All-spin logic devices

Abstract

Illustrative embodiments of all-spin logic devices, circuits, and methods are disclosed. In one embodiment, an all-spin logic device may include a first nanomagnet, a second nanomagnet, and a spin-coherent channel extending between the first and second nanomagnets. The spin-coherent channel may be configured to conduct a spin current from the first nanomagnet to the second nanomagnet to determine a state of the second nanomagnet in response to a state of the first nanomagnet.

Description

BACKGROUND

The present disclosure relates generally to All-Spin Logic Devices from which all-spin architectures may be constructed. The possible role of spin as an alternative to charge for logic applications has been recognized. Most spin-based proposals, however, use spin only as an internal variable (i.e., the terminal quantities of each individual logic gate are still charge-based). It has also been pointed out that collective entities like magnets may require significantly less switching energy than an equivalent number of non-interacting spins. Recently, there has been significant experimental progress in implementing different forms of magnet-based switching circuits. For instance, in the Magnetic Quantum Cellular Automata (MQCA) architecture, each magnet controls the switching of a neighboring magnet through its magnetic field. In MQCA circuits, interconnects are made of an array of magnets, which also require clocking by external magnetic fields. This scheme is not well-suited for general purpose computing due to the limited control over the nearest-neighbor interconnections.

SUMMARY

According to one aspect, an all-spin logic device may comprise a first nanomagnet, a second nanomagnet, and a spin-coherent channel extending between the first and second nanomagnets. The spin-coherent channel may be configured to conduct a spin current from the first nanomagnet to the second nanomagnet to determine a state of the second nanomagnet in response to a state of the first nanomagnet.

In some embodiments, the first and second nanomagnets may have identical switching characteristics. The all-spin logic device may further comprise a tunnel barrier disposed at an interface between the spin-coherent channel and one of the first and second nanomagnets. The second nanomagnet may comprise a free layer having an easy axis and a fixed layer having an easy axis, where the easy axis of the fixed layer is perpendicular to the easy axis of the free layer.

In other embodiments, the first and second nanomagnets may each be electrically coupled to one un-clocked supply voltage. The second nanomagnet may have a greater spin-torque conductance relative to the spin-coherent channel than the first nanomagnet. The spin-coherent channel may comprise a ground terminal positioned closer to the first nanomagnet than to the second nanomagnet.

In still other embodiments, second nanomagnet may be electrically coupled to a floating voltage. Alternatively, the second nanomagnet may be grounded. The first and second nanomagnets may each have an energy barrier of at least one-quarter electron-volt and may each comprise less than 10⁶ Bohr magnetons.

According to another aspect, an all-spin logic circuit may comprise a first nanomagnet having an input side and an output side, a second nanomagnet having an input side and an output side, a third nanomagnet having an input side and an output side, a first spin-coherent channel configured to conduct a spin current generated from the input side of the first nanomagnet to the output side of the second nanomagnet, and a second spin-coherent channel configured to conduct a spin current generated from the input side of the second nanomagnet to the output side of the third nanomagnet.

In some embodiments, the first, second, and third nanomagnets may have identical switching characteristics. The all-spin logic circuit may further comprise a third spin-coherent channel configured to conduct a spin current generated from the input side of the third nanomagnet to the output side of the first nanomagnet. The output sides of the first, second, and third nanomagnets each have a greater spin-torque conductance than the input sides of the first, second, and third nanomagnets.

In other embodiments, the first spin-coherent channel may comprise a ground terminal positioned closer to the input side of the first nanomagnet than to the output side of the second nanomagnet, and the second spin-coherent channel may comprise a ground terminal positioned closer to the input side of the second nanomagnet than to the output side of the third nanomagnet. The first and second spin-coherent channels may be separated by an isolation layer. The first, second, and third nanomagnets may each be electrically coupled to one unclocked supply voltage.

According to yet another aspect, a method may comprise applying a voltage to a first nanomagnet to generate a first spin current in response to a magnetization direction of the first nanomagnet and routing the first spin current along a first spin-coherent channel to a second nanomagnet to determine a magnetization direction of the second nanomagnet.

In some embodiments, the magnetization direction of the second nanomagnet does not determine the magnetization direction of the first nanomagnet. The method may further comprise applying a voltage to a third nanomagnet to generate a second spin current in response to a magnetization direction of the third nanomagnet and routing the second spin current along the first spin-coherent channel to the second nanomagnet to determine the magnetization direction of the second nanomagnet in response to a superposition of the first and second spin currents.

In other embodiments, the method may further comprise applying a voltage to the second nanomagnet to generate a second spin current in response to the magnetization direction of the second nanomagnet and routing the second spin current along a second spin-coherent channel to a third nanomagnet to determine a magnetization direction of the third nanomagnet. The method may also comprise applying a voltage to the third nanomagnet to generate a third spin current in response to the magnetization direction of the third nanomagnet and routing the third spin current along a third spin-coherent channel to the first nanomagnet to determine the magnetization direction of the first nanomagnet. Applying a voltage to the first nanomagnet and applying a voltage to the second nanomagnet may comprise applying one un-clocked supply voltage to both the first and second nanomagnets.

In still other embodiments, the method may further comprise applying a clocked supply voltage to a fixed layer of the second nanomagnet to place a free layer of the second nanomagnet in a neutral state while receiving the first spin current from the first spin-coherent channel. The method may further comprise applying a floating voltage to the second nanomagnet while receiving the first spin current from the first spin-coherent channel. The method may further comprise grounding the second nanomagnet while receiving the first spin current from the first spin-coherent channel. The magnetization direction of the second nanomagnet may oscillate with the magnetization direction of the first nanomagnet.

BRIEF DESCRIPTION OF THE DRAWINGS

The invention described herein is illustrated by way of example and not by way of limitation in the accompanying figures. For simplicity and clarity of illustration, elements illustrated in the figures are not necessarily drawn to scale. For example, the dimensions of some elements may be exaggerated relative to other elements for clarity. Further, where considered appropriate, reference labels have been repeated among the figures to indicate corresponding or analogous elements.

FIG. 1A illustrates one embodiment of an all-spin logic device (ASLD).

FIG. 2 illustrates another embodiment of an ASLD in which each nanomagnet includes a free layer and a fixed layer with perpendicular easy axes.

FIG. 3A illustrates an energy landscape of one embodiment of a bistable nanomagnet that may be used in the ASLD of FIG. 2.

FIG. 3B illustrates one embodiment of clocked voltages that may be applied to the nanomagnets of the ASLD of FIG. 2.

FIG. 3C illustrates logical switching of a bistable output nanomagnet having the energy landscape of FIG. 3A.

FIG. 4A illustrates the deflection required for logical switching of an output nanomagnet as a function of bias magnitude for various alignments of an input nanomagnet.

FIG. 4B illustrates one embodiment of input-output transfer characteristics for an ASLD.

FIG. 5A illustrates a phase diagram for a single-pulse switching scheme.

FIG. 5B illustrates a phase diagram for a two-pulse switching scheme.

FIG. 6A illustrates one embodiment of logic gates comprising a number of ASLD, during a first phase of an exemplary clocking cycle.

FIG. 6B illustrates the logic gates of FIG. 6A, during a second phase of the exemplary clocking cycle.

FIG. 6C illustrates the logic gates of FIG. 6A, during a third phase of the exemplary clocking cycle.

FIG. 7A illustrates another embodiment of an ASLD in which the spin-coherent channel comprises a ground terminal positioned closer to an input nanomagnet than to an output nanomagnet.

