A lire sur: http://www.gizmag.com/physics-cambridge-spintronics-3d-microchips/26056/
By Brian Dodson February 1, 2013
Physicists at the University of Cambridge have used spintronics to move data between layers of a 3D microchip (Image: Shutterstock)
Image Gallery (2 images)
Image Gallery (2 images)
A major obstruction to the development of practical 3D microchips
is moving data and logic signals from one layer of circuitry to
another. This can be done with conventional circuitry, but is quite
cumbersome and generates a good deal of heat inside the 3D circuit.
Physicists at the University of Cambridge have now developed a
spintronic shift register that allows information to be passed between
different layers of a 3D microchip.
Eventually, using current chip architecture, Moore's law will run out of steam, as problems with lithography, materials, and cooling become more difficult to overcome. At some point in time, the development of truly three-dimensional circuitry (probably) grown on silicon will start to look like the simpler direction to follow for the next decade or two. If you can grow 100 layers of circuitry in an interconnected stack, not only have you provided about 15 years progress on Moore's law, but you have also avoided some of the major bottlenecks facing stupendously large scale integration.
One conceptual problem facing research into 3D integrated circuits is that there seems to be no good way to transfer information between levels of circuitry using electronics. Such a chip might end up with most of its circuitry dedicated to passing information back and forth between layers. Accordingly, Professor Russell Cowburn and his group at the University of Cambridge have developed a vertically layered spintronics shift register to move data vertically between circuitry layers.
Spintronics is mostly known to the public as the basis for super-high density hard disk drives, magnetic random access memory, and racetrack memory.
In such applications, an electric current all of whose electrons spins are parallel (known as spin-polarized current) is sent through an ultrathin magnetic layer. The size of the current that passes through the magnetic layer depends on whether the orientation of the magnetic field of the layer is parallel or opposed to the spin direction of the spin-polarized current. Here an electric current is controlled by a magnetic field, and can be used to read out data from, for example, the magnetic coating of a hard disk drive.
To every action there is an equal and opposite reaction. If the interaction between the spin-polarized current and the magnet changes the current, then properly applied spin-polarized current can change the magnetic state of the magnetic layer. If you can change the magnetic state of a material in which data is represented by magnetic domains, memory and logic operations can be carried out on the data.
Professor Cowburn and his team prepared stacks of alternating ultrathin layers of various metals to demonstrate how spintronics might facilitate vertical transport of data and logic states between 3D circuitry layers. There are two types of metallic layers in the stack.
First is a perpendicularly oriented magnetic layer consisting of a cobalt-iron-boron magnetic layer placed between platinum layers. This structure is less than about 2 nm in thickness, or about 6-8 atomic layers, depending on crystal structure. The second type of layer is a ruthenium layer less than a nanometer in thickness.
A very thin sheet of magnetic material (such as the first kind of layer) essentially has two possible states, spins up and spins down – it simply requires too much energy to tilt the spins. It is more favorable for them to orient perpendicular to the plane of the magnetic layer. Another advantage of such ultrathin magnetic films is that domain walls (the region separating a spin-up domain from a spin-down domain) are very narrow – about 6 nm in width – leading to the potential for very small devices.
The ruthenium layer provides an interaction between neighboring magnetic layers that couples their magnetic spins. The strength, and even the direction of that interaction depends on the thickness of the various layers, which can be tailored to produce one of a number of desirable effects.
In practice, an alternating stack of the first and second type of layers is grown, perhaps to include hundreds of layers. The stack is placed in a magnetic field, which alters the energies of the magnetic materials in all the layers. The thicknesses of the magnetic and the coupling layers are varied slightly within the stack so that the location of a bit of data moves up one magnetic layer for every two field reversals. That is, a spin-up domain on magnetic layer 14, after the magnetic field is reversed twice, now appears on magnetic layer 15. This is the basic mode of operation of this lab demonstration.
And a lab demonstration it is. Rapidly flipping magnetic fields is not good for super-large-scale integrated circuitry, electronic or spintronic. The researchers hope to improve this demonstration by moving the magnetically encoded data by flowing spin-polarized currents from layer to layer, thereby causing the domain structure to climb within the stack. This sort of interaction is also likely to result in a bidirectionally moving stack – the lab demonstrator is "hard-wired" so that the information only works upward.
