic tunnel junction according to claim 6, wherein the storage layer contains one or more layers of materials able to absorb the amorphising element present in the storage layer and to ensure structural transitions between the different magnetic materials comprised in the storage layer.

8. The magnetic tunnel junction according to claim 1, wherein its section is circular or quasi-circular and the characteristic dimension is the diameter of the section.

9. The magnetic tunnel junction according to claim 8, wherein the diameter is less than 50 nm.

10. The magnetic tunnel junction according to claim 1, wherein: the tunnel barrier layer is made of MgO, AlOx, AlN, SrTiO.sub.3, HfOx or any other insulating oxide or nitride; the storage layer includes: a layer made of alloy of cobalt, iron and amorphising element of thickness between 1 and 4 nm in contact with the tunnel barrier; a layer of a material able to absorb boron at the moment of post-deposition annealings, of 0.2 to 0.4 nm thickness; a magnetic layer with low Gilbert dampening.

11. The magnetic tunnel junction according to claim 1, wherein: said magnetic tunnel junction has a stability factor dependent on the energy barrier and the temperature of use of the magnetic tunnel junction; the composition and the thickness of the storage layer are chosen such that the absolute value of the derivative of the thermal stability factor compared to a characteristic dimension of a planar section of the tunnel junction is less than 10 nm.sup.1, the derivative of the stability factor being calculated at a temperature T.sub.m, the temperature T.sub.m being the average temperature of use of the magnetic tunnel junction.

12. A magnetic random access memory point of spin-transfer torque or STT-MRAM type including a magnetic tunnel junction according to claim 1.

13. A magnetic random access memory point of spin-orbit transfer or SOT-MRAM type including a magnetic tunnel junction according to claim 1.

14. A magnetic random access memory point with voltage controlled writing including a magnetic tunnel junction according to claim 1.

15. A non-volatile element of a logic component including a magnetic tunnel junction according to claim 1.

16. A method for manufacturing a magnetic tunnel junction according to claim 1 comprising: depositing all of the layers by physical vapour deposition; etching the storage layer or thick magnetic layer by reactive ion etching; etching the other layers by ion beam etching.

17. The magnetic tunnel junction according to claim 2, wherein the Curie temperature is above 800 C.

18. The magnetic tunnel junction according to claim 5, wherein the Curie temperature is above 800 C.

19. The magnetic tunnel junction according to claim 6, wherein the amorphising element is boron.

20. The magnetic tunnel junction according to claim 10, wherein the layer of a material able to absorb boron is a layer of Ta, Mo, W, or Hf.

Description

BRIEF DESCRIPTION OF THE FIGURES

[0102] Other characteristics and advantages of the invention will become clear from the description that is given thereof below, for indicative purposes and in no way limiting, with reference to the appended figures, among which:

[0103] FIG. 1A: Schematic representation of a STT-MRAM stack with single MgO barrier and a non-magnetic upper electrode according to the prior art;

[0104] FIG. 1B: Schematic representation of a standard STT-MRAM stack with double MgO barrier and a non-magnetic upper electrode according to the prior art;

[0105] FIG. 2: Temperature evolution of the thermal stability factor of p-STT-MRAMs of the prior art with out-of-plane magnetisation for tunnel junctions of diameter between 100 nm and 44 nm;

[0106] FIG. 3: Temperature dependency of the interfacial anisotropy of first order (uniaxial in sin.sup.2 , angle between the magnetisation of the storage layer and the normal to the plane of the layers) and of second order (in sin.sup.4 ) of W/CoFeB/MgO multilayers with different thicknesses of CoFeB. Taken from Kyoung-Min Lee et al., Temperature dependence of the interfacial magnetic anisotropy in W/CoFeB/MgO, AlP Advances 7, 065107 (2017);

[0107] FIG. 4: Thermal variation of the magnetisation of bulk cobalt;

[0108] FIG. 5: Thermal stability factor (grey levels) as a function of the diameter and the thickness of the storage layer;

[0109] FIGS. 6A-6C: Schematic representation of three examples of a magnetic tunnel junction PSA-STT-MRAM according to the invention;

[0110] FIG. 7: Schematic representation of the magnetic tunnel junction according to a first embodiment PSA-MRAM-1 of the invention;

[0111] FIG. 8: Schematic representation of the magnetic tunnel junction according to a second embodiment PSA-MRAM-2 of the invention;

[0112] FIG. 9: Energy barrier as a function of temperature between 40 and +260 C. for 4 different MRAMs: FeCoB (1.2 nm) with D=31 nm, FeCoB (1.4 nm)/Co (12 nm) with D=12.8 nm, Co (16.6 nm) with D=11.4 nm and Py (26.8 nm) with D=19 nm.

[0113] FIGS. 10A-10B: Implementation of the tunnel junction according to the invention in a cSOT-MRAM memory cell; a) Top configuration; b) Bottom configuration;

[0114] FIG. 11: Implementation of the junction according to the invention in a voltage controlled MRAM (VC-MRAM). The bottom configuration is also possible;

[0115] FIG. 12: Example of non-volatile logic component using tunnel junctions according to the invention. Here a non-volatile latch;

[0116] FIG. 13: Method for manufacturing a tunnel junction according to the invention;

[0117] FIG. 14A: Graphs of M.sub.S.sup.FM opti and t.sup.FM opti calculated for a magnetocrystalline anisotropy of K.sub.u.sup.FM=0; these curves correspond to the monolayer stack FM, for a stability .sub.300=60;

[0118] FIG. 14B: Identical to FIG. 14A with the sole exception that the magnetocrystalline anisotropy is equal to K.sub.u.sup.FM=0.3 10.sup.6 J/m.sup.3;

[0119] FIG. 14C: Graphs of M.sub.S.sup.FM opti (right vertical scale) and t.sup.FM opti (left vertical scale) calculated for a magnetocrystalline anisotropy of K.sub.u.sup.FM=0. These curves correspond to the FeCoB/FM stack with M.sub.S.sup.FeCoB=1.0 10.sup.6 A/m, t.sup.FeCoB=1.4 nm and K.sub.S.sup.FeCoB=1.4 mJ/m.sup.2, for a stability .sub.300=60;

[0120] FIG. 14D: Identical to FIG. 14C with the sole exception that the magnetocrystalline anisotropy is equal to K.sub.u.sup.FM=0.3 10.sup.6 J/m.sup.3;

[0121] FIG. 15: Representation of as a function of diameter for all M.sub.S (colour code) and for different thicknesses L (colour axes). Diagram calculated from the macrospin model for a cylinder with K.sub.S=0 J/m.sup.2 and K.sub.u=0 J/m.sup.3;

[0122] FIG. 16: Graphs representing in colour code the variability of the thermal stability factor (and thus also of the writing current) as a function of the diameter of the cell for two given variabilities of the diameter of the cell (variability associated with the etching method: D=1 nm left column, D=2 nm, right column). In these figures, 3 parameters have been varied , K.sub.s and K.sub.u. The vertical scales each represent one of these parameters, the two others being fixed. For each set of values of parameters , K.sub.s and K.sub.u, D, the thickness L and the magnetisation M.sub.S are chosen to correspond to the optimal parameters {L.sup.opti, M.sub.S.sup.opti}.

