SPM force feedback

This page is more about force feedback for the human operator in order to develop an intuition for nanoscale forces.
This is not so much about force feedback for the automatic control loop.
The Δf signal is usually used directly used for that in "constant Δf mode" (wiki-TODO: (double check)).
Constant height mode is commonly used for nc-AFM.

Why not just feed forces to machine learning AI?

Q: Why not hook it up to machine learning AI? Doesn't that make force feedback for human operators obsolete?
No. But hooking up AI can (and should) be done too. This does not make force feedback for human operators obsolete.
There is still the need for communicating the results of the AI (in terms of it's understanding) to human operators and
vice versa communicating guiding envelopes for the AI's further desired actions.
Communicating forces directly as what they are, forces,
rather than via some other indirect "visualization/audioialization/textualitation"
should very much be the most efficient (and thus most enjoyable) communication channel.

nc-AFM – perhaps suprising facts

nc-AFM (atoic force microscopy) does not measure forces but stiffnesses instead

A historic naming accitent. Just like nc (non contact) is one too.
The nc-AFM technique is actually measuring surface stiffness rather than surface forces.
So it is actually stiffness feedback rather than force feedback.

Forces can be reconstructed. Since the delta_f signal is averaging over the oscillation amplitudes range
it is not a simple integration but rather some more complex method needs to be used.

• Sader-Jarvis method or
• Giessibl matrix method.

In the small amplitude limit the math should become equivalent with a simple integration again fro any used method. (TODO: Sow these limits for the two methods eventually. as this may help in intuition.)

nc-AFM gives no force images, all Δf images (roughly stiffness images)

Getting the force from the Δf signal is in essence in core a hidden integration.
This means that for every force data-point one needs an entire spectra of Δf values across different heights (~30 points).
Given that (constant height) nc-AFM is rather slow to begin with in comparison to STM,
this makes nc-AFM way too slow to be usable for making true images of forces (wiki-TODO: check if this really has never been done).
Rather reconstruction of forces is only suitable for single point spectra.

nc-AFM Δf stiffness image brightness is not necessarily unique for height above surface

Unlike with STM where current is monotonously growing with shrinking diastance,
in case of nc-AFM the signal is not monotonously growing.
Looking at the typical curves (Morse potential, Lennard Jones potential)
one finds that for all but the rather deep repulsive regime there are:

• two distances of same energy,
• two distances of same attractive force in first derivative, and
• two distances of same stiffness in second derivative

With every derivative the inner (or nearer to the surface) distance moves farther out away from the surface.

Since nc-AFM effectively measures stiffnesses rather than forces (second derivative of energy)
the second/inner branch of the curve is especially far out and possible accessibly for probing. (TODO: To analyze.).
Thus when probing a bit deeper fro imaging then
one might get images that contain areas of the same brightness but of different branches.
Something to be aware and maybe worry about in the case of probing deeper.
Point is avoiding accidental misinterpretation of images.

The precence of the two branches should not affect the force determination methods
as they are basically an integration not caring about the branch. (wiki-TODO: To confirm.)

The whole ting applies for fores too. But …

• In case of foce spectra curves this is glaringly obvious.
• As a lower derivative this does not push the inner branch as far out.
  The inner branch is at the inflection point of the energy which may be at a depth avoidably foe many mechanosynthesic reactions even.

Existing attempts & lack of (linear) correspondence to real forces

See the paper in external links.
The issue there was that in none of the cases with nc-AFM
there was just a linearly scaled mapping of the real forces.
Not even a mapping of the much faster to attain frequency shifts.
(that are roughly proportional to the local surface siffness).

Tip control stability and tip crash safety one reason given reason (wiki-TODO: (recheck)).
Thee should be means to prevent crashes while still leaving a slinearly scaled mapping though.

Major reason for their nonlinear mappings choices may have been their
understandable focus on scanning in the far very weak interaction range.
Rather than a focus on the more close and more tip crash risky interaction range
where the exponential nature tapered off (nearing the inflection point of growth of attractive forces).

Existing 3D force feedback hardware & (un)suitability

VR force feedback is mostly unrelated

Note that this is quite unrelated and orthogonal to VR.

• Force feedback there amouts to mere vibrations and not really forces.
  Granted vibrations may be adjustable in intensity and thus allow a very low resolution analog to forces of one sign and one direction.
  One direction is nc-AFM limit anyway but tther restrictions are way too severe-
• There is no need for a 3D VR environment to benefit from force feedback. In how far a 3D VR environment my be helpful or not si an entirely different topic.

3D pen force feedback devices

(wiki-TODO: Add image of the device for illustration.)

Cheap 3D force feedback devices originally made for gaming (there was one called "Novint Falcon") have way too low resolution to be useful.
(wiki-TODO: Find the article again that made that very clear. Hard.)

Expensive 3D force feedback devices as of 2025 (3k$ to 4k$) seem to be produced by only one company.
"Touch" from 3D Systems. Formerly "Geomagic Touch", "SensAble Phantom Desktop", "SensAble Omni", and maybe other names too.
These seem to have sufficient force resolution to be useful.

DIY force feedback devices

DIY build force feedback devices may be an option too.
Unclear how difficult to built to sufficient specs.
Serial robot arm designs like the afore mentioned "Touch" seem difficult. Other geometries likely easier.
Coupling to a regular computer mouse may be an interesting option. "2D matching a vertical cross section. See: SPM user interfaces
The nc-AFM Force feedback is 1D limited anyway.

Maybe 3D printer standard stepper motors spring scales and a few strings suffice at the hardware side.
Perhaps useful components:

• Polulu – Tic Stepper Motor Controllers for minimizing electroncis hardware
• Amazon – QWORK® Federwaage, Set mit 5 transparenten Federwaagen zum Aufhängen (1N, 5N, 10N, 20N, 30N) (upper end?)
• some thin braided fishing line - no stretch stiffness is key

There are some existing projects on the web (2025) but none seem suitable ATM.

Related

• Scanning probe microscopy
• SPM software
• SPM user interfaces

External links

• (open access) 2023 – Haptic sensation-based scanning probe microscopy: Exploring perceived forces for optimal intuition-driven control
  Freeman M, Applestone R, Behn W, Brar V. Haptic sensation-based scanning probe microscopy: Exploring perceived forces for optimal intuition-driven control. Ultramicroscopy. 2024 Jan;255:113856. doi: 10.1016/j.ultramic.2023.113856. Epub 2023 Sep 22. PMID: 37783064.
• 3D Systems Touch haptic device
  https://www.3dsystems.com/haptics-devices/touch