EP1595263A2

EP1595263A2 - Spin detection magnetic memory

Info

Publication number: EP1595263A2
Application number: EP03814485A
Authority: EP
Inventors: Viatcheslav Safarov
Original assignee: Centre National de la Recherche Scientifique CNRS; Universite de la Mediterranee
Current assignee: Centre National de la Recherche Scientifique CNRS; Universite de la Mediterranee
Priority date: 2002-12-27
Filing date: 2003-12-22
Publication date: 2005-11-16
Also published as: WO2004061856A2; WO2004061856A3; CN1745430A; JP2006512763A; AU2003303658A1; AU2003303658A8; FR2849526B1; FR2849526A1

Abstract

The invention relates to a spin detection magnetic memory disposed on a semiconductor junction (103) formed by 2 adjacent zones, the first (101) and the second (102) zones having a conductivity which is respectively a first and second type; said memory comprises a first (110) and a second (120) connection cell disposed on the sides of said junction (103), each cell being provided with a magnetization module (111-112, 121-122). At least one of said cells comprises a polarization electrode (113, 123) on top of the magnetization module.

Description

The present invention relates to a magnetic memory with spin detection.

Magnetic memories on silicon, also called “MRAM” for the English term “Magnetic Random Access Memory” have experienced very rapid development in recent years and reference may be made to this subject for example in US patent US Pat. They indeed have many advantages such as the non-volatility of the "FLASH" memory, the speed of the "SRAM" memory and the density of the "DRAM" memory. In addition to these numerous advantages, they also offer very low voltage operation.

However, the MRAM manufacturing process is complex. It requires very precise control of certain parameters as well as the use of uncommon materials. These memories thus prove difficult to industrialize profitably and, so far, only a few prototypes have been produced.

A first type of spin transistor such as that described in US Pat. No. 5,654,566 is presented as a field effect transistor except that the source, respectively the drain, are replaced by an injector, respectively a detector. of spin-polarized electrons, both made of magnetic magnetic material.

Spin polarized electrons are injected from the injector into the transistor channel. They drift under the effect of the electric field applied between the injector and the detector. The grid is used to manipulate spins (change their orientation) during the path from the injector to the detector. The electrical potentials of the three elements, injector, gate and detector, are conditioned by the operation of the transistor, so that it is not possible to freely modify them to optimize the injection of spin-polarized electrons in the channel, nor to optimize the detection of these same electrons. The bipolar type spin transistor such as that taught by American patent US 5,962,905 is also known. Here, the emitter and the base are each covered with a magnetized magnetic layer. Although these two elements are separated by a semiconductor junction, the range of adjustment of their potential remains very limited. The present invention thus relates to a magnetic memory with spin detection in which the injection and / or the detection of electrons polarized in spin are significantly improved.

According to the invention, the memory is arranged on a semiconductor junction formed by two adjacent zones, the first and the second zone having a conductivity of a first and a second type respectively, this memory comprising a first and a second connection cells arranged on either side of this junction, each cell being provided with a magnetization module; in addition, at least one of these cells has a polarization electrode in addition to its magnetization module.

The addition of an electrode near the magnetization module makes it possible to modify the polarization of this module without unduly disrupting the operation of the memory.

Preferably, one of the magnetization modules adjoins the semiconductor junction.

According to a preferred embodiment, at least one of these magnetization modules comprises a buffer layer in contact with the zone in which it appears, a magnetic layer being disposed on this buffer layer.

Advantageously, this buffer layer is made of an insulating material and, according to an additional characteristic, its thickness is such that it allows conduction by tunnel effect between the magnetic layer and the zone in which it appears.

In addition, the distance between the two magnetization modules of the memory is less than double the spin diffusion length. In addition, the first zone has a p-type conductivity.

The present invention will now appear in more detail in the context of the description which follows of exemplary embodiments given by way of illustration with reference to the appended figure which represents a diagram of the memory with detection of spin according to the invention. With reference to FIG. 1, the magnetic memory is arranged on a semiconductor substrate 100.