FIG. 7B illustrates a state diagram for the nanomagnets of the ASLD of FIG. 7A.

FIG. 7C illustrates one embodiment of a coupled spin-transport/magneto-dynamics model for the ASLD of FIG. 7A.

FIG. 7D illustrates one embodiment of a distributed conductance network model for the ASLD of FIG. 7A.

FIG. 7E illustrates components of the spin voltage in a spin-coherent channel of the ASLD of FIG. 7A.

FIG. 7F illustrates a ratio of torques exerted on the input and output nanomagnets of the ASLD of FIG. 7A as a function of contact length normalized to transfer length.

FIG. 8A illustrates one embodiment of un-clocked supply voltages that may be applied to the input and output nanomagnets of the ASLD of FIG. 7A.

FIG. 8B illustrates a magnetization direction of the output nanomagnet of the ASLD of FIG. 7A in response to the un-clocked supply voltages of FIG. 8A, where the input nanomagnet has a magnetization direction representing a logical zero.

FIG. 8C illustrates a magnetization direction of the output nanomagnet of the ASLD of FIG. 7A in response to the un-clocked supply voltages of FIG. 8A, where the input nanomagnet has a magnetization direction representing a logical one.

FIG. 9 illustrates logical switching voltages for the input and output nanomagnets of the ASLD of FIG. 7A.

FIG. 10A illustrates one embodiment of a cascaded chain of ASLD including first, second, and third nanomagnets.

FIG. 10B illustrates a magnetization direction of each of the first, second, and third nanomagnets of the ASLD of FIG. 10A after a positive supply voltage has been applied, where the input nanomagnet has a magnetization direction representing a logical zero.

FIG. 11A illustrates another embodiment of a cascaded chain of ASLD forming a ring oscillator including first, second, and third nanomagnets.

FIG. 11B illustrates an oscillating magnetization direction of each of the first, second, and third nanomagnets of the ring oscillator of FIG. 11A after a positive supply voltage has been applied.

FIG. 12 illustrates one embodiment of a circuit representation of an ASLD.

FIG. 13A illustrates one embodiment of an ASLD in which the spin-coherent channel comprises a ground terminal positioned equidistant between an input nanomagnet and an output nanomagnet.

FIG. 13B illustrates an oscillating magnetization direction of each of the input and output nanomagnets of the ASLD of FIG. 13A after the output nanomagnet has been grounded and a positive supply voltage has been applied to the first nanomagnet.

FIG. 14 illustrates logical switching speeds of two ASLD including nanomagnets with the same energy barrier, but different numbers of Bohr magnetons (N_s) and different internal uniaxial anisotropies (H_K).

FIG. 15 illustrates the energy barrier of a nanomagnet as a function of the number of Bohr magnetons (N_s) in the nanomagnet and the internal uniaxial anisotropy (H_K) of the nanomagnet.

DETAILED DESCRIPTION OF THE DRAWINGS

While the concepts of the present disclosure are susceptible to various modifications and alternative forms, specific exemplary embodiments thereof have been shown by way of example in the drawings and will herein be described in detail. It should be understood, however, that there is no intent to limit the concepts of the present disclosure to the particular forms disclosed, but on the contrary, the intention is to cover all modifications, equivalents, and alternatives consistent with the present disclosure and appended claims.

References in the specification to “one embodiment,” “an embodiment,” “an example embodiment,” etcetera, indicate that the embodiment described may include a particular feature, structure, or characteristic, but every embodiment may not necessarily include the particular feature, structure, or characteristic. Moreover, such phrases are not necessarily referring to the same embodiment. Further, when a particular feature, structure, or characteristic is described in connection with an embodiment, it is submitted that it is within the knowledge of one skilled in the art to effect such feature, structure, or characteristic in connection with other embodiments whether or not explicitly described.

The present disclosure relates to All-Spin Logic Devices (ASLD) comprising a number of nanomagnets linked by a number of spin-coherent channels. The nanomagnets of the ASLD may be switched between their stable states representing binary data (e.g., right-magnetized or left-magnetized in FIGS. 1, 2, 3C, 6A-C, 7A, 10A, 11A, and 13A) if enough torque is exerted on them. Information stored in the magnetization direction of an input nanomagnet may be used to generate a spin current that can be routed along a spin-coherent channel to any desired location, where the spin current determines the final state of an output nanomagnet based on spin-torque phenomenon. Overall, what is achieved is the switching of the output nanomagnet in accordance with the information provided by the input nanomagnet through a spin current transmitted along the spin-coherent channel.

The presently disclosed ASLD may operate without the use of conventional charge current, even at its terminals. The nanomagnets act as digital spin capacitors that also provide non-volatile memory. The ASLD exhibits the five essential characteristics required for logic applications. First, both input and output information in the ASLD are in the same form, namely magnetization, providing concatenability. Second, the nanomagnets of the ASLD are intrinsically nonlinear (as further described below with reference to FIGS. 3A-F and 4A) and automatically correct any errors in magnetization direction. Third, the resulting input-output transfer characteristics are sharp with large gain (as further described below with reference to FIG. 4B). Fourth, the ASLD may be used to build a complete set of Boolean logic gates (as further described below with reference to FIGS. 6A-C). Fifth, the ASLD exhibits non-reciprocal information transfer from the input magnet to the output magnet. Several methods of minimizing feedback from the output magnet to the input magnet to achieve this non-reciprocity are contemplated, as will be further described below. Thus, architectures employing the ASLD afford versatility comparable to standard charge-based architectures while having the potential for low power operation and continued scaling.

Unlike known MQCA circuits, which use magnetic fields, the ASLD uses spin currents, which may lead to significant increases in both scalability and versatility. As spin currents are not limited to nearest neighbor communication, the ASLD will not be inherently limited to cellular architectures. Instead, the range of communication is limited to a spin diffusion length. At room temperature, using a silicon diffusion constant of D=36 cm²/second and a lifetime of 10 nanoseconds, the spin diffusion length would be approximately 6 microns. Drift in an electric field may be used to increase the diffusion length. The range of a few microns is large enough to allow considerable flexibility and versatility of architecture. Moreover, information is transferred in an ASLD with Fermi velocity, which may be compared to magnon velocity, as in the case of MQCA devices. In some embodiments, larger interconnects (e.g., a few millimeters) in an ASLD-based system could be electrical, current-based interconnects. Thus, an entire circuit may comprise blocks of all-spin logic, which may be connected by electrical interconnects, if necessary.

As compared to conventional CMOS devices, the ASLD has the potential for extremely low switching energies (i.e., the energy dissipated throughout switching). The self-correcting feature provided by magnets may make it possible to reduce the switching energy to several k_BT (where T is the room temperature) per magnet rather than several k_BT per spin. Although the ASLD is based on the physics of spin-torque, it does not require the use of structures or phenomena commonly associated with spin-torque, such as the tunneling magneto-resistance (TMR) of high resistance tunnel junctions. The dissipation in existing spin-torque devices is far in excess of the theoretical minimum. Power dissipation, along with other factors that affect scaling of the ASLD, are discussed in more detail below.

