🔬 applied physics

From Cantilevers to Membranes: Advanced Scanning Protocols for Magnetic Resonance Force Microscopy

This paper demonstrates through numerical simulations that combining strained SiN resonators with a novel multislice, compressed-sensing scan protocol can accelerate Magnetic Resonance Force Microscopy (MRFM) acquisition times by up to two orders of magnitude while maintaining high reconstruction fidelity for volumetric imaging of biological nanostructures.

Original authors: Nils Prumbaum, Christian L. Degen, Alexander Eichler

Published 2026-04-15

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Nils Prumbaum, Christian L. Degen, Alexander Eichler

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to take a high-resolution 3D photograph of a tiny, invisible speck of dust floating in a room. But there's a catch: you can't use a camera. Instead, you have to use a super-sensitive "feeler" (a tiny mechanical sensor) that vibrates when it touches the magnetic "aura" of the atoms inside that speck. This is the basic idea behind Magnetic Resonance Force Microscopy (MRFM).

For decades, scientists have been trying to use this technique to take pictures of biological structures (like viruses or proteins) at the atomic level. It's like trying to see the individual bricks in a wall from a mile away. The problem? It's incredibly slow, and the pictures are often blurry.

This paper, written by researchers at ETH Zurich, proposes a new way to take these pictures that is 100 times faster and much clearer. They do this by changing two main things: the camera lens and the way they scan the object.

Here is the breakdown of their breakthrough using simple analogies:

1. The Old vs. New "Camera Lens" (The Sensor)

The Old Way (The Cantilever): Imagine a diving board. In traditional MRFM, the sensor is like a diving board that wiggles side-to-side (left and right). When the magnetic "aura" of the sample pushes it, it moves sideways.
- The Problem: It's like trying to feel the shape of a ball by brushing your hand against it from the side. You get a good feel for the edge, but it's hard to tell what's happening deep inside the ball.
The New Way (The Membrane): The researchers switched to a sensor that looks like a trampoline or a drum skin. Instead of wiggling side-to-side, it bounces up and down.
- The Analogy: Imagine pressing your finger straight down onto a drum skin. You can feel the tension and shape of the drum much better than if you were just brushing the side.
- The Result: Because the "trampoline" bounces up and down, it picks up the magnetic signals from the sample more efficiently, especially from parts of the sample that are further away. It creates a sharper "fingerprint" of the object.

2. The Old vs. New "Scanning Strategy" (The Protocol)

The Old Way (The XYZ Scan): Imagine you are trying to map a mountain range by walking in a grid. You walk forward, stop, look, walk forward, stop, look. Then you take one step to the side, walk forward, stop, look. You do this for every single inch of height, width, and depth.
- The Problem: This takes forever. Also, you are using the same "flashlight" (magnetic pulse) for every single spot, which isn't very efficient.
The New Way (The Multislice Scan): Instead of walking in a 3D grid, imagine you stand in one spot and look at the mountain through different colored glasses.
- How it works: You stay in one spot on the "trampoline." You take a picture, then you change your glasses (tune the magnetic pulse to a different frequency), take another picture, and repeat this 46 times. Each "color" of glasses sees a different "slice" of the mountain's interior.
- The Benefit: You get a much richer 3D picture by staying still and changing your "view," rather than moving around blindly. It's like listening to a symphony: instead of walking around the orchestra to hear each instrument, you stand in the middle and listen to the different sections (strings, brass, woodwinds) one by one.

3. The "Compressed Sensing" Trick (The Speed Booster)

Even with the new sensor and strategy, taking a picture of every single pixel is still slow. The researchers added a third trick called Compressed Sensing.

The Analogy: Imagine you are trying to guess the shape of a cloud. Do you need to measure every single water droplet? No. If you know clouds are generally fluffy and continuous, you can measure just half (or even a third) of the droplets and use a smart computer algorithm to "fill in the blanks" for the rest.
The Result: The computer is smart enough to reconstruct the full, high-quality image even if they skip measuring 50% to 80% of the data points. This cuts the time needed in half (or more) without losing detail.

The Grand Finale: What Does This Mean?

By combining the Up-and-Down Trampoline Sensor, the Multislice Scanning Strategy, and the Smart "Fill-in-the-Blanks" Algorithm, the researchers predict they can take a 3D picture of a biological nanostructure (like a virus) 100 times faster than before.

Old Speed: Imagine it takes 100 hours to take one picture.
New Speed: It would take just 1 hour.

Why is this a big deal?
Currently, we can't see the detailed 3D structure of many biological molecules because the process is too slow and the images are too blurry. If this technique works in the real world (as the simulations suggest), it could revolutionize medicine and biology, allowing us to see how viruses and proteins are built at the atomic level, which is the first step toward designing better drugs and cures.

In short: They swapped a wobbly diving board for a bouncy trampoline, stopped walking in circles to start looking through different colored glasses, and taught the computer to guess the missing pieces. The result? A super-fast, super-clear 3D camera for the microscopic world.

1. Problem Statement

Magnetic Resonance Force Microscopy (MRFM) is a technique capable of 3D imaging of nuclear spin densities in nanoscale objects, with potential applications in structural biology (e.g., viruses, proteins). However, despite significant milestones (such as single-electron spin detection and 1D sub-nanometer imaging), MRFM has not yet achieved near-atomic resolution for volumetric biological imaging.

The primary bottlenecks are:

Acquisition Time: Traditional 3D scanning protocols are extremely slow, often requiring hours or days to acquire sufficient signal-to-noise ratio (SNR).
Sensor Limitations: Traditional pendulum-style cantilevers suffer from non-contact friction, static bending, and lower resonance frequencies compared to newer resonator types.
Reconstruction Fidelity: The convolution of the sample with the Point Spread Function (PSF) of the sensor leads to information loss, particularly for internal structures, making image reconstruction ill-posed and sensitive to noise.

