Periodic Reporting for period 2 - CHIRON (Spin Wave Computing for Ultimately-Scaled Hybrid Low-Power Electronics)
Reporting period: 2019-05-01 to 2022-04-30
CHIRON’s effort in hybrid CMOS-spin-wave computing fits into a wider research effort in the semiconductor device community to find new ways to extend Moore’s law beyond the physical scaling limitations of CMOS transistors. Some of the approaches target the replacement of (complex) logic gates rather than individual transistors. This follows current design automation methods that use logic gates as building blocks for circuit design. If these logic gates are formed by individual transistors but rely on different principles, for example using different physical entities to encode information such as magnetic degrees of freedom used in spintronics, the logic gates may follow different scaling laws and can potentially be made much smaller than implementations based on transistors. Hence, the long-term goal of CHIRON’s consortium members is to realize spin-wave circuits that are (much) smaller than their CMOS counterparts.
CHIRON has thus set four concrete objectives. The target specifications of objectives 1–3 followed a roadmap with two technology nodes (TN) during the project.
The first objective concerns the magnetoelectric transducers, both for spin-wave generation as well as detection. Both the proof-of-principle of a scalable transducer with an FBAR design as well as the specifications clearly go beyond the state-of-the-art at the start of the project.
The second objective addresses the scalability of the spin-wave majority gates and the inverters. Miniaturized spin-wave circuits (majority gates, inverters) require the usage of spin waves with short wavelengths. High group velocities improve the computing throughput. Low attenuation lengths are of interest for large circuits although circuit area scaling will alleviate the issue.
The third objective combined objectives 1 and 2 to achieve miniaturized majority gates and inverters. This constitutes a proof of concept of a scalable/scaled spin-wave computing technology. The targeted output voltage would be in a range close to being detected by conventional CMOS periphery.
The fourth and final objective intends to fill the gaps in the circuit design methodology with the goal to develop a methodology to design “arbitrary” logic circuits in spin-wave technology. Such concrete circuits can then be used to benchmark the hybrid SW-CMOS technology in terms of area and throughput. When magnetoelectric transducers are included in the benchmarking methodology, also the computing energy can be estimated.
The above discussions show how CHIRON’s objectives aim at extending the state of the art in multiple aspects from device demonstrators to circuit design concepts. An achievement of all objectives would in principle allow CHIRON to fabricate a spin-wave circuit with electrical input and output transducers. Such a circuit was not yet the goal of CHIRON but will be addressed in a follow-up project, as discussed below. Nonetheless, all elemental building blocks for the circuit would have been demonstrated together with a clear methodology how to combine them in circuits. In the following, we will discuss in detail how close we have come to this final project goal, which objectives have been achieved and which issues remain after the end of CHIRON.
CHIRON has thus set four concrete objectives. The target specifications of objectives 1–3 followed a roadmap with two technology nodes (TN) during the project.
The first objective concerns the magnetoelectric transducers, both for spin-wave generation as well as detection. Both the proof-of-principle of a scalable transducer with an FBAR design as well as the specifications clearly go beyond the state-of-the-art at the start of the project.
The second objective addresses the scalability of the spin-wave majority gates and the inverters. Miniaturized spin-wave circuits (majority gates, inverters) require the usage of spin waves with short wavelengths. High group velocities improve the computing throughput. Low attenuation lengths are of interest for large circuits although circuit area scaling will alleviate the issue.
The third objective combined objectives 1 and 2 to achieve miniaturized majority gates and inverters. This constitutes a proof of concept of a scalable/scaled spin-wave computing technology. The targeted output voltage would be in a range close to being detected by conventional CMOS periphery.
The fourth and final objective intends to fill the gaps in the circuit design methodology with the goal to develop a methodology to design “arbitrary” logic circuits in spin-wave technology. Such concrete circuits can then be used to benchmark the hybrid SW-CMOS technology in terms of area and throughput. When magnetoelectric transducers are included in the benchmarking methodology, also the computing energy can be estimated.