Referring now to FIG. 1, one illustrative embodiment of an ASLD 100 includes first and second nanomagnets 102 and a spin-coherent channel 104. As noted above, binary data is stored in a non-volatile manner in the state (i.e., magnetization direction) of each of the nanomagnets 102. The spin-coherent channel 104 extending between the nanomagnets 102 may comprise a metal or a semiconductor, with the latter having a higher spin coherence length. In the illustrative embodiment of FIG. 1, the left nanomagnet 102 acts as an input and the right nanomagnet 102 acts as an output. When a supply voltage is applied, the input magnet 102 generates a spin-current 106 that is routed along the spin-coherent channel 104 to the output magnet 102 to determine the state of the output magnet 102 by spin-torque phenomena. One advantage of the ASLD 100 is that, in some embodiments, the first and second nanomagnets 102 may have identical switching characteristics (i.e., the nanomagnets 102 would both switch if subjected to the same spin-transfer torque) and, hence, may be identical structures.

The ASLD 100 may also include one or more isolation layers 108 that separate the spin-coherent channel 104 from other spin-coherent channels in the architecture, preventing unwanted cross-talk between the channels. The isolation layers 108 may be comprise electrostatic barriers (e.g., oppositely doped semiconductor) and/or insulation layers (e.g., an oxide). Where an isolation layer 108 is placed under a nanomagnet 102, the nanomagnet 102 may interface with two (or more) spin-coherent channels 104. As such, each nanomagnet 102 may have an input side 110 (i.e., a “talking” side) that transmits information via spin currents 106 and an output side 112 (i.e., a “listening” side) that receives information via spin currents 106, providing non-reciprocity.

In some embodiments, the ASLD 100 may further include a tunnel barrier 114 at the interface between the input side 110 of each nanomagnet 102 and the spin-coherent channel 104 to the increase spin injection efficiency (while an ohmic contact at the interface between the output side 112 of each nanomagnet 102 and the spin-coherent channel 104 prevents back-injection of spins). The tunnel barriers 114 may comprise oxides and/or Schottky barriers. In other embodiments, the tunnel barriers 114 may be disposed in other portions of the spin-coherent channel 104. It will be appreciated that, while the tunnel barriers 114 may be included in any embodiment of the presently disclosed ASLD to increase spin injection efficiency, the tunnel barriers 114 need not be included in every embodiment.

Referring now to FIG. 2, another illustrative embodiment of an ASLD 200 is shown. The ASLD 200 of FIG. 2 has a similar configuration to the ASLD 100 of FIG. 1, except that the first and second nanomagnets 102 each comprise a free layer 102A and a fixed layer 102B with perpendicular easy axes (while the easy planes of the two layers 102A-B are aligned in parallel). The magnetization direction of the free layer 102A represents the binary data stored in each nanomagnet 102. In some embodiments of the ASLD 200, the free layer 102A of a bistable nanomagnet 102 may have an energy landscape similar to that shown in FIG. 3A, in which the two stable states A and C represent binary data. State B of the free layer 102 is a high energy “neutral” state. When the free layer 102A is not under the influence of an external torque, the internal field of the bistable nanomagnet 102 drives its magnetization direction to one of the energy minima, i.e., state A or state C.

The fixed layer 102B may be used to place the magnetization direction of the free layer 102A in a neutral state (e.g., state B of FIG. 3A), ready to receive information via a spin current 106 (when an appropriate supply voltage is applied to the fixed layer 102B, as described further below). Thus, in the ASLD 200, the energy required to switch the output magnet 102 is provided by the supply voltage, and only information is conveyed by the spin current 106. In some embodiments, the fixed layer 102B may comprise a pair of anti-parallel layers, so that the dipolar field of fixed layer 102B does not interfere with the switching process. In other embodiments, the nanomagnets 102 of the ASLD 200 may be implemented using known materials with perpendicular magnetic anisotropy (PMA). Structures which utilize materials with PMA and a planar fixed layer may be used to diminish the effect of the demagnetizing field, which increases the minimum switching current density. Simulations of the ASLD 200 suggest that the amplitude of the Slonczewski term (represented on the horizontal axis of FIG. 5B, discussed below) would be considerably lower if the effect of the demagnetizing field was diminished.

The ASLD 200 illustrated in FIG. 2 has been simulated using the a model based on the Landau-Lifshitz-Gilbert (LLG) equation, appropriately modified to include the established physics of the interactions of nanomagnets with spin-polarized currents (i.e., spin-torque). While large magnets have multiple domains, nanomagnets with sizes less than 100 nm in each dimension may be approximated as a single domain having a macro-spin whose dynamics are commonly modeled using the LLG equation (where {circumflex over (m)} is the normalized magnetic moment):

$\begin{array}{cc}\frac{d\hat{m}}{dt}=-\gamma \hat{m}\times {\overrightarrow{H}}_{\mathrm{eff}}+\alpha \hat{m}\times \frac{d\hat{m}}{dt}.& \left(1\right)\end{array}$ H_eff is a conservative field that includes all internal and external fields and can be written as the gradient of the potential energy with respect to the normalized magnetization components (i.e., H_eff=−(1/M_sV){right arrow over (∇)}E, where M_sV is the magnetic moment if the unit volume V is magnetized to saturation). A nanomagnet may be switched by applying an easy axis field which exceeds the minimum field of switching H_c=2K_u/M_s, where K_u is the effective second order uniaxial anisotropy constant.

A spin current may be used to toggle a nanomagnet between its two stable states. This spin current interacts with the magnetization of the nanomagnet and exerts a torque on it. Equation (1) should therefore be modified to include this effect:

$\begin{array}{cc}\frac{d\hat{m}}{dt}=-\gamma \hat{m}\times {\overrightarrow{H}}_{\mathrm{eff}}+\alpha \hat{m}\times \frac{d\hat{m}}{dt}-\gamma H_{||}\hat{m}\times \left(\hat{m}\times \hat{M}\right)-\gamma H_{\perp }\hat{m}\times \hat{M}.& \left(2\right)\end{array}$ {circumflex over (M)} and {circumflex over (m)} are unit vectors in the direction of the fixed and free layers' magnetizations, respectively. H_∥≡τ_∥/M_sV and H_⊥=≡τ_⊥/M_sV are magnitudes of the magnetic fields that enter the magneto-dynamics equation due to interaction with a spin current. The quantities τ_∥ and τ_⊥ are in units of energy and are directly related to the torque exerted by the spin current. This analysis assumes that the spin current's momentum is fully absorbed by the macro-spin and that its polarization is redirected in the direction of the macro-spin once it passes a few atomic layers inside the nanomagnet. This analysis also assumes that H_∥ and H_⊥ are independent of the relative angle between the two magnetic layers.

For logic applications, a different switching scheme employing two spin currents may be used: a larger spin current providing the energy and a smaller spin current providing the bias that determines the final state of switching. To include the effect of both spin currents, Equation (2) is modified accordingly:

$\begin{array}{cc}\frac{d\hat{m}}{dt}=-\gamma \hat{m}\times {\overrightarrow{H}}_{\mathrm{eff}}+\alpha \hat{m}\times \frac{d\hat{m}}{dt}-\gamma \hat{m}\times \left(\hat{m}\times \left(H_{||}^{\mathrm{clk}}{\hat{M}}_{\mathrm{clk}}+h_{||}^b{\hat{m}}_b\right)\right)-\gamma \hat{m}\times \left(H_{\perp }^{\mathrm{clk}}{\hat{M}}_{\mathrm{clk}}+h_{\perp }^b{\hat{m}}_b\right),& \left(3\right)\end{array}$ where “clk” denotes the torques due to clock spin current and “b” denotes the torques due to the bias spin current. {circumflex over (M)}_clk represents the direction of the fixed magnetic layer, which is along the hard axis of the output free layer, while {circumflex over (m)}_b represents the direction of the bias, which is along the easy axis of the input free layer. The normalized magnetic moment of the output free layer is represented by {circumflex over (m)}. The role of the clock torques is to soften the magnetization by putting it along its hard axis. If there is bias present, it gives the magnetization a tilt from this neutral state: upon removing the clock spin current, magnetization will relax to the state dictated by the tilt. The spin current may be generated by applying a lower voltage, V_bias, to an input nanomagnet 102 of the ASLD 200 and the clock torque may be supplied by applying a higher voltage, V_supply, to an output nanomagnet 102 of the ASLD 200, as illustrated in FIG. 3B. While both torque pulses are illustrated as being turned ON and OFF at the same time in FIG. 3B, the V_bias pulse-width can be lager than that of V_supply, as illustrated by the dotted line. In other words, exact synchronization is not necessary.

