Wave Domain Computing with Stacked Intelligent Metasurfaces

You are reading version 1 of this direction, frozen 2026-10-04. The latest version is version 2.

Version 1 of 238 papers69 conceptsfrozen 2026-10-04

As wireless networks transition toward 6G, they face a "power wall." Traditional Massive MIMO systems rely on power-hungry digital baseband processors and complex radio-frequency (RF) chains to perform the matrix mathematics required for beamforming and interference cancellation. As the number of antennas and the bandwidth increase, the energy cost of these digital operations becomes unsustainable.

A new paradigm, pioneered by researchers like Jiancheng An 0001, Marco Di Renzo, and Mérouane Debbah, proposes a radical solution: Wave Domain Computing. Instead of converting electromagnetic (EM) waves into digital bits to perform calculations, we can use the physical propagation of the waves themselves to execute mathematical operations at the speed of light.

From Digital Logic to Physical Diffraction

The core engine of this paradigm is the Stacked Intelligent Metasurface (SIM). While a standard Reconfigurable Intelligent Surface (RIS) uses a single layer of "meta-atoms" to reflect signals, a SIM cascades multiple programmable layers in a 3D structure.

As an EM wave passes through these layers, each meta-atom applies a specific phase shift. According to Rayleigh-Sommerfeld diffraction theory, the wave naturally spreads and interferes with itself as it travels between layers. By precisely tuning the phase shifts of thousands of meta-atoms across multiple layers, the SIM acts as a physical analog computer.

In this architecture, the "calculation" is the propagation itself. By the time the wave exits the final layer, the physical structure has performed the equivalent of a complex matrix transformation—such as precoding for multiuser MIMO—without a single digital floating-point operation.

The Electromagnetic Neural Network

One of the most powerful ways to understand SIM is as an Electromagnetic Neural Network (EMNN). In a digital deep neural network, data passes through layers of neurons where weights are applied. In a SIM, the layers of meta-atoms act as physical "hidden layers," and the programmable phase shifts act as the "weights."

As shown by Yida Zhang and Guojun Huang, these physical parameters can be trained using error backpropagation, just like a standard AI model. This allows the hardware to learn complex tasks:

  • Semantic Encoding: Guojun Huang demonstrated that a SIM can perform image recognition by mapping input data directly to the transmission coefficients of the first layer. The SIM then "classifies" the data by focusing EM energy into specific spatial beams representing different categories.
  • Semantic Alignment: Mario Edoardo Pandolfo and Kyriakos Stylianopoulos used SIMs to align mismatched latent spaces between different AI models directly in the wave domain, enabling heterogeneous devices to communicate without heavy digital translation.

Precision Control in the Near-Field

As 6G moves toward higher frequencies (THz) and larger antenna arrays, communication increasingly happens in the near-field, where waves are spherical rather than flat. Traditional beamforming (pointing in a direction) is no longer sufficient; we need beamfocusing (focusing on a specific 3D point).

Linglong Dai and Xing Jia have shown that the "architectural depth" of a SIM—the number of layers—provides the necessary degrees of freedom to manipulate these complex spherical wavefronts. By stacking layers, the system can physically "null" interference between users who are located in the same direction but at different distances, a feat that is difficult for single-layer surfaces.

Computing Mathematical Transforms at Light Speed

Beyond communication, SIMs can be programmed to perform fundamental mathematical transforms. Yong Liang Guan and Chau Yuen demonstrated a SIM that physically computes a 2D Discrete Fourier Transform (DFT).

In this setup, the SIM is optimized so that incoming waves from different angles are automatically focused onto specific receiver probes. This allows for Direction-of-Arrival (DOA) estimation with near-zero processing delay. The system doesn't "calculate" the angle; the physical geometry of the optimized metasurface ensures that the wave simply lands on the correct sensor based on its incoming frequency and phase.

Why It Works: The Efficiency of Passive Processing

The primary advantage of this direction is the massive reduction in hardware complexity. In a traditional system, every antenna element usually requires a dedicated RF chain (mixers, converters, amplifiers). In a SIM-aided transceiver, as explored by Chao Xu, the "heavy lifting" of spatial multiplexing is done by the passive (or low-power) metasurface layers.

This allows a base station to support many users with only a few active RF chains, drastically lowering the cost and energy consumption of the network. By treating the wireless environment and the hardware as a single, programmable computational medium, Stacked Intelligent Metasurfaces turn the physics of wave propagation into the computer of the 6G era.

Papers in this version

History & citationVersion 2 · 2026-10-04

Current · Permanent link

  1. Version 2 (latest) · 2026-10-04 · 50 papers
  2. Version 1 · 2026-10-04 · 38 papers
Cite this direction

Latest version (a living page):

@misc{lacuna_wave_domain_computing_fb0db678,
  title = {wave-domain computing},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/wave-domain-computing/rdr_fb0db6785da746fa8f9c99cd53e60cec},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-22},
}

This exact version (version 1, frozen text):

@misc{lacuna_wave_domain_computing_with_stacked_intel_fb0db678_v1,
  title = {Wave Domain Computing with Stacked Intelligent Metasurfaces},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/wave-domain-computing-with-stacked-intelligent-metasurfaces/rdr_fb0db6785da746fa8f9c99cd53e60cec/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-22},
}