The substrate 100 has a first zone 101 on which a first connection cell 110 is arranged. This first zone 101 has a conductivity of a first type, type p in this case, while the rest of the substrate which constitutes a second zone 102, presents a conductivity of the second type, of type n in this case. The separation of the two zones thus forms a semiconductor junction 103.

The first connection cell 110 is here a spin polarized electron injector. It comprises a first magnetization module formed by a first buffer layer 111 in contact with the first zone 101 and by a first magnetic layer 112 disposed on this first buffer layer.

Preferably, this first magnetization module is disposed in the immediate vicinity of the semiconductor junction 103.

The spin-polarized electrons are injected from the first magnetic layer 112 into the first zone 101.

To inject and detect spin-polarized electrons, you need materials that have a strong electronic spin polarization: ferromagnetic materials are naturally good candidates. These materials can be insulating, semiconducting or metallic. For electronic devices such as memories, it is preferable to use ferromagnetic metals because ferromagnetic semiconductors are materials which have been synthesized recently and their technology is not yet well mastered. In addition, the Curie temperature of these materials is quite low, less than 300 ° K, and these materials therefore cannot be used at room temperature. On the other hand, conductive ferromagnetic materials have very high Curie temperatures, well above 300 ° K. Their technology is well mastered and there is a wide range of ferromagnetic metals (pure and alloys) with varied magnetic properties (coercive field, magnetic anisotropy ...). The injection of electrons from ferromagnetic metals can be done in different ways, in particular by means of a tunnel junction. Indeed, experiments carried out with ferromagnetic metals show that the electrons emitted by these metals through tunnel junctions are strongly polarized in spin. Thus, preferably, the first buffer layer 111 is made of an insulating material such as silicon dioxide or alumina.

It has a sufficiently fine thickness, from a fraction of a nanometer to a few nanometers, so that the conduction between the first magnetic layer 112 and the first zone 101 is dominated by the tunnel effect. The stack of first zone 101, first buffer layer 111, first magnetic layer 112, therefore constitutes a tunnel junction. So that this tunnel junction can be polarized directly, the first magnetization cell comprises a polarization electrode 113 in ohmic contact with the first zone 101. It suffices that a relatively low voltage, of the order of a few volts, is applied between the first magnetic layer 112 and the first zone 101 to curve the strips in the semiconductor 101 near the interface with the first buffer layer 111. The injection of electrons into the conduction strip of this semi -conductor is then insured.

On the other hand, a second connection cell 120 acts as a detector of spin-polarized electrons. Arranged on the second zone 102, it comprises a second magnetization module which, preferably, is formed of a second buffer layer 121 in contact with the second zone 102 and of a second magnetic layer 122 disposed on this second buffer layer . It was mentioned above that a tunnel junction makes it possible to significantly increase the injection efficiency of polarized electrons. Such a junction similarly improves the detection of these polarized electrons because the probability of an electron passing through a ferromagnetic material through this junction very strongly depends on its spin orientation. Thus, advantageously, the second buffer layer 121 is made of insulating material to produce a second tunnel junction materialized by the stacking of the second zone 102, second buffer layer 121, second magnetic layer 122.

To improve the spin selectivity of the detection, it is preferable to have a second polarization electrode 123 in ohmic contact with the second zone 102. As an example, the potential difference between the second magnetic layer 122 and this second electrode polarization is of the order of a few volts.

The current injected by the first connection cell 110 to the second connection cell is spin polarized. In other words, it is mainly made up of electrons of a spin type, "spin up" or "spin down". The rate of current polarization is determined by the band structure of the magnetic material at the interface with the buffer layer. The spin polarization depends on the orientation of the magnetization of the ferromagnetic metal. The injected current / has two components G + and G-, each representing the electron current respectively of "spin up" or "spin down".

At the second connection cell, the injected current is subdivided into a detection current picked up by the second magnetic layer 122 and a leakage current picked up by the second polarization electrode 123.

The detection current and the leakage current depend on the relative magnetization of the two magnetization modules.

We denote i _p , respectively i _ap , the detection current when the magnetizations of the two modules are parallel, respectively antiparallel. We also note j _p , respectively j _ap , the leakage current when the magnetizations of the two modules are parallel, respectively antiparallel.