This type of logic switching using spin currents, using a free layer 102A and a fixed layer 102B with perpendicular easy axes, is illustrated in FIG. 3C. Irrespective of the initial state of the output nanomagnet 102, its final state is determined solely by the bias provided by the spin current 106. In theory, if the clock spin current is large enough to put the nanomagnet along its hard axis, then any infinitesimal bias spin current can determine the final state of switching. In practice, however, a finite spin current is needed to overcome noise and imperfections. In other words, the value of the deflection A in FIG. 3C should be larger than any deflection due to noise and other sources of error. The dependence of Δ on various pertinent parameters is discussed and quantified below. FIG. 3C was plotted using Equation (3) with the following parameters M_s=1100 emu/cm³ (for Co₆₀Fe₂₀B₂₀), α=0.01, H_c=100 Oe, H_⊥=0.5H_c, H_∥=2.5H_c, H_⊥=0.8H_c, H_∥=5.8H_c, h_∥^b=0.05H_c, h_⊥^b=0.01H_c, and, for volume calculations only, V=180×90×2 nm³.

Since nanomagnets 102 have internal fields that prefer either of two stable states along the easy axis of the free layer 102A, they inherently digitize information which allows for large gains and error correction, as shown in FIG. 3C. After removal of the voltage pulses shown in FIG. 3B, the internal field of the output nanomagnet 102 drives the magnetization to the m_z=1 or m_z=−1 states. As noted above, this process will be successful only if the deflection A of the mid-states of magnetization (the plateaus in FIG. 3C) from m_z=0 is larger than noise fluctuations. Using an equipartition theorem, one can find the most probable deviation angle θ_rms from the easy axis at thermal equilibrium using (where E_ani is the effective anisotropy energy barrier):

$\begin{array}{cc}\sin \left({\theta }_{rms}\right)={\left(\frac{k_{B}T}{2E_{\mathrm{ani}}}\right)}^{\frac{1}{2}}.& \left(4\right)\end{array}$ Noting that (m_z)_rms=cosine(θ_rms) and using a barrier height of E_ani=40 k_BT, the fluctuations in the normalized component of magnetization along the easy axis are less than 0.01. Changes in the deflection Δ as a function of the bias magnitudes h_∥^b and h_⊥^b for various alignments of the input nanomagnet 102 are illustrated in FIG. 4A, in which each pair of curves corresponds to a particular alignment of the input {circumflex over (m)}_b. In each pair, the upper curve corresponds to h_⊥^b=(⅓)h_∥^b and the lower curve corresponds to h_⊥^b=(⅕)h_∥^b. This graph demonstrates that the ASLD 200 has a wide margin for tolerating errors. This can also be seen in FIG. 4B which illustrates the input-output transfer characteristics of the ASLD 200. In practice, there will be a region W, designated by the width δm, for which operation becomes indeterministic. The actual value of δm depends on the values of the biases h_∥^b and h_⊥^b that can be achieved.

The spin current density needed to overcome thermal fluctuations of about 0.01 may also be calculated. Using typical parameters of 170×60×2 nm³ for the volume of the nanomagnet 102, 800 emu/cm³ for saturation magnetization M_s, with an anisotropy barrier height of approximately 40 k_T (for a stability of approximately 10 years), and a damping parameter of α=0:01, a spin current density of about 2.5×10⁴ A/cm² is needed to overcome thermal fluctuations. This spin current density is about two orders of magnitude smaller than the spin current density needed to switch the output nanomagnet 102. Again, in the ASLD 200, the energy needed for switching the output nanomagnet 102 essentially comes from V_supply and not V_bias.

For the graph of FIG. 3C, a specific pair of H_∥^clk and H_⊥^clk was chosen for use in Equation (3). However, the actual ratio of H_∥^clk to H_⊥^clk depends on experimental circumstances and changes as the applied voltage across the nanomagnet 102. What is needed for operation of the ASLD 200 is a pair of H_∥^clk and H_⊥^clk that can align the magnetization along its hard axis. Since H_∥^clk enters Equation (3) as a non-conservative field, it is not possible to predict the final equilibrium state analytically. Simulations were performed over a wide range of H_∥^clk and H_⊥^clk to find various combinations of these quantities that result in a successful logic switching event.

A phase diagram for a single-pulse switching scheme was plotted using Equation (2) and is illustrated in FIG. 5A. This figure shows the case of anti-parallel to parallel switching under the influence of both H_∥ and H_⊥. These two variables were varied independently to produce the two-dimensional grid shown in FIG. 5A with approximately 18,000 data points. The white areas 500 of the grid represent the regions for which switching occurs. If H_∥ was zero, with H_⊥=τ_⊥/M_sV independent of θ, the last term of Equation (2) behaves as an ordinary magnetic field (as compared to the first term), and hence the magnet switches when H_⊥=Hc. In nanomagnets, however, the dominant term is the third term of Equation (2), which is the primary reason behind switching. Moving along the horizontal axis with H_⊥=0, it is recognizable that switching occurs around H_∥≈1.6H_c. The exact value depends on various parameters, such as H_c and the ratio of this value to the demagnetizing field, as well as the damping parameter α. If both H_∥ and H_⊥ are non-zero, then they could assist (or hinder) each other in switching. In experiments, both contributions are usually present. A known experimental value 502 is shown on the diagram in FIG. 5A, demonstrating agreement between this model and experiments. The experimental value 502 corresponds to the minimum experimentally required value for switching; that value resides in the white switching region 500 close to the black non-switching region.

A phase diagram for a two-pulse switching scheme was also plotted using Equation (3) and is illustrated in FIG. 5B. Equation (3) contains four parameters (H_∥^clk, H_⊥^clk, h_∥^b, and h_⊥^b) that depend on the strength of the clock and bias spin currents. In a two dimensional phase diagram, one can freely vary two of these parameters. The effect of h_∥^b, and h_⊥^b was discussed above with regard to when the clock spin torque is just large enough to align the free layer of a nanomagnet along its hard axis. Assuming that these biases are infinitesimally small, the dependence of switching on H_∥^clk and H_⊥^clk may be independently assessed. The switching is successful if the magnetization of the free layer (i.e., the logic bit) is stabilized in a direction slightly tilted from the hard axis towards the direction of the bias (i.e., the z direction). This way the internal field of the nanomagnet will relax the magnetization to the intended stable state once V_supply is removed. The white areas 500 in FIG. 5B show the regions for which this condition is satisfied. For both FIGS. 5A and 5B, the parameters used were: M_s=1100 emu/cm³, α=0.008, H_c=45 Oe, and, for volume calculations only, V=250×70×2 nm³.