DETAILED DESCRIPTION OF THE INVENTION

[0123] Unless stated otherwise, a same element appearing in the different figures has a single reference.

[0124] FIGS. 6A-C schematically represents three examples of magnetic tunnel junction PSA-STT-MRAM according to the invention.

[0125] According to alternative a), the tunnel junction according to the invention includes a reference layer of fixed out-of-plane magnetisation RL, a tunnel barrier TB and a storage layer SL having a vertical thickness/(characteristic planer dimension) aspect ratio comprised between 0.8 and 8. The reference layer RL has a fixed perpendicular magnetisation represented by a single arrow. The storage layer has a perpendicular magnetisation switchable between two magnetisation states normal to the plane of the layers, represented by a double arrow. In the so-called top configuration a), the storage layer SL is above the tunnel barrier TB and the reference layer RL below. In the so-called bottom configuration b), the storage layer is below the tunnel barrier TB and the reference layer RL above. For reasons of quality of growth and ease of manufacture, configuration a) is preferable in so far as possible.

[0126] The configuration of FIG. 6 c) also represents a tunnel junction according to the invention in which the sides of the junction are slightly tapered instead of being virtually vertical. This shape with tapered sides may come from the etching of the junction by ion beach etching (IBE). Even if quantitatively the shape anisotropy of the storage layer SL is modified on account of this truncated cone shape, a vertical shape anisotropy may also be obtained and the other parameters may be adapted to this shape by those skilled in the art according to the present invention.

[0127] The ferromagnetic material constituting the storage layer SL and the shape of this layer (thickness and lateral dimension) are chosen such that the shape anisotropy of the storage layer SL is positive and above the interfacial anisotropy existing at the interface with the tunnel barrier. Two configurations may be used. In the so-called top configuration a), the storage layer is above the tunnel barrier and the reference layer below. In the so-called bottom configuration b), the storage layer is below the tunnel barrier and the reference layer above. For reasons of quality of growth and ease of manufacture, configuration a) is preferable in so far as possible.

[0128] FIG. 7 describes a first embodiment of a magnetic tunnel junction according to the invention or PSA-MRAM-1 in which the storage layer SL is composed of a single ferromagnetic layer FM, that is to say constituted of a single homogenous material.

[0129] The tunnel junction illustrated in FIG. 7 includes from bottom to top: [0130] a lower electrical contact layer T1, [0131] a metal growth layer M, [0132] a synthetic antiferromagnetic element SAF constituted of two ferromagnetic layers antiferromagnetically coupled through a coupling layer RKKY for example of Ru of 0.3 nm to 0.8 nm thickness, [0133] a structural transition layer TS able to absorb boron, for example made of Ta or W, [0134] a reference layer typically made of CoFeB of 1 to 3 nm thickness, [0135] a tunnel barrier usually made of MgO of between 0.6 and 2 nm thickness, [0136] the storage layer having a vertically elongated aspect ratio, the storage layer being formed of a homogenous material giving a strong tunnel magnetoresistance and preferably with a low Gilbert dampening such as an alloy based on Co, Fe and B. This layer FM is surmounted by a metal layer making it possible to absorb the amorphising element (here B) and also serving as upper electrical contact and optionally as hard mask during the etching of the memory point.

[0137] The bottom symmetrical configuration is also possible.

[0138] Starting from its base, the tunnel junction PSA-MRAM-1 includes on a substrate T1 a seed and growth layer M intended to favour the adherence of the tunnel junction on its substrate and to promote a good growth texture for the remainder of the junction. It next includes a synthetic antiferromagnetic layer SAF well known to those skilled in the art, constituted of two ferromagnetic layers of which the magnetisations are coupled in an antiparallel manner through a thin layer RKKY ensuring an antiparallel coupling between the magnetisations of the ferromagnetic layers. This coupling layer is constituted of a material such as ruthenium or iridium of suitably chosen thickness so that the coupling is antiparallel. For example, for ruthenium, the thicknesses giving antiparallel coupling are typically between 0.3 nm and 0.8 nm. The ferromagnetic layers are constituted of multilayers such as (Co/Pt) or (Co/Pd) or CoPt or CoPd alloys or L10 ordered alloys of FePt or FePd type known to have an out-of-plane magnetic anisotropy. This synthetic antiferromagnetic multilayer SAF is surmounted by a thin nanocrystalline metal layer TS typically made of Ta, Hf, W or Mo having several aims: 1) to ensure a structural transition between the synthetic antiferromagnetic multilayer generally of face centred cubic (fcc) structure of third order symmetry in the favoured growth direction (111) and the layer of CoFeB alloy in contact with the tunnel barrier which is of cubic centred structure thus of fourth order symmetry; 2) to absorb boron or the amorphising element if said element is other than boron (for example Zr, Nb, etc.) outside of the CoFeB layer in contact with the MgO tunnel barrier during the post-deposition annealing necessary for the good crystallisation of the MgO barrier and its electrodes. This layer is sufficiently thin (typically between 0.2 nm and 0.4 nm) to provide strong ferromagnetic coupling between the synthetic antiferromagnetic multilayer and the CoFeB layer.

[0139] This structural transition layer TS is itself surmounted by a layer of CoFeB alloy of thickness typically between 1 and 2 nm, which constitutes the reference electrode RL of the tunnel junction. The magnetisation of this layer remains fixed out-of-plane during the operation of the memory.

[0140] This layer is surmounted by the MgO tunnel barrier TB of typical thickness of 0.6 to 2 nm. This thickness makes it possible to control the resistance x area (RA) product of the junction. In STT-MRAMs, this RA product is generally chosen between 1 and 10 .Math.m.sup.2 so that the tunnel junction has a resistance comparable to that of the selection transistor in ON mode which is connected to the tunnel junction.

[0141] The tunnel barrier TB is surmounted by the magnetic storage electrode SL constituted in this first embodiment of a single layer of magnetic material made of alloy based on Co, Fe, Ni and of an amorphising element, preferably boron but other amorphising elements could be used such as Zr or Nb. The thickness of this layer will be chosen between 0.5 and 8 times the targeted diameter for the tunnel junction after nanostructuring. The alloy will be chosen with a Curie temperature above 400 C. and preferably above 600 C. or even 800 C. so that the variation of its magnetisation over the range 40 C. to 260 C. is not too important. One way of increasing the Curie temperature of the alloy is to increase its concentration of cobalt of which the Curie temperature is 1121 C. However, the Gilbert dampening depends on the concentration of the alloy, being minimum towards concentrations where iron is in proportion 3 times greater than cobalt (Ultra-low magnetic damping of a metallic ferromagnet, Martin A. W. Schoen Danny Thonig, Michael L. Schneider, T. J. Silva, Hans T. Nembach, Olle Eriksson, Olof Karis and Justin M. Shaw, Nat. Phys. 12, 839 (2016)). Since the Gilbert dampening has a direct influence on the writing current, it will be necessary to choose the concentration of the alloy both as a function of the need to minimise the thermal variation and the tolerance on the value of the writing current.