The transmission probabilities of the “spin up” and “spin down” electrons in the second magnetic layer are characterized by the coefficients α + and α-, the probability of passing to the ohmic contact being characterized by the coefficient β which is independent of spin polarization.

In parallel configuration, the different currents are linked to the n + and n concentrations. of spin up and spin down electrons as follows:

7 = G ₊ + G_; i _p = a ₊ n ₊ + a_ n_; } _p = β [n ₊ + n_)

In steady state and always for a parallel configuration of the relative magnetizations of the injectors and detectors we have:

G ₊ = ₊ n ₊ + βn ₊ ; G _ = a _ n _ + βn _

_{n +} = ^{G +} _n : n - = —G- = - α + + β α - + β

_Λ a. i _n =. n ₊ + _n_ = ^ G. -f -. G ^{p +} a ₊ + β ⁺ _ + β

When the injector and detector magnetizations are in antiparallel configuration, the detected current is modified compared to the parallel configuration: G_ = a ₊ n ₊ + βn ₊ ; G ₊ = a_n_ + βn_

n- _ G n ₊ = - ^ α- + β α + + β

i _ap = a ₊ n ₊ + a_n_ = a _Λ .

G_ + G + a ₊ + β _ + β

It follows that:

* ^p „+ z ^• _ + ^ap a ₊ + β ⁺ + CJ - ⁱ + - a_ ^ + β (G ₊₊ G_)

We introduce the following notations to quantify the asymmetries rant injected, the detection and the detected current:

G ₊ = G + ΔG; G- = G-ΔG; ₀ ώG = - ^ - ^ etΔ? = ^ - ^

The quantity which characterizes the sensitivity of the detector is Aili, relative variation of the detection current for two magnetization configurations:

Assuming that (Δa) ² is significantly less than a, we obtain the following relationship:

AG / G characterizes the spin polarization of the injected electrons. Δa / a characterizes the anisotropy of transmission of the detector. The ratios AG / G and Δα / α are worth a few tenths of a unit, around 0.4 for an alloy of iron and cobalt. The sensitivity limit therefore depends solely on the properties of the ferromagnetic structures. It is reached for a "or for /" j. The detector therefore has a lower current (compared to the injected current), but with maximum sensitivity to spin polarization in the semiconductor. For a detected current representing 10% of the injected current, the sensitivity of the detector will be equal to 90% of the limit sensitivity, given by the choice of ferromagnetic materials.

By qualifying as a collector the space appearing between the two magnetization modules, this collector contains a non-negligible concentration of non-polarized electrons of spin when the injected current is zero. As spin-polarized electrons are injected, these electrons gradually replace non-polarized electrons. In the collector in steady state, a spin polarization distribution P is established which has the following form:

P (x) = P (0) exp (- x / L _s ) where x is the distance from the electron to the semiconductor junction 103 and L _s the spin diffusion length.

with D the electron diffusion coefficient and τ _s the spin relaxation time.

It is therefore preferable that the distance d which separates the two magnetization modules is less than the scattering length L _s , although this distance d may be greater, twice the scattering length L _s for example, this at detriment to the sensitivity of the detector. In silicon at room temperature, the diffusion coefficient of the carriers and the spin relaxation time are high enough for the electrons to retain their spin over diffusion lengths of several microns. The relaxation times of conduction electron spins, measured by so-called "RPE" techniques (for Electronic Paramagnetic Resonance), are of the order of 10 ^"8 s. This leads to a value of L _s , the length of diffusion, of the order of a few microns. For distances d less than L _s , spin relaxation is therefore a negligible phenomenon and spin becomes a characteristic specific to each electron.