To make general purpose digital logic circuits, a minimal set of Boolean logic operations from which all other logical functions can be constructed is needed. A complete minimal set is composed of a basic binary operator like logical AND or logical OR and the unary operator NOT. AND and OR gates can be combined with inversion to make NAND and NOR gates respectively, which are universal logic gates. As will be explained shortly, a minimal set of Boolean logic gates can be constructed using a number of ASLD. When using the ASLD 200, the transfer of information in a chain of concatenated gates can be achieved by proper clocking As will be described further below, when using other embodiments of the ASLD, clocking may not be necessary.

One illustrative embodiment of a digital logic circuit including a number of ASLD 200 is illustrated in FIGS. 6A-C. In this illustrative embodiment, both gate A and gate B are have two data inputs 600, one data output 602, and a fixed input 604 (each comprising one nanomagnet 102). The spin-coherent channels 104 linking the nanomagnets 102 are separated by isolation layers 108. In gate A, both data inputs 600 (as well as the fixed input 604) are in a “send” mode. The final state of the data output 602, in a “receive” mode, is determined by the superposition of spin currents injected from the inputs 600, 604 into the spin-coherent channel 104. The polarity of V_bias determines whether the operation is a COPY or a NOT, because the ferromagnetic contact can be used in a majority spin injection mode or in a minority spin accumulation mode. As such, the gate can function as either an AND/OR logic gate or a NAND/NOR logic gate. The state of the fixed input 604 toggles the functionality between AND and OR, or between NAND and NOR. In the illustrative embodiment, gate B has a similar composition to gate A.

Three phases of clocking and signal transmission from one gate to the next are illustrated sequentially in FIGS. 6A-6C. Information travels in response to the application of appropriate voltages to each stage of nanomagnets 102. Each nanomagnet 102 is in the “send” mode when V_bias is applied to it. Each nanomagnet 102 is in the “receive” mode when V_supply is applied to it. Finally, each nanomagnet 102 is in an “idle” mode if it is grounded. The “idle” mode serves to prevent a nanomagnet 102 from interfering with the writing process on other nanomagnets 102 with which it shares a spin-coherent channel 104. Except for the fixed inputs 604, which don't use the “receive” mode, all nanomagnets 102 go through the three stages of receive, send, and idle in a cyclic fashion, as is illustrated in FIGS. 6A-6C. If advantageous, even the fixed inputs 604 can be reconfigured to allow toggling between AND and OR, or between NAND and NOR. Where reconfiguration is not a desired feature, it is not necessary to have a fixed layer 102B with 90° polarization in the nanomagnets 102 serving as the fixed inputs 604.

Referring now to FIG. 7A, another illustrative embodiment of an ASLD 700 is shown. The ASLD 700 of FIG. 7A has a similar configuration to the ASLD 100 of FIG. 1, except that the spin-coherent channel 104 includes a ground terminal 702 that is positioned closer to the first nanomagnet 102 than to the second nanomagnet 102. As described further below, this positioning of the ground terminal 702 causes the first nanomagnet 102 to act as an input and the second nanomagnet 102 to act as an output. As with the ASLD 100, the first and second nanomagnets 102 of the ASLD 700 may have identical switching characteristics in some embodiments. As shown in FIG. 7A, the first and second nanomagnets 102 are each electrically coupled to a supply voltage, V_ss. In the illustrative embodiment, the supply voltage is unclocked, i.e., constantly applied during operation. It will be appreciated that, in some embodiments, the supply voltage may be intermittently turned off to reduce power consumption of the ASLD 700. Even when the supply voltage is removed, however, the nanomagnets 102 will retain their states, acting as a non-volatile memory.

As described above with reference to FIG. 1, each nanomagnet 102 may have an input side 110 (i.e., a “talking” side) that transmits information and an output side 112 (i.e., a “listening” side) that receives information. An isolation layer 108 may separate two spin-coherent channels 104 below the nanomagnet 102 (one spin-coherent channel 104 being associated with the input side 110 and another spin-coherent channel 104 being associated with the output side 112). Therefore, each nanomagnet 102 can interact independently with different nanomagnets 102 on either side, allowing for a number of cascaded ASLD 700.

FIG. 7B illustrates a state diagram for two nanomagnets 102 connected by a spin-coherent channel 104. When the supply voltage is absent, the two nanomagnets 102 in FIG. 7A can exist in any of four possible states, denoted as 00, 01, 10, 11 (the input nanomagnet 102 being the first bit and the output nanomagnet 102 being the second bit). As the supply voltage is increased beyond a positive threshold value, however, the only stable states are those for which the two nanomagnets 102 are anti-parallel, namely 01 and 10. On the other hand, if the supply voltage is negative beyond a certain threshold value, the only stable states are those for which the two nanomagnets 102 are parallel, namely 00 and 11. These stable states are the ones shown in FIG. 7B with arrows pointing back to themselves. The operation with a positive supply voltage can be viewed as a NOT operation, while the operation with a negative supply voltage can be viewed as a COPY operation. It should also be appreciated that the transition arrows in the table are always “vertical” and never “horizontal.” Due to the non-reciprocity of the ASLD 700, it is always the output nanomagnet 102 that changes its magnetization appropriately to reach a stable state, never the input nanomagnet 102. In other words, the two nanomagnets 102 of the ASLD 700, even when identical structures, function as the input and the output respectively, with an input-output isolation that is characteristic of transistors.

In the ASLD 700, the asymmetric positioning of the ground terminal 702 results in the first nanomagnet 102 effectively shielding the second nanomagnet 102 from communicating with the ground terminal 702, thereby reducing the charge current and consequently the spin current ({right arrow over (I)}_S) injected by the second nanomagnet 102 as compared to the first nanomagnet 102. As a result the torque exerted on the second nanomagnet 102, given by {right arrow over (I)}_S={circumflex over (m)}₂×({right arrow over (I)}_S2×{circumflex over (m)}₂), is greater than the torque exerted on the first nanomagnet 102. This effect is captured by the distributed conductance network model illustrated in FIG. 7D, which is based on the coupled spin-transport/magneto-dynamics model shown in FIG. 7C. As shown in FIG. 7D, the ASLD 700 may be modeled as a number of conductance elements (each comprising a 4 by 4 tensor), which accounts for the spatial variation of quasi-Fermi levels in the spin-coherent channel 104, for charge as well as each spin component.

As one illustrative example, FIGS. 7E and 7F show the numerical results for the specific case when the first nanomagnet 102 is along z and the second nanomagnet 102 is along x. FIG. 7E shows how a higher injection by the first nanomagnet 102 results in a larger z spin voltage distribution in the vicinity of the second nanomagnet 102 compared to the x spin voltage in the vicinity of the first nanomagnet 102. The ability of the input nanomagnet 102 to shield the output nanomagnet 102 from the ground terminal depends on the ratio of the magnet contact length (L_a) to the transfer length of the contact which is of the order: L_rg=1/√{square root over ((gρ)/(WA))}{square root over ((gρ)/(WA))}, where g is the contact conductance, ρ is the channel resistivity, W is the width of the channel, and A is the cross-sectional area of the channel. The ratio of spin-torques, I_S2^⊥/I_S1^⊥, follows from V_Z/V_X, as can be seen by comparing FIGS. 7E and 7F. FIG. 7E was simulated for the particular case of L_c˜L_rg, and the corresponding point can be identified in FIG. 7F.

