Designing a Nanotube Strain Sensor

← Back to AIMS

By Charles Xie ✉

A strain gauge is a conductor whose resistance changes when it is stretched. The metal-foil gauges glued to bridges and aircraft wings change their resistance by about two percent for every percent of stretch, a ratio called the gauge factor. A semiconducting carbon nanotube can do far better, because stretching it changes its band gap, and the number of electrons that can carry current in a semiconductor depends exponentially on the gap. Heyd, Charlier and McRae computed in 1997 that stress along the axis shifts the gap of a nanotube, and Yang and Han derived the rule in 2000: the shift per unit strain is proportional to cos 3θ, where θ is the chiral angle of the nanotube, so zigzag nanotubes respond most and armchair nanotubes not at all, and the sign of the shift depends on whether n − m leaves a remainder of 1 or 2 on division by three. Tombler and coworkers pressed a suspended nanotube with the tip of an atomic force microscope in 2000 and watched its conductance drop a hundredfold. Minot and coworkers measured the gaps of individual nanotubes under strain in 2003 and found shifts of both signs, as the rule predicts, and Cao, Wang and Dai stretched metallic, nearly metallic and semiconducting nanotubes the same year. Stampfer and coworkers built displacement sensors from single nanotubes in 2006, with a gauge factor more than a thousand times that of metal foil.

In AIMS a student can turn that physics into a design problem. The example below is a conversation in four turns: ask for the most sensitive nanotube that stays semiconducting to 3 % strain, ask whether the model can be trusted for the nanotube it picked, ask how the nanotubes one can actually buy compare, and stretch the chosen nanotube in a tensile test to see how much stress 3 % takes. The AI agent records the goal, charts the band gap of several nanotubes against strain in one sweep, builds the winner atom by atom, and reports whether the goal was met.

Why Stretching Moves the Gap

In the π tight-binding model each electron hops between the p_z orbitals of bonded carbon atoms, and the hopping energy decreases as the bond lengthens. In an unstrained nanotube the three bonds at every atom are equal, 0.142 nm long with a hopping energy of 2.7 eV. With three equal hoppings the valence and conduction bands of graphene touch at the K points of its Brillouin zone. Rolling the sheet into a nanotube restricts the electron to a set of parallel lines across that zone, one for each wavelength that fits around the circumference, and the nanotube is metallic when a line passes through K and semiconducting otherwise, with a gap set by the distance from K to the nearest line. The previous article works through that picture.

Stretching the nanotube along its axis lengthens the three bonds by different amounts. A bond along the axis stretches by the full strain. A bond around the circumference shortens, because the nanotube narrows as it stretches, by a fraction of the strain called the Poisson ratio, about 0.2 for a nanotube. A bond between the two directions changes by an amount in between. The three hoppings are no longer equal, and the point where the bands touch moves away from K. The lines stay where they were: a line is fixed by the number of wavelengths that fit around the circumference, and that number does not change when the circumference stretches. So the touching point moves between fixed lines, and the gap, which is set by the distance from the touching point to the nearest line, changes.

How fast the gap changes depends on the direction in which the touching point moves. In a zigzag nanotube one bond lies along the axis, the point moves straight across the lines, and the gap changes fastest. In an armchair nanotube the point moves along its own line, so the gap does not change, and an armchair nanotube stays metallic at any strain. Between the two the response is proportional to cos 3θ, where θ is the chiral angle, 0° for zigzag and 30° for armchair. The sign depends on which side of the touching point the nearest line lies. In nanotubes where n − m leaves a remainder of 1 on division by three, tension moves the point away from the nearest line and the gap widens. Where the remainder is 2, tension moves the point toward the line, the gap narrows, and at a large enough strain it closes. A metallic nanotube, with a line through its touching point, opens a gap as soon as the point moves off the line, unless it is armchair. Yang and Han wrote the rule as a formula: the gap changes by 3·t·(1 + ν)·ε·cos 3θ, where t is the hopping energy, ν the Poisson ratio and ε the strain, with the hopping taken to decrease as the inverse square of the bond length. For a zigzag nanotube with t = 2.7 eV and ν = 0.2 that is about 100 meV per percent of strain.

