Magnonic integrated circuits have been demonstrated
A team of researchers has successfully constructed a basic building block of computer circuits using magnons to convey information, in place of electrons. The “magnonic half-adder,” described in Nature Electronics, requires just three nanowires, and far less energy than the latest computer chips.
Published
A disturbance in the local magnetic order of a solid body can propagate across a material just like a wave. This wave is named spin wave and its quanta are known as magnons. Over the last years, physicists are searching for a way to use magnons to carry and process information instead of electrons as it is done in electronics. This technology opens access to a new generation of computers in which data is processed without motion of any real particles like electrons. This leads to a decrease in the accompanying heat loss and, consequently, to lower energy consumption. Moreover, unique linear and nonlinear magnon properties allow for the utilization of alternative computing concepts resulting in a decrease of the footprint of logic elements and in a boost of the performance of modern processors. The field of science addressing these questions is known nowadays as magnonics and we have invested many years in helping to expand it.
In 2014 we opened a new direction of “all-magnon” computing within magnonics by the demonstration of the first magnon transistor [Nat. Commun. 5, 4700, (2014)]. The density of magnons flowing from the transistor's Source to its Drain in this three-terminal device could be decreased one thousand times via the injection of magnons in the Gate. The interaction between magnon flows was very efficient due to the strong natural nonlinearity of magnons and was consequently enhanced by using an artificial magnetic material – the magnonic crystal. The most important achievement in that work was to prove that it is possible to control one magnon in a magnonic circuit by another magnon without any conversion of data into electric signals. Thus, one can potentially build a magnonic circuit consisting of millions of elements in which all information is constantly encoded into the spin-wave phase or amplitude. Nevertheless, at that time, it was just an idea that was far from realization. The prototype transistor had millimeter sizes (while the CMOS technology just went down to 14 nm feature size) and the energy consumption was in the nJ range (while CMOS was already sub-fJ). Thus, the first thing we had to do to prove the feasibility of the “all-magnon” approach was to decrease the sizes of the elements down to the nano-scale and to find another, more energy-efficient, nonlinear mechanism to process data. Fortunately, the ERC funded my Starting Grant “MagnonCircuits” and we got a chance to address this challenge.
Now, after many years of intensive investigations, we can report that the mission is accomplished. This manuscript reports on the realization of a nano-scaled magnonic directional coupler, which is a multi-functional device and can be used as a universal building block of a magnonic circuit. So, from the functional point of view, this directional coupler is much more powerful than the original magnon transistor. The device is based on yttrium iron garnet single-mode waveguides with a width of 350 nm. We use the amplitude of the spin waves to encode information and to guide it to one of the two outputs of the coupler depending on the signal magnitude, frequency and the applied magnetic field. Using micromagnetic simulations, we also proposed an integrated magnonic half-adder that consists of two directional couplers and we investigate its functionality for information processing within the magnon domain. This device is amazing because it consists of 3 magnetic nanowires and fancy spin-wave nonlinear physics only, and replaces 14 CMOS transistors (namely hundreds of elements). The analysis has also shown that the energy consumption of this device lies in the aJ energy range as we expected many years ago. Finally, we performed the benchmarking of the device and compared it with 7 nm CMOS, which just recently appeared on the market. It demonstrates that 30-nm-based magnonic half-adder has the same footprint as 7 nm-based-CMOS half-adder, consumes around ten times less energy, but, unfortunately, is slower.
The operational principle of the half-adder. The information is encoded into spin-wave intensity: red corresponds to logic “1”, blue to logic “0”. When only one of the inputs is „1“, the second directional coupler operates in the linear regime and the wave transmits towards the top output “S”. If both inputs have “1” signal, the directional coupler switches to the nonlinear regime and the output wave reaches the bottom “C” output. Any combination of the input signals satisfies the truth table of a half-adder.
Many talented researchers significantly contributed to the realization of the magnonic directional coupler. But I would specially like to underline the contribution of Dr. Qi Wang (currently with me at the University of Vienna), the lead author of the paper, who has proposed this fantastic half-adder and performed most of the investigations. Moreover, on this way, we were going shoulder-to-shoulder with Jun.-Prof. Philipp Pirro (TU Kaiserslautern) and supervised this research jointly.
The proposed magnonic circuit is dedicated to the processing of binary, so-called digital, information used nowadays in any smartphone or computer. Nevertheless, the potential of the magnonic directional coupler goes far beyond the binary data in the area of more powerful unconventional computing. Philipp will use it to build a neuromorphic magnon computer with the operational principle inspired by the functionality of our brain. I plan to cool the system down to extremely small temperatures to get access to the fascinating quantum properties of magnons. The coupler opens indeed fantastic opportunities.Further information can also be found here: https://www.youtube.com/watch?v=P8HHKDVzIMg
Copyrights of the poster image: Niels Paul Bethe, SYNC audiovisual design.
No comments yet.