The above discussions show how CHIRON’s objectives aim at extending the state of the art in multiple aspects from device demonstrators to circuit design concepts. An achievement of all objectives would in principle allow CHIRON to fabricate a spin-wave circuit with electrical input and output transducers. Such a circuit was not yet the goal of CHIRON but will be addressed in a follow-up project, as discussed below. Nonetheless, all elemental building blocks for the circuit would have been demonstrated together with a clear methodology how to combine them in circuits. In the following, we will discuss in detail how close we have come to this final project goal, which objectives have been achieved and which issues remain after the end of CHIRON.
The main shortcoming of CHIRON is the failure to conduct all-electrical spin-wave transmission experiments with magnetoelectric transducers. While their operation has been successfully demonstrated with Brillouin light scattering spectroscopy, which unambiguously revealed spin-wave emission by the transducer, the electrical detection of the emitted spin waves has failed, despite three different approaches (magnetoelectric transducers, inductive antennas, inverse spin-Hall effect) being employed. We note that both inductive antennas as well as the inverse spin-Hall effect have been successfully used within CHIRON to detect spin waves emitted by and inductive antenna. Moreover, Brillouin light scattering experiments indicate that the intensity of the spin waves emitted by a magnetoelectric transducer is comparable to that of spin waves emitted by an inductive antenna. Therefore, it appears that the lack of signal in the electrical transmission experiments using a magnetoelectric transducer does not originate in insufficient sensitivity or spin wave emission efficiency but may rather stem from the different mechanisms of spin wave generation by magnetoelectric transducers.
This has been partly confirmed by a combination of finite-element and micromagnetic simulations that show the excitation of very complex magnetization dynamics and asymmetric modes that cannot be detected by antennas. To improve the situation, CHIRON has tried to detect spin waves emitted by a magnetoelectric transducer using the inverse spin-Hall effect. However, so far, these experiments have not been successful neither. We note that devices with antenna transducers fabricated on the same chip showed a clear inverse spin-Hall voltage. This is currently not well understood. It should however be mentioned that typical models for the inverse spin-Hall detection typically assume ferromagnetic resonance or first-order spin wave modes and not the complex magnetization dynamics simulated for magnetoelectric transducers. This topic will be continued by CHIRON partners after the end of the project.
This has been partly confirmed by a combination of finite-element and micromagnetic simulations that show the excitation of very complex magnetization dynamics and asymmetric modes that cannot be detected by antennas. To improve the situation, CHIRON has tried to detect spin waves emitted by a magnetoelectric transducer using the inverse spin-Hall effect. However, so far, these experiments have not been successful neither. We note that devices with antenna transducers fabricated on the same chip showed a clear inverse spin-Hall voltage. This is currently not well understood. It should however be mentioned that typical models for the inverse spin-Hall detection typically assume ferromagnetic resonance or first-order spin wave modes and not the complex magnetization dynamics simulated for magnetoelectric transducers. This topic will be continued by CHIRON partners after the end of the project.
Comparing its objectives to the state of the art at the beginning of the project, it is clear that CHIRON’s activities extend the state of the art. Although spin wave computing has received considerable attention in the magnonics community, the topic has nonetheless been driven mainly by CHIRON with only little scientific competition to date. At the beginning of CHIRON, one key researcher at UNIKL, Andrii Chumak, left the group to start his own group at University of Vienna. CHIRON continued however the collaboration with his group on a pro bono basis, leading to many interesting results. It is fair to say that CHIRON with its outside collaborators have been by far the most advanced consortium working on spin wave computing.
During the project, the scientific field of magnetoelectrics has also gained considerably in interest. Most work is however still dedicated to interactions between spin waves and SAWs. The work of CHIRON on scalable magnetoelectric FBAR-like transducers has remained rather unique, possibly due to the processing challenges for nanoscale devices.
During the project, the scientific field of magnetoelectrics has also gained considerably in interest. Most work is however still dedicated to interactions between spin waves and SAWs. The work of CHIRON on scalable magnetoelectric FBAR-like transducers has remained rather unique, possibly due to the processing challenges for nanoscale devices.