FIGS. 8A-C illustrate a typical simulation result for the ASLD 700 using the coupled spin-transport/magneto-dynamics model of FIG. 7C and the dynamics of magnetization ({circumflex over (m)}) described by the LLG equation:

$\begin{array}{cc}\frac{d\hat{m}}{dt}=-\gamma \hat{m}\times {\overrightarrow{H}}_{\mathrm{int}}+\alpha \hat{m}\times \frac{d\hat{m}}{dt}+\frac{I_s^{\perp }}{{\mathrm{qN}}_s}.& \left(5\right)\end{array}$ Each nanomagnet 102 is described by a separate LLG equation. N_s is the net number of Bohr magnetons in one of the nanomagnets 102, given by N_s=M_sΩ/μ_B (where M_s is the saturation magnetization, Ω is the volume, and μ_B is the Bohr magneton). The solid curves in FIGS. 8B and 8C represent the state of the output nanomagnet 102, and the dashed curves represent the state of the input nanomagnet 102 as a function of the dimensionless time τ in units of qN_s/I_sc. I_sc is the critical spin current needed for switching, given by:

$\begin{array}{cc}{I}_{\mathrm{sc}}=\frac{2q}{\hslash }\left(2\alpha E_b\right)\left(1+\frac{H_d}{2H_K}\right),& \left(6\right)\end{array}$ where ℏ is the reduced Plank's constant, E_b is the anisotropy energy barrier, H_a is the demagnetizing field, and H_K is the uniaxial anisotropy field.

As shown in FIG. 8A, the supply voltage applied to the ASLD 700 begins at a negative value, switches to positive value, and returns to a negative value. In FIG. 8B, the ASLD 700 was initialized in the parallel state 00. With a negative initial supply voltage, the parallel states are stable and the state 00 is retained. When the supply voltage is made positive, the output nanomagnet 102 becomes the NOT of input nanomagnet 102 and the anti-parallel state 01 is reached (which is now stable). The ASLD 700 again reaches the parallel state 00 when the supply voltage is changed back to a negative value. Similarly, in FIG. 8C, the ASLD 700 was initialized in the parallel state 11. With a negative initial supply voltage, the parallel states are stable and the state 11 is retained. When the supply voltage is made positive, the output nanomagnet 102 becomes the NOT of input nanomagnet 102 and the anti-parallel state 10 is reached (which is now stable). The ASLD 700 again reaches the parallel state 11 when the supply voltage is changed back to a negative value.

Even if both nanomagnets 102 of the ASLD 700 initially start to switch after one of the transitions of the supply voltage in FIG. 8A, the output nanomagnet 102 gets going faster and soon the input nanomagnet 102 goes back to its initial state, since the overall objective of both nanomagnets 102 is the same: an anti-parallel configuration for positive supply voltages, or a parallel configuration for negative supply voltages. As the input nanomagnet 102 is closer to the ground terminal 702, it is more efficient in generating the spin currents that cause the output nanomagnet 102 to switch states. As illustrated in FIG. 9, the critical voltage (V_c1) required to switch the input nanomagnet 102 is greater in magnitude than the critical voltage (V_c2) required to switch the output nanomagnet 102. The difference between these critical voltages arises from the asymmetry in the ASLD 700 (i.e., for a symmetrical ASLD, V_c1=V_c2). In other words, the asymmetry of the ASLD 700 dictates that a higher voltage is required to generate enough spin current to switch the input nanomagnet 102. The difference between the critical voltages (V_c1−V_c2) is a direct measure of the non-reciprocity in the system.

As shown in FIG. 9, V_c1 and V_c2 delineate the range of supply voltages into three distinct regimes. When V_SS<V_c2, neither nanomagnet 102 has enough spin current to switch, and all four states (i.e., 00, 10, 11, 01) are stable. When V_c2<V_SS<V_c1, only the output nanomagnet 102 has enough spin current to switch, and the nanomagnets 102 stabilize in either parallel or anti-parallel states. When V_SS>V_c1, there is sufficient torque acting on both nanomagnets 102 to switch either of them, but only the output nanomagnet 102 switches. The induced asymmetry ensures that the torque on the output nanomagnet 102 is greater than the torque on the input nanomagnet 102. Additional schemes for creating this asymmetry in other embodiments of the ASLD are discussed below.

Referring now to FIG. 10A, a number of ASLD 700 may be cascaded to construct digital circuits. In the illustrative embodiment of FIG. 10A, first and second nanomagnets 102 are linked by a first spin-coherent channel 104, while second and third nanomagnets 102 are linked by a second spin-coherent channel 104. As described above, isolation layers 108 prevent cross-talk between the spin-coherent channels 104. When a positive supply voltage is applied to the circuit of FIG. 10A, each nanomagnet 102 tries to invert the other nanomagnets 102 connected to it. However, the positioning of the ground terminal 702 of each spin-coherent channel 104 enforces a directed transfer of information (i.e., magnetization) in a specific order: 1→2→3. Consequently, the circuit shown in FIG. 10A behaves as a chain of inverters. This is illustrated in FIG. 10B, which shows the states of the first, second, and third nanomagnets 102 over time in individual plots. Initially, both the first nanomagnet 102 and the second nanomagnet 102 start to switch. However, the first nanomagnet 102 dominates, and the second nanomagnet 102 gets inverted from logical “0” to a logical “1.” Subsequently, the second nanomagnet 102 inverts the third nanomagnet 102, and the system becomes stable. It should be appreciated that the first, second, and third nanomagnets may have identical switching characteristics, in some embodiments, and that no clocking is needed for the circuit of FIG. 10A.

Another illustrative embodiment of a digital circuit including a number of cascaded ASLD 700 is illustrated in FIG. 11A. The circuit of FIG. 11A is similar to the circuit of FIG. 10A, except that the third and first nanomagnets 102 are also linked by a third spin-coherent channel 104, forming a ring oscillator. When a supply voltage is applied to the ring oscillator, each nanomagnet 102 will invert the state of the next nanomagnet 102 in a cyclic fashion. This ring oscillator allows sufficient feedback from the third nanomagnet 102 to drive a first nanomagnet 102 having identical switching characteristics, without any clocking circuitry. The switching edges of the first, second, and third nanomagnets 102 are identified by vertical dashed lines in FIG. 11B and show the step by step process of information transfer from one nanomagnet 102 to the next. It is contemplated that the ASLD 700 may be used to construct additional digital circuits (e.g., a set of Boolean logic gates).

In addition to the ground terminal 702 of the ASLD 700 described above, other asymmetries may be used to provide non-reciprocity to an ASLD. The degree of non-reciprocity of an ASLD can be arrived at by defining a “spin-torque conductance” (g_s) relating the spin-torque component of the current at each nanomagnet in the ASLD to the supply voltage:|{right arrow over (I)}_s1^⊥|={circumflex over (m)}₁×({right arrow over (I)}_s1×{circumflex over (m)}₁)|=g_s1V_SS  (7),|{right arrow over (I)}_s2^⊥|=|{circumflex over (m)}₂×({right arrow over (I)}_s2×{circumflex over (m)}₂)|=g_s2V_SS  (8).Of the two nanomagnets in an ASLD, the nanomagnet with the greater spin-torque conductance (g_s) will function as the output.