[0142] This ferromagnetic layer is surmounted by one or more metal layers T2. A first aim of this metal layer is to absorb the amorphising element outside of the magnetic storage electrode during the post-deposition annealing necessary for the good crystallisation of the tunnel barrier and the magnetic electrodes. This layer able to absorb the amorphising element may be made of tantalum (Ta), tungsten (W), molybdenum (Mo), hafnium (Hf), etc.

[0143] Optionally, several laminations of this layer able to absorb boron or the amorphising element may be introduced into the thickness of the ferromagnetic storage layer to improve the re-crystallisation of the alloy. For example a CoFeB layer of 16 nm surmounted by a Mo layer of 4 nm could be replaced by a (CoFeB 4 nm/Mo 2 nm).sub.4/Mo multilayer of 2 nm.

[0144] The ferromagnetic storage layer SL may also be surmounted by a second oxide layer (second tunnel barrier) intended to reinforce the perpendicular anisotropy of the storage layer. In this case, in addition to the shape anisotropy of the storage layer, this also benefits from the interface anisotropy not only at the interface with the tunnel barrier but also at the interface with this second oxide layer. It is then necessary to see to it that one or more layers able to absorb the amorphising element of the storage layer are inserted into the storage layer as described previously. As is known to those skilled in the art, the second tunnel barrier must have a sufficiently low resistance*area product compared to the first tunnel barrier that provides the tunnel magnetoresistance signal so as not to too reduce this signal due to the resistance in series associated with this second tunnel barrier. Other functions of the upper metal layer(s) T2 are to enable an electrical contact at the summit of the structure and optionally to serve as hard mask for etching during the etching of the tunnel junction.

[0145] The structure described in FIG. 7 is presented in top configuration. It goes without saying that the bottom configuration may also be produced by reversing the stack, that is to say with the storage layer under the tunnel barrier and the reference layer made of the SAF layer above the barrier.

[0146] FIG. 8 illustrates a magnetic tunnel junction PSA-MRAM-2 according to a second embodiment of the invention.

[0147] The magnetic junction tunnel PSA-MRAM-2 according to the invention includes from bottom to top: [0148] a lower electrical contact layer T1; [0149] a metal growth layer M; [0150] a synthetic antiferromagnetic SAF constituted of two ferromagnetic layers antiferromagnetically coupled through a coupling layer RKKY for example Ru of 0.3 nm to 0.8 nm thickness; [0151] a structural transition layer TS able to absorb boron for example made of Ta or W; [0152] a reference layer RL typically made of CoFeB; [0153] a tunnel barrier TB usually made of MgO; [0154] a storage layer SL having a vertically elongated aspect ratio, the storage layer being formed of an interfacial part I for example made of FeCoB, a thin transition layer TR, for example made of Ta or W or Mo, and a thick ferromagnetic layer FM; [0155] a metal layer T2 serving as upper electrical contact and optionally as hard mask during the etching of the memory point.

[0156] The bottom symmetrical configuration is also possible.

[0157] In the embodiment described in FIG. 8, the storage layer SL, instead of being constituted of a homogenous material, is constituted of an assembly of magnetic and non-magnetic layers. In particular, if the tunnel barrier is made of MgO as is most often the case, it is advantageous to conserve near to the interface with the tunnel barrier TB a layer I of CoFeB alloys preferably rich in cobalt to reduce the interfacial anisotropy, of 1.5 to 3 nm thickness, making it possible to obtain a strong tunnel magnetoresistance. As non-limiting example of alloys rich in Co may be cited:

[0158] Co.sub.80B.sub.20;

[0159] Co.sub.72Fe.sub.8B.sub.20;

[0160] Co.sub.56Fe.sub.24B.sub.20;

[0161] or more generally: (Co.sub.xFe.sub.(1-x)).sub.(1-y)B.sub.y with x comprised between 0.5 and 1 and y comprised between 0 and 0.40, boron being able to be substituted all or part by another amorphising element (for example Zr, Nb, V) and Fe being able to be partially substituted by nickel.

[0162] This thin magnetic layer may be associated with a thick layer of NiFe or alloys based on Co, Fe, Ni and optionally other additives (V, Al, Mo, etc.) having a low Gilbert dampening. This layer is noted FM. Advantageously, the two magnetic layers are separated by a non-magnetic layer TR for example of Mo, W, Hf or Ta of 0.2 to 0.4 nm thickness making it possible to absorb boron from the CoFeB layer during the post-deposition annealing necessary for the good crystallisation of the MgO barrier and the CoFeB electrodes, as is known in STT-MRAMs of the prior art.

[0163] In the second embodiment described in FIG. 8, the thick ferromagnetic storage layer SL includes at least two different magnetic materials. The first magnetic material in contact with the tunnel barrier makes it possible to obtain a strong tunnel magnetoresistance. Strong tunnel magnetoresistance is taken to mean a magnetoresistance at least above 100%. As in conventional p-STT-MRAMs, this interfacial layer I will be preferentially made of alloy based on Co, Fe and B or other amorphising element. Since in the present invention it is sought to reduce the relative role of the interface anisotropy, an interfacial alloy rich in cobalt will preferably be chosen here because it is known that the anisotropy at the Co/MgO interface is lower than at the Fe/MgO interface.

[0164] As non-limiting examples of alloys rich in Co may be cited:

[0165] Co.sub.80B.sub.20;

[0166] Co.sub.72Fe.sub.8B.sub.20;

[0167] Co.sub.56Fe.sub.24B.sub.20;

[0168] or more generally: (Co.sub.xFe.sub.(1-x)).sub.(1-y)B.sub.y with x comprised between 0.5 and 1 and y comprised between 0 and 0.40, boron being able to be substituted all or part by another amorphising element (for example Zr, Nb, V) and Fe being able to be partially substituted by nickel.

[0169] A second magnetic material FM is then deposited in the part of the storage layer further away from the tunnel barrier. This second material is chosen to have a low Gilbert dampening . Low Gilbert dampening is taken to mean a dampening less than 0.02. Indeed, the current necessary for writing by STT being proportional to the Gilbert dampening averaged over the whole volume of the layer, it is preferable to minimise this average dampening. Materials with low Gilbert dampening are Permalloy (Ni.sub.80Fe.sub.20), CoFe alloys of concentration close to Co.sub.25Fe.sub.75, Heusler alloys. These two magnetic materials being able to have different crystallographic structures, a structural transition layer TS able to absorb the amorphising element during post-deposition annealing could be introduced between the two layers of the two magnetic materials as represented in FIG. 8. Thanks to this choice of material at the interface with the tunnel barrier TB and in the volume of the storage layer SL, it will be sought that the shape anisotropy is at least 2 times greater and preferably 4 times greater than the interface anisotropy to reduce the temperature dependency of the total anisotropy.

