Periodic Reporting for period 1 - CoSpiN (Coherent Spintronic Networks for Neuromorphic Computing)
Período documentado: 2022-05-01 hasta 2024-10-31
Neuromorphic computing uses networks of artificial neurons highly interconnected by artificial synapses to perform vast data processing tasks with unmatched efficiency, as needed, for instance, for pattern recognition or autonomous driving tasks. The synaptic connections play a paramount role to create better hardware realizations of these networks. However, it is very complex to realize large interconnectivity by electronic circuity. COSPIN overcomes this connectivity constraint by using the eigen-excitations of the magnetic system - the spin waves - to connect state-of-the-art artificial neurons based on spintronic auto-oscillators.
COSPIN’S main goal is to create and experimentally validate innovative physical building blocks for a novel nano-scaled, all-spintronic network structure which incorporates all necessary properties for neuromorphic computing including high nonlinearity, interconnectivity and reprogrammability.
By design, COSPIN works at the boundary between oscillator-based computing and wave-based computing. It uses interference, frequency-multiplexing, and time-modulation techniques as well as spin-wave amplification to significantly increase the connectivity between neurons. Reprogramming of the network is implemented by a direct physical link to magnetic memory solutions as well as by reconfiguring spin-wave circuits. By using coherent wave interference and nonlinear wave interaction, COSPIN paves the way for novel coupling phenomena for complex artificial neural networks far beyond the state-of-the-art of current hardware realizations.
Using cutting-edge micromagnetic simulations enhanced by inverse design methods, the artificial networks will be designed and tested prior to their nano-fabrication. Experimental investigations will be mainly carried out using micro-focused Brillouin light scattering. This allows for local investigation of the individual neurons and synapses, and significantly simplifies the interpretation of the network dynamics.
COSPIN’S main goal is to create and experimentally validate innovative physical building blocks for a novel nano-scaled, all-spintronic network structure which incorporates all necessary properties for neuromorphic computing including high nonlinearity, interconnectivity and reprogrammability.
By design, COSPIN works at the boundary between oscillator-based computing and wave-based computing. It uses interference, frequency-multiplexing, and time-modulation techniques as well as spin-wave amplification to significantly increase the connectivity between neurons. Reprogramming of the network is implemented by a direct physical link to magnetic memory solutions as well as by reconfiguring spin-wave circuits. By using coherent wave interference and nonlinear wave interaction, COSPIN paves the way for novel coupling phenomena for complex artificial neural networks far beyond the state-of-the-art of current hardware realizations.
Using cutting-edge micromagnetic simulations enhanced by inverse design methods, the artificial networks will be designed and tested prior to their nano-fabrication. Experimental investigations will be mainly carried out using micro-focused Brillouin light scattering. This allows for local investigation of the individual neurons and synapses, and significantly simplifies the interpretation of the network dynamics.
In the two years of the project, we have advanced in multiple ways towards the goal of a coherent spintronic platform for neuromorphic computing (“CoSpiN-platform”) which unifies magnonic and auto-oscillators components. Three major directions of our work can be distinguished:
1) The development and creation of new highly nonlinear magnonic components for neuromorphic magnonics such as the first magnonic leaky integrate-and-fire Neuron based on the nonlinear folder effect.
2) The design and fabrication of spintronic auto-oscillators compatible with magnonic circuits including a new understanding of magnonic coupling effects taking place in the auto-oscillators themselves.
3) The design of the hybrid system including the fabrication of first prototypes.
In addition, we started to investigate several promising approaches to enhance the CoSpiN platform further, e.g. the coupling of spintronic auto-oscillators via phonons to create long-range linear interactions.
1) The development and creation of new highly nonlinear magnonic components for neuromorphic magnonics such as the first magnonic leaky integrate-and-fire Neuron based on the nonlinear folder effect.
2) The design and fabrication of spintronic auto-oscillators compatible with magnonic circuits including a new understanding of magnonic coupling effects taking place in the auto-oscillators themselves.
3) The design of the hybrid system including the fabrication of first prototypes.
In addition, we started to investigate several promising approaches to enhance the CoSpiN platform further, e.g. the coupling of spintronic auto-oscillators via phonons to create long-range linear interactions.