FIG. 12 illustrates a simplified conductance model that may be used to analyze an ASLD structure. This model allows us to obtain an exact analytical expression that matches the distributed numerical model in FIG. 7D for near ballistic channels. Each of the conductances in this model will be a 4×4 matrix that relates 4×1 voltages and currents that include charge and the three spin components, i.e. [c,z,x,y]^T. For a ferromagnet in the ‘z’ direction, the interface conductances are given by G_F and G_0F:

$\begin{array}{cc}{G}_F=\left(\begin{array}{cccc}{g}_F& g_{\alpha }& 0& 0\\ g_{\alpha }& g_F& 0& 0\\ 0& 0& 0& 0\\ 0& 0& 0& 0\end{array}\right),G_{0F}=\left(\begin{array}{cccc}0& 0& 0& 0\\ 0& 0& 0& 0\\ 0& 0& g_{\beta }& -g_{\gamma }\\ 0& 0& g_{\gamma }& g_{\beta }\end{array}\right),& \left(9\right)\end{array}$ where g_F is the conductance of the F_M/interface region, g_α=Pg_F, and P is the effective polarization of the FM interface. g_β and g_γ refer to the effective spin “mixing” conductance of the interface and describe the Slonczewski and field-like components of spin torque, respectively. g_γ may be set to zero because the field-like term is generally very small in all-metallic structures. The lead to the ground terminal may be assumed to be unpolarized with the absence of any spin-orbit interaction effects (thereby, equally affecting all spin components) and can be described by a matrix:

$\begin{array}{cc}{G}_N=\left(\begin{array}{cccc}{g}_0& 0& 0& 0\\ 0& g_{0P}& 0& 0\\ 0& 0& g_{0P}& 0\\ 0& 0& 0& g_{0P}\end{array}\right),\mathrm{where}{g}_0=\frac{A}{\rho L}\mathrm{and}{g}_{0P}=\frac{A}{\rho L}\coth \left(\frac{L}{\lambda }\right).& \left(10\right)\end{array}$ ρ, λ, L and A refer to the resistivity, spin-diffusion length, length and cross section of the lead, respectively. Solving the conductance model of FIG. 12 provides an expression for the degree of non-reciprocity:

$\begin{array}{cc}\frac{g_{s2}}{g_{s1}}\equiv \frac{I_{s2}^{\perp }}{I_{s1}^{\perp }}=\left(\frac{g_{\beta 2}}{g_{\beta 1}}\right)\frac{\begin{array}{c}\left(2g_{\alpha 2}{\cos }_{\frac{\theta }{2}}^2\right)\left(g_{\beta 1}-g_{F1}\right)+\\ g_{\alpha 1}\left(g_{0P}+g_{F2}\right)+g_{\alpha 2}{g}_{F1}+\left(g_{\alpha 1}-g_{\alpha 2}\right)g_{\beta 1}\end{array}}{\begin{array}{c}\left(2g_{\alpha 2}{\cos }_{\frac{\theta }{2}}^2\right)\left(g_{\beta 2}-g_{F2}\right)+\\ g_{\alpha 2}\left(g_{0P}+g_{F1}\right)+g_{\alpha 1}{g}_{F2}+\left(g_{\alpha 2}-g_{\alpha 1}\right)g_{\beta 2}\end{array}}.& \left(11\right)\end{array}$ For two perpendicular magnets (θ=π/2), the expression simplifies to:

$\begin{array}{cc}\frac{g_{s2}}{g_{s1}}=\frac{g_{\alpha 1}{g}_{\beta 2}\left(g_{\beta 1}+g_{0P}+g_{F2}\right)}{g_{\alpha 2}{g}_{\beta 1}\left(g_{\beta 2}+g_{0P}+g_{F1}\right)}.& \left(12\right)\end{array}$

Equation (12) may used to evaluate several possibilities for implementing non-reciprocity in an ASLD. For instance, the non-reciprocity of an ASLD depends on how well the input nanomagnet can inject polarized spin current (g_α1) and how easily the output nanomagnet can relax the non-collinear spins (g_β2), and vice-versa. As described above, the insertion of a tunnel barrier 114 at an interface between the spin-coherent channel 104 and one of the nanomagnets 102 may introduce or improve non-reciprocity. In other embodiments, the input and output nanomagnets 102 may be designed with different interface areas to the spin-coherent channel 104. It is contemplated that any suitable technique for designing different spin-torque conductances at the input and output nanomagnets 102 may be used to introduce or to improve non-reciprocity.

Referring now to FIG. 13A, it is also possible to introduce non-reciprocity in a physically symmetric ASLD 1300 by applying different voltages to the two nanomagnets 102. In the illustrative embodiment of FIG. 13A, the input nanomagnet 102 of the ASLD 1300 is electrically coupled to a supply voltage, while the output nanomagnet 102 is grounded. The non-reciprocity in this case is given by:

$\begin{array}{cc}\frac{g_{S2}}{g_{S1}}=1+\frac{\frac{g_0}{g_F}}{1-\frac{2g_{\alpha }^2{\cos }_{\frac{\theta }{2}}^2}{g_F\left(g_F+g_{0P}+g_{\beta }\right)}}.& \left(13\right)\end{array}$ In general, the nanomagnet 102 with a voltage closer to that of the ground terminal 702 ends up with a higher spin-torque conductance (g_s). The ASLD 1300, however, has no stable state and once the state of the output nanomagnet 102 has switched, the system continues to oscillate deterministically between all the possible states (i.e., 00, 01, 11, 10), as illustrated in FIG. 13B. These oscillations are primarily due to the fact that the nanomagnets 102 try to enforce opposite configurations as one of them is injecting spins while the other is extracting spins. In other embodiments of the ASLD 1300, the output nanomagnet 102 may be electrically coupled to a floating voltage.

There are several factors which determine the scalability of circuits employing the ASLD. The most basic requirement for scalability is small nanomagnets. One limit on the size of each nanomagnet is its thermal stability. The retention time of a magnet is given by:

$\begin{array}{cc}{\tau }_r=\frac{e^{\frac{K_{u}V}{k_{B}T}}}{f_0}.& \left(14\right)\end{array}$ f₀ is called the attempt frequency and, for magnetic thin films used for storage purposes, is on the order of 1 GHz. K_uV (V is the volume) is the height of the energy barrier separating the stable states of a magnet and should be at least 40 k_BT (where T is the room temperature) to give about 10 years of retention time. Many experiments have reported K_u values as high as 10⁷ erg/cm³≈¼ kT/nm³. Based on the aforementioned f₀ value, magnets with volumes on the order of 100 nm³ are stable at room temperature, which translates to only a few nm in each dimension and indicates the potential for an extremely small footprint.

Power dissipation is an additional factor that should be taken into account for extremely high density large scale computational circuits and is believed to be the main road block for further downsizing of CMOS technology. The intrinsic switching energy (i.e. the energy dissipated throughout switching) of a magnet is roughly on the order of the barrier height, K_uV, which is at least 40 k_BT<1 aJ per magnet, based on the discussion above. While the magnet may be composed of millions of spins, the dissipation for switching the magnet as one giant collective entity is only a few k_BT. In a charge-based transistor, every single electronic charge dissipates a few k_BT throughout switching. In general, switching energy and energy-delay can be written as:

$\begin{array}{cc}{E}_{\mathrm{sw}}={\mathrm{VQ}}_{\mathrm{tot}},E_{\mathrm{sw}}{t}_{\mathrm{sw}}=\frac{V}{I}{Q}_{\mathrm{tot}}^2,& \left(15\right)\end{array}$ where V and I are the charge voltage and current, respectively, and t_sw is the switching delay. Q_tot=It_sw is the total charge involved in a switching event. Equation (15) permits a simple comparison with charge-based devices, where Q_tot is the amount of charge being switched.