[0170] The other constituent layers of the stack are similar to those discussed for the first embodiment of FIG. 7.

[0171] FIG. 9 illustrates the thermal variation of the thermal stability factor in tunnel junctions of the prior art and in those according to the present invention.

[0172] FIG. 9 shows the thermal variation of the barrier height E.sub.b which determines the value of the thermal stability factor =E.sub.b/k.sub.BT, for different types of storage layers according to the first embodiment (storage layer entirely made of Co, PSA-MRAM-1/Co, or entirely made of NiFe, PSA-MRAM-1/Py), for the second embodiment (Interfacial FeCoB layer associated with a thick volume layer made of cobalt, PSA-MRAM-2) and for a storage layer according to the prior art (EdA).

[0173] The benefit obtained from this invention in terms of thermal variation of the thermal stability factor is clearly shown in FIG. 9.

[0174] In the paragraph below, the thermal variations of thermal stability factors illustrated in FIG. 9 are compared on the basis of modelling for different compositions of storage layers in accordance with the first or second embodiments and for a storage layer in accordance with the prior art.

[0175] In this comparison, the variation of magnetisation as a function of temperature is described by:

[00020]$M_S\left(T\right)=M_S\left(0\right)\left[1-{\left(\frac{T}{T_c}\right)}^{\frac{3}{2}}\right]$

[0176] where M.sub.S(0) is the magnetisation at 0 K. The variation of anisotropy is given by:

[00021]$K\left(T\right)=K\left(0\right).Math.{\left(1-b.Math.\frac{T}{T_c}\right)\left[\frac{M_S\left(T\right)}{M_S\left(0\right)}\right]}^3$

[0177] where K(0) is the anisotropy at 0 K. The term b represents a thermal expansion coefficient, useful for taking account of the expansion of the crystalline lattice of the thin layers. Thus, b=0 will be taken for the thick layers (L>5 nm) and b=0.4 for the thin layers (L<5 nm). When the storage layer is composed of several ferromagnetic layers coupled together, a property X (X=M.sub.S, K.sub.S, K.sub.u, ) of the layer as a whole is given by:

[00022]$\{\begin{array}{c}L=\underset{i}{.Math.}.Math.L^i\\ X=\underset{i}{.Math.}.Math.X^i.Math.\frac{L^i}{L}\end{array}$

[0178] where the index i refers to the layer i. In addition, the following numerical values have been used: [0179] M.sub.S.sup.Co(0)=1.6 10.sup.6 A/m, magnetisation at 0 K of bulk Co. [0180] M.sub.S.sup.FeCoB(0)=1.6 10.sup.6 A/m, magnetisation at 0 K of the FeCoB thin layer. [0181] M.sub.S.sup.Py(0)=0.9 10.sup.6 A/m, magnetisation at 0 K of bulk Permalloy (Py). [0182] K.sub.S.sup.FeCoB(0)=3 mJ/m.sup.2, anisotropy at 0 K of the MgO/FeCoB thin layer. [0183] T.sub.c.sup.FeCoB=1100 K, Curie temperature of the FeCoB thin layer. [0184] T.sub.c.sup.Co=1388 K, Curie temperature of bulk Co. [0185] T.sub.c.sup.Py=826 K, Curie temperature of Permalloy.

[0186] The thermal variations of the barrier height have been studied between 40 C. and +260 C. to cover both the field of automobile applications and also the question of memory retention at the moment of soldering the chip (solder reflow compliance). In this comparison, the diameter and thickness values are chosen such that each device has a thermal stability =80 at 300 K.

[0187] The prior art EdA corresponds to the FeCoB stack (1.2 nm) with D=60 nm. It is observed that, in these conditions, the perpendicular anisotropy varies extremely rapidly and that in particular it is no longer maintained above 230 C. [0188] The two cases of the first embodiment PSA-MRAM-1/Co and PSA-MRAM-1/Py correspond respectively to the stacks Co(16.6 nm) with D=11.4 nm and Py(26.8 nm) with D=19 nm. In these cases, the anisotropy varies much more slowly compared to the prior art, in particular in the case of Co which has a very high Curie temperature (1121 C.). [0189] The case of the second embodiment PSA-MRAM-2 corresponds to the FeCoB(1.4 nm)/Co(12 nm) stack with D=12.8 nm. The dependency of its anisotropy is logically comprised between that of pure Co and that of the prior art since it involves a mix between the two technologies.

[0190] The PSA-MRAMs described in the present application thus have a dependency on their anisotropy as a function of temperature much lower than that of the prior art. In order to minimise this dependency, it is preferable to use the second embodiment and to choose for the volume part of the storage layer materials with high Curie temperature such as iron or cobalt or alloys containing a large proportion of these materials, in particular cobalt while seeing to it that the Gilbert dampening of the material chosen is not too great, that is to say remains typically less than 0.02. As regards the interfacial material of the storage layer in contact with the tunnel barrier, an interfacial material will be chosen having a low surface anisotropy since surface anisotropy varies more rapidly with temperature than shape anisotropy. This can be done with alloys based on Co and iron and amorphising element such as boron, rich in Co because the anisotropy at the oxide/cobalt interface is lower than at the oxide/Fe interface.

[0191] As non-limiting examples of alloys rich in Co may be cited:

[0192] Co.sub.80B.sub.20;

[0193] Co.sub.72Fe.sub.8B.sub.20;

[0194] Co.sub.56Fe.sub.24B.sub.20;

[0195] or more generally: (Co.sub.xFe.sub.(1-x)).sub.(1-y)B.sub.y with x comprised between 0.5 and 1 and y comprised between 0 and 0.40, boron being able to be substituted all or part by another amorphising element (for example Zr, Nb, V) and Fe being able to be partially substituted by nickel.

[0196] This interfacial alloy will be chosen to provide a strong amplitude of TMR as is the case with CoFeB alloys. As described previously, these two magnetic materials could be separated by a thin layer of material such as Ta, W or Mo, able to absorb the amorphising element and to ensure a structural transition between the two materials. Optionally, the storage layer may contain several laminations of such a material able to absorb the amorphising element.