“This is a great example of the power of advanced materials science," says Professor Cowburn. "Traditionally, we would use a series of electronic transistors to move data like this. We’ve been able to achieve the same effect just by combining different basic elements such as cobalt, platinum and ruthenium. This is the 21st century way of building things – harnessing the basic power of elements and materials to give built-in functionality.”
Source: University of Cambridge
Eventually, using current chip architecture, Moore's law will run out of steam, as problems with lithography, materials, and cooling become more difficult to overcome. At some point in time, the development of truly three-dimensional circuitry (probably) grown on silicon will start to look like the simpler direction to follow for the next decade or two. If you can grow 100 layers of circuitry in an interconnected stack, not only have you provided about 15 years progress on Moore's law, but you have also avoided some of the major bottlenecks facing stupendously large scale integration.
One conceptual problem facing research into 3D integrated circuits is that there seems to be no good way to transfer information between levels of circuitry using electronics. Such a chip might end up with most of its circuitry dedicated to passing information back and forth between layers. Accordingly, Professor Russell Cowburn and his group at the University of Cambridge have developed a vertically layered spintronics shift register to move data vertically between circuitry layers.
Spintronics is mostly known to the public as the basis for super-high density hard disk drives, magnetic random access memory, and racetrack memory.
In such applications, an electric current all of whose electrons spins are parallel (known as spin-polarized current) is sent through an ultrathin magnetic layer. The size of the current that passes through the magnetic layer depends on whether the orientation of the magnetic field of the layer is parallel or opposed to the spin direction of the spin-polarized current. Here an electric current is controlled by a magnetic field, and can be used to read out data from, for example, the magnetic coating of a hard disk drive.
To every action there is an equal and opposite reaction. If the interaction between the spin-polarized current and the magnet changes the current, then properly applied spin-polarized current can change the magnetic state of the magnetic layer. If you can change the magnetic state of a material in which data is represented by magnetic domains, memory and logic operations can be carried out on the data.
Professor Cowburn and his team prepared stacks of alternating ultrathin layers of various metals to demonstrate how spintronics might facilitate vertical transport of data and logic states between 3D circuitry layers. There are two types of metallic layers in the stack.
First is a perpendicularly oriented magnetic layer consisting of a cobalt-iron-boron magnetic layer placed between platinum layers. This structure is less than about 2 nm in thickness, or about 6-8 atomic layers, depending on crystal structure. The second type of layer is a ruthenium layer less than a nanometer in thickness.
A very thin sheet of magnetic material (such as the first kind of layer) essentially has two possible states, spins up and spins down – it simply requires too much energy to tilt the spins. It is more favorable for them to orient perpendicular to the plane of the magnetic layer. Another advantage of such ultrathin magnetic films is that domain walls (the region separating a spin-up domain from a spin-down domain) are very narrow – about 6 nm in width – leading to the potential for very small devices.
The ruthenium layer provides an interaction between neighboring magnetic layers that couples their magnetic spins. The strength, and even the direction of that interaction depends on the thickness of the various layers, which can be tailored to produce one of a number of desirable effects.
In practice, an alternating stack of the first and second type of layers is grown, perhaps to include hundreds of layers. The stack is placed in a magnetic field, which alters the energies of the magnetic materials in all the layers. The thicknesses of the magnetic and the coupling layers are varied slightly within the stack so that the location of a bit of data moves up one magnetic layer for every two field reversals. That is, a spin-up domain on magnetic layer 14, after the magnetic field is reversed twice, now appears on magnetic layer 15. This is the basic mode of operation of this lab demonstration.
And a lab demonstration it is. Rapidly flipping magnetic fields is not good for super-large-scale integrated circuitry, electronic or spintronic. The researchers hope to improve this demonstration by moving the magnetically encoded data by flowing spin-polarized currents from layer to layer, thereby causing the domain structure to climb within the stack. This sort of interaction is also likely to result in a bidirectionally moving stack – the lab demonstrator is "hard-wired" so that the information only works upward.
“This is a great example of the power of advanced materials science," says Professor Cowburn. "Traditionally, we would use a series of electronic transistors to move data like this. We’ve been able to achieve the same effect just by combining different basic elements such as cobalt, platinum and ruthenium. This is the 21st century way of building things – harnessing the basic power of elements and materials to give built-in functionality.”
Source: University of Cambridge
Aucun commentaire:
Enregistrer un commentaire