(Nanowerk Spotlight) As our digital devices handle increasingly complex computations, scientists have looked to physics for inspiration on new computing paradigms. Rather than shuttling electrical signals across silicon like conventional processors, an intriguing approach encodes information in electromagnetic or acoustic waves propagating through space. These wave-based computers can solve problems at incredible speeds — theoretically up to the speed of light itself.
Such extreme velocity results because data processing happens intrinsically within the wave medium through deliberate interference patterns. There are no cascaded logic gates dragging down performance as in digital circuits. Waveforms contain both amplitude and phase data, enriching the information capacity compared to simple on-off binary digits. And without needing repeated analog-to-digital conversions, wave computers avoid a major bottleneck curtailing the evolution of traditional computing architectures.
Experts have sought to build practical wave-based systems for years, but inherent challenges around complexity, customizability and manufacturability have inhibited progress. Realizing practical wave-based computers has proven enormously difficult. Prior concepts demanded complex optimization algorithms yielding unmanufacturable designs. Alternate approaches based on configurable circuit arrays needed impractical numbers of phase shifters and amplifiers. All these setups were application-specific, lacking flexibility for broad problem solving.
Now, scientists from Southeast University in China have achieved a major breakthrough using metasurfaces – a key photonic technology promising revolutionary control over electromagnetic waves. Their pioneering metasurface computer effectively performs rapid analog matrix calculations that would cripple much faster digital supercomputers.
Schematic diagram of the metasurface-based CME solver, which is composed of a 2N-port transmission network and N identical 4-port couplers. (Reprinted with permission by Wiley-VCH Verlag) (click on image to enlarge)
Matrix equations feature prominently across science and engineering, modeling everything from machine learning optimizers to structural mechanics simulations. Solving them digitally demands substantial computing power, motivating the intense interest in analog wave-based architectures.
The crux of the new metasurface solver lies in its computational electromagnetic surface, comprising a grid of 1,176 delicately tuned elements that modify incident waves’ amplitude and phase. This reprogrammable nanophotonic medium actively transforms input signals into desired output data, physically embedding the math inside the metasurface.
To operate the solver, complex matrix equations get converted into two components – a coefficient matrix and a constant vector. These data impress onto electromagnetic waves entering two input ports. As the signals propagate through the metasurface region, they undergo intricate interference computations via scattered reflections. The final result emerges at the output ports, encoded onto outgoing waves.
Remarkably, this entire process finishes almost instantaneously as the waves transit the setup near light speed. There are no systematic logic gate delays like in digital processors. The metasurface computer also consumes far less power than silicon equivalents, sharply reducing operating costs.
Crucially, the operational principle allows solving arbitrary complex matrix equations just by varying the metasurface design and input signals. The same hardware platform thus adapts to diverse problems with no fundamental architecture changes. This programmability grants significant versatility lacking in previous wave computers needing bespoke designs even for basic math operations.
Because tuning individual metasurface elements has proven difficult thus far, the current prototype demonstrates a non-reconfigurable solver for fixed equations. However, rapid progress in dynamic metasurface technology points toward fully software-defined metasurface computers that researchers reconfigure on-demand. The paper also notes that operating at higher frequencies would reduce the overall size enabling larger metasurfaces to solve bigger matrix calculations.
Additionally, the scalable planar geometry of metasurfaces makes expanding the architecture more viable than prior 3D metamaterial structures attempted for wave computing. If robustly developed, such rapid reconfigurable metasurface matrix solvers could markedly transform sectors needing heavy numerical analysis from weather prediction to optimization research.
The pioneering metasurface solver establishes a long-awaited bridge between real-time wave computing and practical programmability. While the initial prototype handles a limited matrix size of 5 x 5, more elements could enumerate higher dimensions. In fact, metasurfaces’ scalable planar geometry makes large problem solving viable more straightforwardly than bulky 3D metamaterial structures attempted previously.
The researchers comprehensively validated their design via simulations and measurements, accurately solving several test matrix equations. Across four simulation test cases, the meta-computer yielded solutions with reasonably low error rates averaging 21%. The experiments on a fabricated prototype further verified the architecture’s feasibility for a 3×3 matrix, successfully computing solutions to eight distinct matrix problems. Quantitatively, these measured solutions showed below 25% error on average – on par with initial benchmarks of competing electronic analog computing schemes. The dominant errors stemmed from tolerances in nanofabrication and challenges precisely reading the output data.
Both factors should improve substantially by leveraging state-of-the-art micro-nanofabrication facilities and high-precision metrology equipment. With further refinement, metasurface computers could surpass digital techniques for specialized tasks needing extreme speeds.
With further refinement, metasurface computers could surpass digital processors for specialized tasks needing extreme speeds like radar imaging, scientific modeling, and data analytics. Intriguingly, their high efficiencies may also suit low-power edge computing applications. The breakthrough work lays vital foundations for metasurface computing by tackling previous showstopper bottlenecks in complexity, customizability, and physical realizability. If robustly developed, such rapid analog matrix solvers could markedly transform sectors needing heavy numerical analysis from weather prediction to optimization research.
Get our Nanotechnology Spotlight updates to your inbox!
Thank you!
You have successfully joined our subscriber list.
Become a Spotlight guest author! Join our large and growing group of guest contributors. Have you just published a scientific paper or have other exciting developments to share with the nanotechnology community? Here is how to publish on nanowerk.com.