[0197] In the case where the storage layer is inserted between two tunnel barriers, then the storage layer could be divided into 3 parts, a first interfacial magnetic layer in contact with the first barrier of same nature as described above in the case of a single barrier, surmounted by a second magnetic volume layer of same nature as described above in the case of a single tunnel barrier, surmounted by a second interfacial magnetic layer in contact with the second tunnel barrier, it also having low interface anisotropy. These different magnetic layers could be separated by several layers of material such as Ta, W or Mo, able to absorb the amorphising element and to ensure a structural transition from one material to the other. It should be noted that the use of two tunnel barriers here has a limited interest in so far as the second tunnel barrier used in the prior art serves to increase the perpendicular anisotropy of interfacial origin thanks to the presence of two oxide/magnetic metal interfaces whereas, in the present invention, it is sought on the contrary to reduce the contribution of the interfaces to the total anisotropy. In the present invention, it will thus be in general preferable to be limited to a single tunnel barrier.

[0198] In a third embodiment, the magnetic tunnel junction according to the invention is integrated in a SOT-MRAM cell.

[0199] The device may be produced in top configuration, illustrated in FIG. 10a or bottom configuration, illustrated in FIG. 10b.

[0200] The magnetic junction PSA-SOT according to the invention illustrated in FIG. 10a includes from bottom to top: [0201] a lower electrical contact layer or terminal T1, [0202] a metal growth layer M, [0203] a synthetic antiferromagnetic layer SAF constituted of two ferromagnetic layers antiferromagnetically coupled through a coupling layer RKKY for example of Ru of 0.3 nm to 0.8 nm thickness, [0204] a structural transition layer TS able to absorb boron for example made of Ta or W, [0205] a reference layer RL typically made of CoFeB, [0206] a tunnel barrier TB usually made of MgO, [0207] a storage layer SL having a vertically elongated aspect ratio, the storage layer being formed of an interfacial part I for example made of FeCoB, a thin transition layer TR, for example made of Ta or W or Mo, and a thick ferromagnetic layer FM.

[0208] A metal line LM parallel to the plane of the layers and in contact with the storage layer SL and comprising a second electrical contact or terminal T2 and a third electrical contact or terminal T3. As is known to those skilled in the art, SOT-MRAM cells have 3 terminals, T1, T2 and T3. During writing, the magnetisation of the storage layer SL is switched by making a current pulse circulate in a conductive line LM parallel to the plane of the layers. The material of the conductive line LM is chosen to have a high spin Hall angle. This line may be for example made of non-magnetic metal such as Pt, Ta, W, Ir or made of antiferromagnetic metal such as IrMn or PtMn or other heavy material with high spin orbit. A spin orbit effect called spin Hall effect taking place in this conductive line LM makes it possible to inject a spin polarised current into the storage layer SL. This spin current makes it possible to switch the magnetisation of the storage layer SL. As is known to those skilled in the art, when the magnetisation of the storage layer SL is out-of-plane, it is necessary to apply a static longitudinal field (substantially parallel to the conductive line) to make the switching deterministic. This field may be created by inserting at the base or at the summit of the stack a layer of hard material radiating a static field on the storage layer or by using the exchange anisotropy field produced at the interface between the storage layer and the conductive line of heavy metal when it is antiferromagnetic as described in the article: S. Fukami, C. Zhang, S. Dutta Gupta, A. Kurenkov and H. Ohno, Magnetization switching by spin-orbit torque in an antiferromagnet-ferromagnet bilayer system, Nat. Mat. 15, 535 (2017).

[0209] The device may be produced in top (FIG. 10a) or bottom (FIG. 10b) configuration.

[0210] In a fourth embodiment, the magnetic tunnel junction is used in a MRAM cell with voltage controlled writing. Such a tunnel junction with electric voltage controlled writing VC-MRAM is illustrated in FIG. 11.

[0211] The tunnel junction with electric voltage controlled writing VC-MRAM according to the invention includes from bottom to top: [0212] a lower electrical contact layer T1, [0213] a metal growth layer M, [0214] a synthetic antiferromagnetic layer SAF constituted of two ferromagnetic layers antiferromagnetically coupled through a coupling layer RKKY for example of Ru of 0.3 nm to 0.8 nm thickness, [0215] a structural transition layer TS able to absorb boron for example made of Ta or W, [0216] a reference layer RL typically made of CoFeB, [0217] a tunnel barrier TB usually made of MgO, [0218] a storage layer SL having a vertically elongated aspect ratio, the storage layer being formed of an interfacial part I for example made of FeCoB, a thin transition layer TR, for example made of Ta or W or Mo, and a thick ferromagnetic layer FM. [0219] a metal layer T2 serving as upper electrical contact and optionally as hard mask during the etching of the memory point. The symmetrical bottom configuration is also possible.

[0220] As is known to those skilled in the art, the magnetic properties of certain magnetic materials may be controlled by electrical field. In particular, the interfacial anisotropy at the oxide/magnetic metal interface may be modulated by the application of an electrical field through the oxide as described for example in: Induction of coherent magnetization switching in a few atomic layers of FeCo using voltage pulses, Y. Shiota, T. Nozaki, F. Bonell, S. Murakami, T. Shinjo, and Y. Suzuki, Nature Materials 11, 39 (2012). Also, the volume properties of multiferroic magnetic materials may be modulated by electrical field. Thus the storage layer may be conceived by playing on the modulation of the interface anisotropy, either by playing on the modulation of one volume magnetisation, or both, such that the application of a voltage to the terminals of the tunnel junction modifies the anisotropy from out-of-plane to in-plane initiating a rotation of the magnetisation from out-of-plane to in-plane. Given the processional movement induced by this initial rotation, by controlling the duration of the voltage pulse so that it is of the order of half the precession period, it is possible to switch the magnetisation from the upwards direction to the downwards direction or vice versa. The reading is carried out under lower voltage making it possible to measure the magnetic state of the junction by the tunnel magnetoresistance, the voltage being sufficiently low so as not to perturb the magnetic state of the junction. Such a tunnel junction with electric voltage controlled writing is illustrated in FIG. 11. Generally speaking, these junctions have a barrier thickness slightly greater than in junctions with spin transfer writing to minimise the current circulating through the barrier during writing.

[0221] In a fifth embodiment, illustrated in FIG. 12, the junction according to the invention is inserted as non-volatile element into a logic component of different possible natures. Different components and non-volatile logic circuits have been proposed over the last few years based on conventional tunnel junctions using very thin storage layers (1 to 3 nm thickness). These same components and circuits may be produced by substituting these tunnel junctions of the prior art by the tunnel junctions according to the invention. As an example, FIG. 12 shows a non-volatile latch including two magnetic tunnel junctions according to the invention, represented in blue. This circuit is adapted from the patent application US2015/0036415 A1.

[0222] FIG. 13 describes a method P for manufacturing the tunnel junction PSA-MRAM-1 or PSA-MRAM-2 according to the invention.