The memory according to the invention can in particular be manufactured in the following manner. The process up to the contact part is a traditional CMOS manufacturing process. Before opening the contacts or after the contact has been filled with a metal, an additional step is introduced. An insulator a few nanometers thick is deposited; this insulator can be silicon dioxide, alumina or any other known dielectric. Next, the ferromagnetic material is deposited, an alloy of cobalt and iron for example. The two constraints imposed on the materials are to have an abrupt interface with the dielectric while maintaining a high electronic polarization at the interface. The thickness of deposited magnetic material can vary from a few tens to a few hundred nanometers. Then is carried out the deposit of a traditional metal such as copper or aluminum, or any other material ensuring good electrical continuity. The circuit is then polluted mechanically and chemically in order to leave only the magnetic material in the injection and detection zones. The process can then resume its traditional course. The memory is written or erased with a magnetic field which returns the magnetization of the first 112 or of the second 122 magnetic layer. Since the current passing through the detector depends on the relative orientation of the magnetizations of the injectors and detectors, the magnetic state of the cell is read with the current passing through the detector. As in the state of the art, the writing of the memory can be accomplished by the passage of a current through two insulated metallic conductors which cross over the magnetic layer which it is magnetize.

When a saturation current crosses these two conductors, the magnetic field generated at the intersection of these is sufficient to change the magnetization configurations from a parallel state to an antiparallel state. The saturation current is chosen so that the combined magnetic field exceeds the critical field of the ferromagnetic metal, predominantly determined by magnetic anisotropy. In addition, if this saturation current is applied to only one of the two conductors, the magnetic field generated is insufficient to return the magnetization. Finally, the arrangement of the conductors is such that the field generated by the saturation current is very localized. This field is less than the field necessary to modify the magnetization of other magnetic elements which would be located near the intersection of the two conductors. The two possible directions of magnetization then define two possible logical states (commonly denoted 0 or 1) of the memory.

Naturally, several individual memories or units such as that described above can be combined to produce a storage unit. The structure of this unit allows it to be integrated with elementary components such as transistors, diodes or capacitors. These components make it possible to manipulate the reading current through the various units and thus lead to a random access storage unit ("RAM"). Currently, all of the non-volatile memories (“EEPROM”,

"FLASH", "FeRAM", "MRAM") use manufacturing processes that do not are not standard. Manufacturing requires the addition of four to five levels of masking, resulting in an additional cost of about 20%.

According to the invention, it becomes possible to manufacture non-volatile memories with a conventional CMOS process, without considerably increasing the different levels of masking. In addition, compared to “Flash” memories, the memory according to the invention operates at low voltage and does not require a charge pump. This is a decisive advantage for mobile applications.

The invention is particularly suitable for so-called “System On Chip” or “SOC” technology. SOC technology integrates all of the components on a single chip: micro-controller, “SRAM” and “DRAM” memories, dedicated logic, “MEMS”, chemical sensors and of course, non-volatile memories. It is therefore necessary to have the most standardized manufacturing process possible. The embodiment of the invention presented above was chosen for its concrete character. However, it would not be possible to exhaustively list all the embodiments covered by this invention. In particular, any means described may be replaced by equivalent means without departing from the scope of the present invention.

Claims

1) Spin detection magnetic memory arranged on a semiconductor junction 103 formed of two adjacent zones, the first 101 and the second 102 having a conductivity of a first and a second type respectively, comprising a first 110 and a second 120 connection cells arranged on either side of said junction 103, each cell being provided with a magnetization module 111-112, 121-122, characterized in that at least one of these cells comprises a polarization electrode 113, 123 in addition to said magnetization module.

2) Memory according to claim 1, characterized in that one of said magnetization modules 111-112 adjoins said junction.

3) Memory according to any one of claims 1 or 2, characterized in that at least one of said magnetization modules comprises a buffer layer 111, in contact with said area 101, a magnetic layer 112 being disposed on this layer buffer.

4) Memory according to claim 3, characterized in that said buffer layer 111 is made of an insulating material.

5) Memory according to claim 4, characterized in that the thickness of said buffer layer 111 is such that it allows conduction by tunnel effect between said magnetization layer and said area.

6) Memory according to any one of the preceding claims, characterized in that the distance between the two magnetization modules 111-112, 121-122 is less than twice the length of spin diffusion.

7) Memory according to any one of the preceding claims, characterized in that said first zone 101 has a p-type conductivity.