A shift of 100 meV is a large signal. The number of electrons excited across a gap at temperature T is proportional to exp(−E_g/2kT), and 2kT is 51 meV at room temperature, so a shift of 100 meV changes that number by a factor of about seven. By this estimate one percent of stretch changes the resistance of the nanotube sevenfold, a gauge factor in the hundreds, against 2 for a metal foil.

Eight Nanotubes on One Chart

The first turn asks for the most sensitive nanotube that stays semiconducting to 3 % strain. The agent records the goal, then sweeps eight nanotubes from 0 to 3 % in steps of a quarter of a percent: the zigzag nanotubes (7,0), (8,0), (10,0) and (11,0), the chiral nanotubes (8,2), (9,2), (10,2) and (7,3), one of them, (8,2), metallic. For each nanotube the sweep solves the infinite nanotube by zone folding with the stretched bond lengths, which takes a fraction of a second for any chirality, and reports the gap at rest, the gap at 3 % and the sensitivity, the slope of the gap against strain at zero. The four zigzag nanotubes respond at 159 to 168 meV per percent, the chiral ones less, in proportion to cos 3θ: the (7,3) nanotube, 17° from zigzag, responds at 105 meV per percent, 0.65 of the zigzag value, and cos 51° is 0.63. The sign follows the family. The gaps of (7,0), (10,0), (9,2) and (7,3) widen under tension, and those of (8,0), (11,0) and (10,2) narrow. The metallic (8,2) opens a gap of 0.39 eV at 3 %. The agent picks (8,0), whose sensitivity of 168 meV per percent is the largest in magnitude, and whose gap narrows from 1.27 to 0.79 eV at 3 %, still a semiconductor. The agent builds a 3 nm segment of the nanotube, loads it into the lab, and reports the goal met.

The table collects the fourteen nanotubes swept across the first three turns. The sensitivity is the slope of the gap at zero strain, in the model of the example, whose hopping energy decreases faster with bond length than the inverse square of the textbook rule; the section on the model below gives the factor.

A Thin Nanotube and a Safer One

The second turn objects. The (8,0) nanotube is 0.63 nm across, and the π model treats the bonds of a nanotube that thin as if the sheet were flat. In a real nanotube that narrow, the curvature mixes the in-plane σ orbitals into the π states and changes the gap. The tool itself flags the nanotube as below the diameter at which the model can be trusted, about 1 nm. The agent sweeps four wider zigzag nanotubes, (13,0), (14,0), (16,0) and (17,0), all 1 nm across or more. Their sensitivities are the same as the thin ones to within a few percent, 161 to 166 meV per percent, because the sensitivity depends on the chiral angle and all four are zigzag. The agent chooses (13,0): 1.02 nm across, responding at 161 meV per percent, 4 % below the (8,0), with a gap that widens from 0.74 to 1.09 eV at 3 %.

The chart of this turn shows a second effect. The gap of the (16,0) nanotube widens to 0.91 eV at 2 % strain and then narrows, to 0.79 eV at 3 %. The gap of the (13,0) does the same, reaching 1.13 eV at 2.75 % before it begins to narrow. In a nanotube whose gap widens, the touching point moves away from the nearest line and toward the next one, and halfway between the two lines the next one becomes the nearest. From then on the gap narrows. At rest the touching point sits one third of the line spacing from the nearest line, and at the turning point it sits one half, so the largest gap a nanotube of this family can reach is about 1.5 times its gap at rest: 0.90 eV for the (16,0), 1.10 eV for the (13,0). A wider nanotube reaches its turning point at a smaller strain, because its lines are closer together. The agent chose the (13,0) over the (16,0) as the better balance of size and sensitivity. The chart gives a second reason: the (16,0) stops responding inside the 3 % working range of the sensor.

The Nanotubes You Can Buy

The third turn asks about the nanotubes sold today. Samples enriched in a single chirality are mostly (6,5) and (7,5), and the question is how a sensor made from them would compare with the (13,0). The agent sweeps the three together. The (6,5) responds at 32 meV per percent and the (7,5) at 41, a fifth and a quarter of the (13,0). Both are close to the armchair angle, 27° and 24.5°, where cos 3θ is 0.16 and 0.28. The sign again follows the family: the gap of the (6,5), with remainder 1, widens from 1.02 to 1.11 eV, and the gap of the (7,5), with remainder 2, narrows from 0.94 to 0.83 eV. Both stay semiconducting to 3 %, and both are thinner than 1 nm, so the model's caution about curvature applies to them as it did to the (8,0). The zigzag nanotubes that respond most are scarce in the sorted samples sold today.