2. Methodology

The authors propose a comprehensive framework combining advanced sensor geometries with novel scanning and reconstruction protocols, validated through numerical simulations.

A. Sensor Geometries

The study compares two distinct mechanical resonator setups:

Cantilever (Pendulum-style): Oscillates laterally (x-direction) parallel to the magnetic field gradient source. The driving force is proportional to $\partial B_z / \partial x$ .
Membrane (Strained SiN): Oscillates vertically (z-direction) perpendicular to the gradient source. The driving force is proportional to $\partial B_z / \partial z$ $\partial B_{z} / \partial z$ .
- Key Insight: While the maximum gradient for membranes is lower than for cantilevers, the gradient decays more slowly with distance. This enhances the signal contribution from deeper parts of the sample, improving the effective PSF for volumetric imaging.

B. Scanning Protocols

The paper introduces and evaluates three scanning strategies:

XYZ Scan (Traditional): The gradient source is scanned in a 3D grid ( $x, y, z$ ). At each point, a single RF inversion pulse defines a single resonant slice. This creates a trade-off between scan depth and sensitivity.
Multislice Scan (Proposed): The gradient source scans only in the $(x, y)$ $(x, y)$ plane. At each $(x, y)$ $(x, y)$ position, $n_f$ $n_{f}$ sequential measurements are taken using different RF inversion pulse parameters (center frequency $f_0$ $f_{0}$ and bandwidth $f_{BW}$ $f_{B W}$ ).
- This generates multiple "slices" of data at different depths without physically moving the resonator in $z$ .
- Noise Reduction: By keeping the resonator stationary during multiple measurements, the thermomechanical noise estimation is improved, reducing total measurement uncertainty by up to a factor of 2 compared to the XYZ scan.
Compressed Sensing (CS): To further accelerate acquisition, the authors apply CS to the Multislice protocol. Instead of sampling every voxel, they use a random sub-sampling mask (probability $p < 1$ ). The full image is recovered by solving a convex optimization problem ( $\ell_1$ -norm minimization) that exploits the sparsity of the sample in the Fourier domain.

C. Reconstruction Algorithm

The reconstruction is formulated as an inverse problem: recovering the spin density $O$ from the measured force variance $I$ , where $I = H \star O + \eta$ ( $H$ is the PSF, $\star$ is convolution).

Regularization: To handle the ill-posed nature of the problem and noise, the authors use Total Variation (TV) regularization to preserve edges while smoothing homogeneous regions.
Solver: They employ the Alternating Direction Method of Multipliers (ADMM) to solve the optimization problem efficiently. This outperforms traditional Landweber iteration in both convergence speed and reconstruction accuracy.

3. Key Contributions

Membrane Superiority: Demonstrated via simulation that membrane resonators (oscillating in $z$ ) provide better reconstruction fidelity than cantilevers for volumetric imaging, particularly when combined with the Multislice protocol.
Multislice Protocol: Introduced a scanning method that decouples depth resolution from mechanical scanning, optimizing the trade-off between sensitivity and depth. It reduces measurement uncertainty and allows for better noise characterization.
Compressed Sensing Integration: Showed that sub-sampling the Multislice data by up to 50% (or more) allows for a proportional reduction in acquisition time without degrading image quality, thanks to the denoising properties of $\ell_1$ -regularization.
Algorithmic Optimization: Developed a robust ADMM-based reconstruction pipeline that converges significantly faster than standard iterative methods (Landweber) and handles the specific noise characteristics of MRFM data.

4. Results

Simulations were performed on a spherical sample (40 nm diameter) with non-homogeneous spin density, mimicking biological specimens.

Reconstruction Fidelity (RMSE):
- The Multislice protocol consistently outperformed the traditional XYZ scan, reducing the Root Mean Square Error (RMSE) by 2–5x.
- Membrane sensors combined with Multislice scanning yielded the lowest RMSE, recovering both the outer shape and internal density fluctuations more accurately than cantilevers.
Noise Performance:
- The Multislice protocol reduced the impact of thermomechanical noise. When the force noise density ( $S_{th}$ ) was lowered (simulating advanced resonators), the error plateaued only when limited by intrinsic spin noise, confirming the protocol's efficiency.
Compressed Sensing Impact:
- Sub-sampling with $p=0.5$ (50% of data) while increasing integration time per point resulted in ~20% lower RMSE compared to a fully sampled scan with shorter integration time, despite having the same total acquisition time.
- Significant detail was preserved even at $p=1/3$ , though fidelity degraded at $p=1/6$ .
Speed-Up Projection:
- The authors project a cumulative acceleration factor of 100x compared to traditional 3D MRFM, derived from:
  - 5x from Multislice scanning.
  - 2x from Compressed Sensing.
  - 3x from Membrane resonators (vs. cantilevers).
  - 4x from potential phase-multiplexing techniques.

5. Significance

This work provides a critical roadmap for achieving high-resolution, volumetric MRFM imaging of biological nanostructures. By shifting from mechanical cantilevers to strained silicon nitride membranes and replacing 3D spatial scanning with a frequency-encoded Multislice approach, the authors demonstrate that the acquisition time bottleneck can be overcome by orders of magnitude.

The combination of advanced resonators, optimized scanning protocols, and sparse reconstruction algorithms makes it feasible to image complex biological objects (like viruses or proteins) at near-atomic scales, a goal that has remained elusive for decades. The proposed methods are directly applicable to current and next-generation MRFM experiments, potentially enabling the structural determination of single biological macromolecules without crystallization.