The Q_tot involved in switching an ASLD may be analyzed using the coupled spin-transport/magneto-dynamics model of FIG. 7C. Using this model, the logical switching of two output nanomagnets was plotted in FIG. 14. As shown in FIG. 14, for the same current, I, the output nanomagnet having a lower number of Bohr magnetons (N_s) and a higher internal uniaxial anisotropy (H_k) switched faster. The two output nanomagnets used in the simulation were otherwise identical. While the charge and spin currents, I_c and I_s, continue to flow so long as a supply voltage is present, the spin-torque current I_st that enters the LLG equation is time-limited: it flows only during the time that the nanomagnet is switching and follows the relationship:

$\begin{array}{cc}{\int }_0^{\infty }{I}_{\mathrm{st}}^{z}dt=f_1\left(2{\mathrm{qN}}_s\right).& \left(16\right)\end{array}$ The factor f₁ is exactly 1 if only a uniaxial field is present. However, f₁ can be less or more than 1 when fields other than uniaxial are involved. The total charge Q_tot will be larger than f₁(2qN_s) and can be written as:

$\begin{array}{cc}{Q}_{\mathrm{tot}}={\int }_0^{t_{\mathrm{sw}}}Idt=\frac{I}{{\overrightarrow{I}}_s}{f}_2{f}_1\left(2{\mathrm{qN}}_s\right).& \left(17\right)\end{array}$ where I is the charge current, I_s is the time-average spin current, and f₂=∫I_sdt/∫I_st^zdt is a factor reflecting the fact that the spin current is larger than the spin-torque current that enters the LLG equation.

Thus, the Q_tot and the switching energy-delay of an ASLD may be improved by lowering the number of Bohr magnetons (N_s) of the nanomagnets of the ASLD, while maintaining a fixed energy barrier (E_b) of at least 10 kT (i.e., at least ¼ eV at room temperature) to sustain nonvolatility of the nanomagnets. As shown in FIG. 15, this requires materials with increased H_K. In some embodiments, the nanomagnets used in the ASLD may comprise CoPd, CoNi, CoPtCr, Co₃Pt, FePd, FePt, CoPt, MnAl, Fe₁₄Nd₂B, and/or SmCo₅, all which are have relatively high H_K values. Data points 1 and 2 in FIG. 15 represent the two nanomagnets simulated in FIG. 14. It will be appreciated that such fixed energy scaling through increased H_K results in increased switching speed without an increase in current.

While the disclosure has been illustrated and described in detail in the drawings and foregoing description, such an illustration and description is to be considered as exemplary and not restrictive in character, it being understood that only illustrative embodiments have been shown and described and that all changes and modifications consistent with the disclosure and recited claims are desired to be protected.

Claims

1. An all-spin logic device comprising: a first nanomagnet;a second nanomagnet; anda spin-coherent channel extending between the first and second nanomagnets, the spin-coherent channel configured to conduct a spin current from the first nanomagnet to the second nanomagnet to determine a state of the second nanomagnet in response to a state of the first nanomagnet.

2. The all-spin logic device of claim 1, wherein the first and second nanomagnets have identical switching characteristics.

3. The all-spin logic device of claim 1, further comprising a tunnel barrier disposed at an interface between the spin-coherent channel and one of the first and second nanomagnets.

4. The all-spin logic device of claim 1, wherein the second nanomagnet comprises: a free layer having an easy axis; anda fixed layer having an easy axis, the easy axis of the fixed layer being perpendicular to the easy axis of the free layer.

5. The all-spin logic device of claim 1, wherein the first and second nanomagnets are each electrically coupled to one un-clocked supply voltage.

6. The all-spin logic device of claim 5, wherein the second nanomagnet has a greater spin-torque conductance relative to the spin-coherent channel than the first nanomagnet.

7. The all-spin logic device of claim 5, wherein the spin-coherent channel comprises a ground terminal positioned closer to the first nanomagnet than to the second nanomagnet.

8. The all-spin logic device of claim 1, wherein the second nanomagnet is electrically coupled to a floating voltage.

9. The all-spin logic device of claim 1, wherein the second nanomagnet is grounded.

10. The all-spin logic device of claim 1, wherein the first and second nanomagnets each have an energy barrier of at least one-quarter electron-volt and each comprise less than 106 Bohr magnetons.

11. An all-spin logic circuit comprising: a first nanomagnet having an input side and an output side;a second nanomagnet having an input side and an output side;a third nanomagnet having an input side and an output side;a first spin-coherent channel configured to conduct a spin current generated from the input side of the first nanomagnet to the output side of the second nanomagnet; anda second spin-coherent channel configured to conduct a spin current generated from the input side of the second nanomagnet to the output side of the third nanomagnet.

12. The all-spin logic circuit of claim 11, wherein the first, second, and third nanomagnets have identical switching characteristics.

13. The all-spin logic circuit of claim 11, further comprising a third spin-coherent channel configured to conduct a spin current generated from the input side of the third nanomagnet to the output side of the first nanomagnet.

14. The all-spin logic circuit of claim 11, wherein the output sides of the first, second, and third nanomagnets each have a greater spin-torque conductance than the input sides of the first, second, and third nanomagnets.

15. The all-spin logic circuit of claim 11, wherein: the first spin-coherent channel comprises a ground terminal positioned closer to the input side of the first nanomagnet than to the output side of the second nanomagnet; andthe second spin-coherent channel comprises a ground terminal positioned closer to the input side of the second nanomagnet than to the output side of the third nanomagnet.

16. The all-spin logic circuit of claim 11, wherein the first and second spin-coherent channels are separated by an isolation layer.

17. The all-spin logic circuit of claim 11, wherein the first, second, and third nanomagnets are each electrically coupled to one un-clocked supply voltage.

18. A method comprising: applying a voltage to a first nanomagnet to generate a first spin current in response to a magnetization direction of the first nanomagnet; androuting the first spin current along a first spin-coherent channel to a second nanomagnet to determine a magnetization direction of the second nanomagnet.

19. The method of claim 18, wherein the magnetization direction of the second nanomagnet does not determine the magnetization direction of the first nanomagnet.

20. The method of claim 18, further comprising: applying a voltage to a third nanomagnet to generate a second spin current in response to a magnetization direction of the third nanomagnet; androuting the second spin current along the first spin-coherent channel to the second nanomagnet to determine the magnetization direction of the second nanomagnet in response to a superposition of the first and second spin currents.

21. The method of claim 18, further comprising: applying a voltage to the second nanomagnet to generate a second spin current in response to the magnetization direction of the second nanomagnet; androuting the second spin current along a second spin-coherent channel to a third nanomagnet to determine a magnetization direction of the third nanomagnet.

22. The method of claim 21, further comprising: applying a voltage to the third nanomagnet to generate a third spin current in response to the magnetization direction of the third nanomagnet; androuting the third spin current along a third spin-coherent channel to the first nanomagnet to determine the magnetization direction of the first nanomagnet.

23. The method of claim 21, wherein applying a voltage to the first nanomagnet and applying a voltage to the second nanomagnet comprise applying one un-clocked supply voltage to both the first and second nanomagnets.

24. The method of claim 18, further comprising applying a clocked supply voltage to a fixed layer of the second nanomagnet to place a free layer of the second nanomagnet in a neutral state while receiving the first spin current from the first spin-coherent channel.

25. The method of claim 18, further comprising applying a floating voltage to the second nanomagnet while receiving the first spin current from the first spin-coherent channel.

26. The method of claim 18, further comprising grounding the second nanomagnet while receiving the first spin current from the first spin-coherent channel.

27. The method of claim 26, wherein the magnetization direction of the second nanomagnet oscillates with the magnetization direction of the first nanomagnet.