The experimental generation of the first all-magnonic leaky integrate-and-fire neuron with build-in tunablity was a breakthrough. We could show that this approach not only allows to create the neuron functionality, but also to tune the properties of the neurons dynamically. This allows to create so-called “dynamical neural networks” where not only the weights of the connections can be tuned, but the entire network can be adjusted. As in detail demonstrated by our collaborators, this approach has great potential for advanced functionalities and performance in comparison to traditional ANN since dynamical neurons offer signicant advantages over existing reservoir computer proposals, providing greater versatility and scalability without signicantly increasing computational costs.
The observation of stochastic magnonic processes and the detection of single magnonic events was a clear advance beyond the state-of-the-art which was also unplanned. It can open an entire new field for magnonics (“stochastic magnonics”). This can also be very important for the development of quantum magnonics since measurements of statistics is a key requirement in this field.
The first demonstration of a hybrid system consisting of a spintronic auto-oscillator and a spin-wave delay line was a demanding breakthrough which we achieved. Even though this system is still electrical coupled, it is the first time that two very different spintronic elements working in the GHz range are connected. Thus, this is also an important step for all-spintronic technology approaches outside the field of neuromorphic computing (e.g. tunable RF signal generation on the nanoscale).
As described in detail above, the first identification of magnon instability in a SAO connected to a simultaneous multi-mode emission of RF signals was an important breakthrough for the understanding of SAO dynamics in general since the gained understanding allows to tailor the properties of future complex SAOs. In addition, it opens a novel way to create multi-frequency magnonic networks where information density can be significantly enhanced by the simultaneous use of distinct signal frequency with a well-defined phase relation.
The first demonstration of the stimulated emission of magnons can have important influence on the field of magnonics in general, since the creation of novel, coherent SW sources on chip could have a great impact similar to the realization of on-chip lasers for the field of photonics.
We have created with our partners the first devices made of a fully spin-polarized half metallic Heusler compound where we could show that the necessary L21 ordering is also preserved in the nanostructure, which was a major problem in the past. The L21 order is associated with an ultralow damping which could make these systems comparable to insulation magnonic systems like YIG nanostructure, but with the additional benefit that electron spin-based effects like the Zhang-Li spin transfer torque can be used in the same device. This opens completely new opportunities to construct magnon-spintronic hybrids.
The creation of a SW-emitting SAO driven by the spin Hall effect on an ultralow-damping garnet was a particular breakthrough for the objectives of the CoSpiN project. This approach requires a material with perpendicular magnetic anisotropy and low damping, two properties which are usually not combined. We fund that Ga:YIG is suited for this approach opening the way for a simplified design of highly interconnected SAOs using coherent SWs.
The observation of stochastic magnonic processes and the detection of single magnonic events was a clear advance beyond the state-of-the-art which was also unplanned. It can open an entire new field for magnonics (“stochastic magnonics”). This can also be very important for the development of quantum magnonics since measurements of statistics is a key requirement in this field.
The first demonstration of a hybrid system consisting of a spintronic auto-oscillator and a spin-wave delay line was a demanding breakthrough which we achieved. Even though this system is still electrical coupled, it is the first time that two very different spintronic elements working in the GHz range are connected. Thus, this is also an important step for all-spintronic technology approaches outside the field of neuromorphic computing (e.g. tunable RF signal generation on the nanoscale).
As described in detail above, the first identification of magnon instability in a SAO connected to a simultaneous multi-mode emission of RF signals was an important breakthrough for the understanding of SAO dynamics in general since the gained understanding allows to tailor the properties of future complex SAOs. In addition, it opens a novel way to create multi-frequency magnonic networks where information density can be significantly enhanced by the simultaneous use of distinct signal frequency with a well-defined phase relation.
The first demonstration of the stimulated emission of magnons can have important influence on the field of magnonics in general, since the creation of novel, coherent SW sources on chip could have a great impact similar to the realization of on-chip lasers for the field of photonics.
We have created with our partners the first devices made of a fully spin-polarized half metallic Heusler compound where we could show that the necessary L21 ordering is also preserved in the nanostructure, which was a major problem in the past. The L21 order is associated with an ultralow damping which could make these systems comparable to insulation magnonic systems like YIG nanostructure, but with the additional benefit that electron spin-based effects like the Zhang-Li spin transfer torque can be used in the same device. This opens completely new opportunities to construct magnon-spintronic hybrids.
The creation of a SW-emitting SAO driven by the spin Hall effect on an ultralow-damping garnet was a particular breakthrough for the objectives of the CoSpiN project. This approach requires a material with perpendicular magnetic anisotropy and low damping, two properties which are usually not combined. We fund that Ga:YIG is suited for this approach opening the way for a simplified design of highly interconnected SAOs using coherent SWs.