[0223] In particular, FIG. 13 illustrates: (a) deposition of all of the constituent layers of the stack, (b) reactive ion etching of the upper electrode, (c) ion etching of the thick layer FM belonging to the storage layer, (d) ion etching of the remainder of the tunnel junction and (e) reactive ion etching of the lower electrode; The method P according to the invention includes the following steps: [0224] Deposition D of all of the layers by physical vapour deposition; [0225] Etching RIE of the storage layer or thick magnetic layer by reactive ion etching; [0226] Etching IBE of the other layers by ion beam etching.

[0227] After deposition of all of the layers by a physical vapour deposition (PVD) method such as cathodic sputtering, a hard mask is defined by lithography and etching on the surface of the stack. The layer constituting the hard mask may for example be made of tantalum, which is easily etched by reactive ion etching.

[0228] Next, advantageously, it will be sought to etch the thick magnetic layer by reactive ion etching. Several articles describe means for etching Permalloy or cobalt by reactive ion etching using reactive gases based on methanol or carbon monoxide potentially combined with argon (Development of methanol based reactive ion etching process for nanoscale magnetic devices, M. T. Moneck and J. G. Zhu, NSTI-Nanotech 2011). Since this magnetic layer is thick, it is particularly advantageous to try etching by reactive ion etching to avoid a lot of deposition of this etched magnetic material. If the thick magnetic layer is constituted of several magnetic elements, it is then possible to etch by reactive ion etching at least the thickest of the two.

[0229] Next the remainder of the structure may be etched by ion beach etching (IBE) in so far as the layers in play are much thinner. This method is compatible with high memory point densities (technological node <10 nm) since the etching of the junction is almost entirely carried out by RIE and that the storage layer may itself be constituted of a thick hard mask for the etching of the lower part of the stack.

[0230] Minimisation of the Variability

[0231] The invention also relates to a magnetic tunnel junction having a stability factor dependent on the energy barrier and the temperature of use of the magnetic tunnel junction in which the composition and the thickness of the storage layer are chosen such that the absolute value of the derivative of the thermal stability factor compared to a characteristic dimension of a planar section of the tunnel junction is less than 10 nm.sup.1, the derivative of the stability factor being calculated at a temperature T.sub.m, the temperature T.sub.m being the average temperature of use of the magnetic tunnel junction.

[0232] Advantageously, such a choice of the parameters of the magnetic tunnel junction makes it possible to reduce variability from memory point to memory point of the anisotropy and of the writing current over a wide operating temperature range.

[0233] An example of temperature range is 40 C. to +150 C. for automobile applications. Energy barrier separating the two magnetisation states of the storage layer is taken to mean the difference in energy between the minimum energy corresponding to the initial stable state of the magnetisation of the storage layer and the maximum energy encountered during the reversal of the magnetisation to the final stable state.

[0234] Thermal stability factor is taken to mean a measurement of the stability of the magnetisation of the storage layer for a given energy barrier and temperature.

[0235] For example, the thermal stability factor may be expressed as =E.sub.b/K.sub.BT, E.sub.b being the energy barrier, T the temperature and K.sub.B Boltzmann's constant.

[0236] Planar section of the tunnel junction is taken to mean a section of the tunnel junction along the plane of the layers forming the tunnel junction.

[0237] Characteristic dimension of a planar section is taken to mean a dimension of said planar section. For example, in the case of tunnel junction having a circular section, the characteristic dimension of a planar section of the tunnel junction may be chosen as being the diameter of the circle. In the case of elliptical section, the characteristic dimension may be chosen as being the small axis or the large axis of the ellipse.

[0238] Hereafter characteristic dimension of a planar section and characteristic planar dimension are synonymous.

[0239] Choice of the composition of the storage layer is taken to mean the choice of at least one of the materials comprised in the storage layer.

[0240] Thickness of the storage layer is taken to mean the size of the storage layer measured along a direction normal to the plane of the layers.

[0241] An important point of this embodiment concerns the optimisation of the choice of the thickness and magnetic parameters of the storage layer to minimise the effects of dispersion of properties from memory point to memory point during the manufacture of a memory chip. Among the following parameters: diameter, thickness, magnetisation at saturation, interface anisotropy and volume anisotropy, mainly the diameter is capable of varying significantly from memory point to memory point on account of the method of etching or nanostructuring the memory points. Indeed, all the other parameters are chosen upstream of the etching of the junctions and are not affected by the latter. The deposition techniques of those skilled in the art are perfectly mastered and do not induce significant spatial variations of its properties within a chip. On the other hand, during the etching of the tunnel junctions, it is inevitable that variations of diameter appear from one junction to the next. These variations of diameter may typically be of the order of 1 to several nanometres. Since the total effective anisotropy mainly or entirely draws its source from the shape anisotropy term, that is to say linked to the very shape of the storage layer, it is inevitable that variations of diameter induce variations of thermal stability, and consequently variations of writing current. In order to minimise the effect of the diameter variations on the stability, it is necessary to be positioned at the point defined by min

[00023]$\left(.Math.\frac{}{D}.Math.\right)$

where min(*) represents the minimum function and |*| the absolute function value. In particular, when this is possible, it is necessary to be positioned at the point defined by

[00024]$\frac{.Math.}{D}=0.$

[0242] FIG. 15 shows as a function of D in the macrospin approximation, for a cylinder with K.sub.S=0 J/m.sup.2 and K.sub.u=0 J/m.sup.3. These curves clearly show that a maximum of (D) is always attainable, whatever the value of the thickness and the M.sub.S of the layer. Being placed around this maximum is advantageous because this shows that the thermal stability factor is then slightly dependent on the exact diameter of the pillar constituted by the magnetic tunnel junction. It is easy to be convinced of this by plotting these same curves for different values of K.sub.S and K.sub.u which, in the operating regime of PSA-MRAMs, the function (D) always has a single maximum. Thus, even in the presence of given interfacial volume anisotropy: K.sub.S and K.sub.u, there exists a single optimal solution {L.sup.opti, M.sub.S.sup.opti} that minimises the effects of a variability in diameter on the stability, for a predefined stability and technological node (i.e. a diameter).

[0243] The calculations below present an analytical resolution of the parameters {L.sup.opti, M.sub.S.sup.opti} in the limit of the approximation of the approximate expression of the demagnetising factors of a uniformly magnetised cylinder mentioned above. The numerical applications will be made for certain anisotropy values K.sub.S and K.sub.u typically achievable with materials of those skilled in the art. The invention is in no way limited to the case of the uniformly magnetised cylinder or to the anisotropy values chosen as an example. In order to simplify the expressions, constants a, b and c are introduced into the expression of the thermal stability factor.