How Much Stress Three Percent Takes

The last turn builds the (13,0) nanotube, 338 carbon atoms and 2.8 nm long, and runs a tensile test on it in the lab. The test clamps the two ends, stretches the nanotube in eight steps to 3 %, and lets the atoms relax at each step under the Tersoff bond-order potential, the force field fitted to the stiffness of graphite and diamond. The stress is the force divided by the cross-section of the wall, taken as a ring of the nanotube's diameter and 0.335 nm thick, the spacing of the layers in graphite. The slope of the stress–strain curve at zero strain gives a Young's modulus of 1.19 TPa, the same to within a few percent as the 1.18 TPa of the (6,6) and (10,0) nanotubes in the example Stretching a Carbon Nanotube, and in the range of 1 TPa measured on real nanotubes. At the last point of the test, 2.8 % strain, the stress is 29.6 GPa, and the curve has begun to bend below its initial slope, so 3 % takes about 31 GPa. In the tensile tests of Yu, Files, Arepalli and Ruoff in 2000, ropes of single-wall nanotubes broke at strains up to 5.3 %, and the same group measured multiwall nanotubes breaking at strains up to 12 %. A sensor working to 3 % stays inside the range that measured nanotubes survive.

Running Experiments by Asking

The student states a design goal in a sentence. The agent records it, chooses which nanotubes to sweep, runs the sweep, reads the chart, builds the winner, and reports against the goal. When the student objects to the winner, the agent sweeps again with the objection as a new constraint. Choosing the eight nanotubes of the first sweep, picking the wider candidates of the second, and running the tensile test are the agent's work.

The Model and Its Limits

The strain sweeps use the nearest-neighbor π tight-binding model with a hopping energy of 2.7 eV at the rest bond length, solved by zone folding: the states of the nanotube are the states of graphene on the allowed lines, and strain enters only through the three hopping energies, each set by the stretched length of its bond. The lines do not move with strain, because the number of wavelengths around the circumference is fixed. The nanotube narrows by a Poisson ratio of 0.2. The hopping energy decreases with bond length as exp(−β(r/r₀ − 1)) with β = 3.37, which is steeper than the inverse-square decay of Yang and Han's formula, where β would be 2. The sensitivities in the table are therefore 1.7 times the textbook values: 160 meV per percent for a zigzag nanotube against about 100. The ratios between nanotubes, the cos 3θ rule and the sign rule do not depend on β, and the sweep accepts a different decay on request.

The model omits what curvature does to the orbitals of a thin nanotube, which is why the agent moved from the (8,0) to the (13,0). It omits torsion, which also shifts the gap and enters Yang and Han's formula as a second term. It omits the interaction between the excited electron and the hole it leaves behind, and it says nothing about the contacts through which a real sensor is read, which contribute a resistance of their own. The tensile test is a different model: the Tersoff potential moves the atoms, at absolute zero and with no thermal vibration, and it would stretch bonds smoothly past the strain at which a real nanotube breaks, so it reports a modulus and a stress but no breaking point. The gap sweeps are for infinitely long nanotubes. The 2.8 nm segment in the lab is the structure that is stretched, and its gap would carry the finite-size effects described in the previous article.

Conclusion

This article uses the strain sensitivity of carbon nanotubes to illustrate a new way to learn science: students converse with an AI agent that turns a plain-language design goal into a quantum-mechanical calculation and hands back the result as a chart. Students find for themselves that the response of a nanotube to stretching is set by its chiral angle and its sign by its family, two rules predicted in 2000 and confirmed by measurement in 2003. They watch the agent pick a winner, object that the winner is too thin for the model, get a safer nanotube that responds as well, learn why the nanotubes on the market respond a fifth to a quarter as well, and stretch the chosen nanotube to see how much stress the working strain takes. By pairing a physically grounded solver with an AI that can set up the sweeps and run them, AIMS lets students carry a design from a rule of physics to a candidate nanotube and its mechanical test.

← Back to AIMS