[00025]$={\mathrm{aLD}}^2\left(1-\frac{3.Math.D}{D+\mathrm{bL}}\right)+{\mathrm{cD}}^2\left(K_u.Math.L+K_S\right).Math..Math.\mathrm{where}$$a=\frac{{}_0.Math.M_S^2}{16.Math..Math.k_B.Math.T}$$b=\frac{4}{\sqrt{}}$$c=\frac{}{4.Math.k_B.Math.T}$

[0244] The optimisation condition is written:

[00026]$\frac{}{D}=\frac{2.Math.}{D}-\frac{3.Math.{\mathrm{abL}}^2.Math.D^2}{{\left(D+\mathrm{bL}\right)}^2}=0.Math.a=\frac{2.Math.{\left(D+\mathrm{bL}\right)}^2}{3.Math.{\mathrm{bL}}^2.Math.D^3}$

[0245] This expression of a is injected into the expression of the thermal stability factor. After simplifications, it is possible to extract a quadratic equation in L.

[00027]$L^2\left[\frac{1}{2}+\frac{3.Math.{\mathrm{cD}}^3.Math.K_u}{4.Math.b.Math..Math.}\right]+L\left[-\frac{5.Math.D}{4.Math.b}+\frac{3.Math.{\mathrm{cD}}^3.Math.K_S}{4.Math.b.Math..Math.}\right]-\frac{D^2}{b^2}=0$

[0246] Among the two roots L.sub., only L.sub.+ is positive and thus correspond to the optimal thickness previously designated L.sup.opti.

[00028]$L^{\mathrm{opti}}=\frac{1}{1+\frac{3.Math.{\mathrm{cD}}^3.Math.K_u}{2.Math.b.Math..Math.}}[.Math.\frac{5.Math.D}{4.Math.b}-\frac{3.Math.{\mathrm{cD}}^3.Math.K_S}{4.Math.b.Math..Math.}+\sqrt{{\left(\frac{5.Math.D}{4.Math.b}-\frac{3.Math.{\mathrm{cD}}^3.Math.K_S}{4.Math.b.Math..Math.}\right)}^2+\frac{2.Math.D^2}{b^2}.Math.\left(1+\frac{3.Math.{\mathrm{cD}}^3.Math.K_u}{2.Math.b.Math..Math.}\right)}]$

[0247] M.sub.S.sup.opti is deduced therefrom:

[00029]$M_S^{\mathrm{opti}}=\sqrt{\frac{16.Math..Math.k_B.Math.T}{{}_0}.Math.\frac{2.Math..Math.{\left(D+{\mathrm{bL}}^{\mathrm{opti}}\right)}^2}{3.Math.{b\left(L^{\mathrm{opti}}\right)}^2.Math.D^3}}$

[0248] As an example, several examples of standard magnetic materials are listed having a magnetisation M.sub.S.sup.FM opti which minimises the variability of .sub.300 for junction diameter values between 10 nm and 25 nm: [0249] FM=cobalt. A layer of 15 nm of Co (M.sub.S=1.446 10.sup.6 A/m, K.sub.u=0 J/m.sup.3) corresponds to the optimal layer for a junction diameter of D=10.8 nm. [0250] FM=iron. A layer of 13.4 nm of Fe (M.sub.S=1.714 10.sup.6 A/m, K.sub.u=0 J/m.sup.3) corresponds to the optimal layer for a junction diameter of D=9.6 nm. [0251] FM=Permalloy (Py). A layer of 23.1 nm of Py (M.sub.S=0.756 10.sup.6 A/m, K.sub.u=0 J/m.sup.3) corresponds to the optimal layer for a junction diameter of D=16.6 nm. [0252] FM=CoFe alloys. These alloys are optimal for junction diameters comprised between 10 and 15 nm. [0253] FM=Co.sub.2FeAl (full-Heusler). A layer of around 6 nm of Co.sub.2FeAl (M.sub.S=1.3 10.sup.6 A/m, K.sub.u=0.3 10.sup.6 J/m.sup.3) corresponds to the optimal layer for diameters of around 25 nm.

[0254] Finally, in the case of a bilayer comprising an interface based on FeCoB of which the properties are assumed to be optimised for other reasons (TMR, RA, spin polarisation, etc.) as well as a volume layer noted FM, the optimal characteristics (optimal thickness L.sup.FM opti and optimal magnetisation M.sub.s.sup.FM opti) of the layer FM are given by:

[00030]$\{\begin{array}{c}{L}^{\mathrm{FM}.Math..Math.\mathrm{opti}}=L^{\mathrm{opti}}-L^{\mathrm{FeCoB}}\\ M_S^{\mathrm{FM}.Math..Math.\mathrm{opti}}=\frac{M_S^{\mathrm{opti}}.Math.L^{\mathrm{opti}}-M_S^{\mathrm{FeCoB}}.Math.L^{\mathrm{FeCoB}}}{L^{\mathrm{opti}}}\end{array}$

[0255] This point corresponds to the maxima of the iso-M.sub.S of FIG. 15 or instead to the minima iso- of FIG. 5.

[0256] FIGS. 14A-D represents L.sup.FM opti and M.sub.S.sup.FM opti as a function of the technological node (identified here at the diameter of the memory point) for a stability =60 at 300 K, for a stack having or not surface anisotropy. Two volume anisotropy values are presented therein.

[0257] The graphs illustrated in FIGS. 14a to 14d represent the graphs of M.sub.S.sup.FM opti and t.sup.FM opti calculated for different anisotropy values K.sub.s and K.sub.u as a function of the diameter of the tunnel junction.

[0258] In order to minimise the dispersion of thermal stability factor and writing current from one memory point to the next, a position will be taken in conditions such that

[00031]$.Math.\frac{}{D}.Math.<\mathrm{crit}$

[0259] Where crit is a positive dispersion tolerance constant. This tolerance is defined by the fact that the distributions of writing voltage, reading voltage and dielectric breakdown voltage have to be well separated as is known to those skilled in the art. Typically, it will be sought to have a dispersion of below 40 (that is to say for example =8020 for a dispersion of diameter of tunnel junctions of 2 nm) i.e.

[00032]$.Math.\frac{}{D}.Math.<10.Math..Math.{\mathrm{nm}}^{-1}.$

However to produce high capacity memories (several Gbits), it will be necessary to have a narrower writing voltage distribution thus a stricter criterion

[00033]$.Math.\frac{}{D}.Math.<3.Math..Math.{\mathrm{nm}}^{-1}$

or even

[00034]$.Math.\frac{}{D}.Math.<1.Math..Math.{\mathrm{nm}}^{-1}.$

This signifies working closer and closer to the optimal condition defined by L.sup.FM opti and Ms.sup.FM opti.

[0260] In the case of a storage layer constituted of a single homogenous material of uniform magnetisation, the condition

[00035]$.Math.\frac{}{D}.Math.<\mathrm{crit}$

is written

[00036]$.Math.\frac{2.Math.}{D}-\frac{3.Math.{\mathrm{abL}}^2.Math.D^2}{{\left(D+\mathrm{bL}\right)}^2}.Math.<\mathrm{crit}$

where the constants a and b have been defined previously.

[0261] FIG. 15 illustrates the zones of parameters (diameter (D), thickness (L)) in which a value

[00037]$.Math.\frac{}{D}.Math.$

is obtained below a certain criterion crit equal to 5, 10, or 20 nm.sup.1 in the figure. The figure is represented in multiscale form making it possible to cover different types of materials represented by their magnetisation in colour code. To establish this figure, the variations of stability are calculated, which implies a given variation of diameter D (that is to say

[00038]$.Math.\frac{}{D}.Math.),$

as a function of the parameters D, , L and M.sub.S and by considering K.sub.S=0 and K.sub.u=0. The value of

[00039]$.Math.\frac{}{D}.Math.$

is then plotted in hatching code as a function of (Y-axis) and of D (X-axis), for all M.sub.S (colour code) and for different values of L (coloured axes). Vertical hatchings indicate that

[00040]$.Math.\frac{}{D}.Math.<5.Math..Math.{\mathrm{nm}}^{-1},$

horizontal hatchings indicate that

[00041]$5<.Math.\frac{}{D}.Math.<10.Math..Math.{\mathrm{nm}}^{-1},$

cross hatchings indicate that

[00042]$10<.Math.\frac{}{D}.Math.<20.Math..Math.{\mathrm{nm}}^{-1}$

and everything that is beyond the hatchings indicates that

[00043]$.Math.\frac{}{D}.Math.>20.Math..Math.{\mathrm{nm}}^{-1}.$

For example, it is reasonable to consider that must not vary by more than 20 for variations of diameter of 2 nm, i.e.

[00044]$.Math.\frac{}{D}.Math.<10.Math..Math.{\mathrm{nm}}^{-1}.$

Thus, the vertical and horizontal hatched zones show all the domain in which it is possible to be placed while satisfying this property. For example, let us consider the case of cobalt (curve iso-M.sub.S=Co) with a thickness L=15 nm (blue axes). The optimal conditions

[00045]$\left(.Math.\frac{}{D}.Math.=0\right)$

of this case correspond to the maximum of the iso-M.sub.S and roughly corresponds to the operating point {D=10.9 nm, =60}. On the other hand, if the criterion is

[00046]$.Math.\frac{}{D}.Math.<10.Math..Math.{\mathrm{nm}}^{-1}$

for example instead of being

[00047]$.Math.\frac{}{D}.Math.=0$

but that it is specifically wished to have =55, then it is possible by moving towards the left on the iso-M.sub.S to find an operating point {D=8.8 nm, =55} which always satisfies the desired criterion. The criterion is no longer satisfied for around D<8.1 nm, which leaves the possibility of having a method variability of 0.7 nm.

[0262] FIG. 16 shows another manner of presenting the influence of the variability of the thermal stability factor for a given variability in cell diameter D (1 nm or 2 nm in FIG. 16) as a function of the diameter of the cell in different surface and volume anisotropy conditions. For each diameter value, and fixed anisotropy parameters Ks and Ku, the optimal conditions of thickness and of magnetisation {L.sup.opti, M.sub.S.sup.opti} are chosen.

[0263] To do so, the results presented in FIG. 16 are calculated in the following manner. The X-axis represents the target diameter D. The Y-axis represents a variable parameter , K.sub.S or K.sub.u, the two others being fixed. For each point of the diagrams, the optimal parameters {L.sup.opti, M.sub.S.sup.opti} are calculated via the method described previously. By using all of these parameters, the relative variation of maximal stability (in % of the target stability ) during a variability of the diameter DD is plotted in colour code. In other words, for all values of diameter [DD; D+D], there is a stability [(1); (1+)]. The diagrams are calculated for D=1 nm (left column) and D=2 nm (right column). These diagrams make it possible to conclude that for the largest diameters (D>15 nm), it may be advantageous to have a surface or volume anisotropy since the latter reduce the variability . On the other hand they only have little influence at small diameters (D<15 nm). In addition, for the largest diameters, the variability is generally less than 2%, which is low and which makes it possible to be positioned relatively far from the optimal conditions presented in the present patent. On the other hand, the smaller the diameter, the more the variability at the optimal conditions increases. As an example, reaches some 40% for D=1 nm at D=2.7 nm, and reaches some 40% for D=2 nm at D=5.3 nm. Thus, if D=2 nm is the uncertainty on the method and =40% the maximum allowed stability variation, then if the target diameter is 5 nm, then it is necessary to be placed very close to the optimal conditions {L.sup.opti, M.sub.S.sup.opti} since elsewhere the variability can only get worse.

[0264] It is to be noted that in certain situations and in particular towards the smallest technological nodes (4 nm), it may become impossible to find a material having an optimal magnetisation M.sub.S.sup.FM opti, whether it is for purely physical reasons (for example no material having the optimal magnetisation if it is too high) or because said materials do not satisfy other criteria (for example necessity of having a low Gilbert dampening). This can take place in particular towards the smallest technological nodes in so far as, generally speaking, the optimal magnetisation of the material of the ferromagnetic storage layer increases when the diameter of the tunnel junction decreases. In this case, if it becomes impossible to satisfy the equality

[00048]$\frac{}{D}=0,$

it will be sought to be placed at the lowest accessible point

[00049]$\left(.Math.\frac{}{D}.Math.\right).$

An analysis of the diagrams such as that given in FIG. 15 leads to the fact that it is necessary to choose the material having the greatest magnetisation M.sub.S (with M.sub.S<M.sub.S.sup.opti), then to adapt the thickness to obtain the desired stability. In this case, the thickness is obtained according to:

[00050]$L=\frac{1}{2.Math.\left({\mathrm{abD}}^2+{\mathrm{bcD}}^2.Math.K_u\right)}.Math.\begin{array}{c}[\left(2.Math.{\mathrm{aD}}^3+b.Math..Math.-{\mathrm{cD}}^3.Math.K_u-{\mathrm{bcD}}^2.Math.K_S\right)+\\ \sqrt{{\left(2.Math.{\mathrm{aD}}^3+b.Math..Math.-{\mathrm{cD}}^3.Math.K_u-{\mathrm{bcD}}^2.Math.K_S\right)}^2+4.Math.\left({\mathrm{abD}}^2+{\mathrm{bcD}}^2.Math.K_u\right).Math.\left(D.Math..Math.-{\mathrm{cD}}^3.Math.K_S\right)}]\end{array}$

[0265] It should be noted that the values of L.sup.FM opti and Ms.sup.FM opti depend on temperature. Advantageously, for a device having to operate over a temperature range of T.sub.min to T.sub.max, for example 0 C. to 85 C., it will be sought to come closer to the optimal conditions at the average temperature (T.sub.min+T.sub.max)/2.

[0266] It should be noted that is it also possible to give to the reference layer a vertically elongated shape in order to increase its perpendicular anisotropy. However a drawback is that then the field radiated by this layer on the storage layer becomes very important such that it can become necessary to compensate this radiated field by the application of a compensation magnetic field.