Switching Light In Three Optical Cycles

Twenty femtoseconds is three cycles of the light you are trying to modulate. At 1.9 µm a single optical period lasts 6.3 fs, so a modulator that opens and closes in 20 fs has not imposed an envelope on the beam — it has interfered with the beam's own carrier wave. That is the regime a group at Purdue University and Johns Hopkins say is reachable in a transparent conducting oxide, in a paper published in Nature Communications on 25 September 2026.

The claim needs one qualification stated up front, because the paper states it in its own first sentence: this is theory. Nothing was built. What the authors did was show, through inverse design and a hot-electron model of the material, that an oscillatory, sign-reversing optical response ought to appear under conditions nobody has quite pushed a transparent conducting oxide into — and that the same mechanism offers an explanation for fast oscillations other groups have already measured and set aside.

Transparent conducting oxides — indium tin oxide, aluminium-doped zinc oxide and their relatives — occupy an unusual position in optics. They are the material on the front of a touchscreen: transparent to visible light and electrically conductive, which are ordinarily contradictory properties. They manage it with a free-electron density of around 10²⁰ to 10²¹ per cubic centimetre, one to two orders of magnitude below a metal like gold at 5.9 × 10²² cm⁻³. Gold's plasma frequency sits in the ultraviolet, which is why it reflects everything below it. Drop the carrier density by one to two orders of magnitude and the plasma frequency falls into the near infrared.

At that frequency the material's permittivity passes through zero. This is the epsilon-near-zero condition, and it is the whole reason TCOs matter to photonics rather than to display manufacturing. Where the real part of the permittivity crosses zero, the refractive index — which goes as its square root — becomes extraordinarily sensitive to small changes in the electron population. A perturbation that would shift the index of glass in the fifth decimal place can shift the index of a TCO at its ENZ wavelength by a substantial fraction of unity. The material sits on a knife edge, and the knife edge is placed by doping: the AZO film modelled in this paper has a carrier concentration of 7.35 × 10²⁰ cm⁻³, which puts the ENZ wavelength at 1425 nm, conveniently inside the band where femtosecond sources are cheap.

The last decade of ENZ research has exploited exactly this. Shine an intense femtosecond pulse on an ITO or AZO film and the index changes by nearly one — an enormous nonlinearity — then relaxes fully in under 500 femtoseconds as the electrons cool. Under 500 fs is fast by the standards of any electro-optic modulator. It is slow by the standards of the light itself.

What moves a TCO's index is not the field of the pump pulse directly. It is the temperature of the electron gas.

The chain is worth stating plainly, because the whole paper hangs on it. The pump is absorbed, dumping energy into the conduction electrons. Electron–electron scattering in these materials is fast — tens of femtoseconds — so the electrons redistribute that energy among themselves long before they hand it to the lattice. For a brief window you have a hot electron gas inside a cold crystal. A hot electron gas occupies higher-energy states. In a material with a parabolic conduction band that would be the end of the story and nothing optical would change, because the effective mass would be the same at every energy. But the conduction band of a doped oxide is not parabolic: it flattens at higher energy, and a flatter band means a heavier electron. The paper's phrasing is that carriers driven to higher energies acquire a substantially larger effective mass, producing a pronounced modification of the plasma frequency.Effective mass is one of the two quantities that set the plasma frequency; carrier density is the other. In the absorber layer the authors hold the carrier density fixed — they impose carrier conservation under intraband optical excitation, explicitly rejecting the naive Sommerfeld treatment that would predict an unphysical thermal increase in carrier density. So in the absorber, heating does exactly one thing to the plasma frequency: it drags it down, by making the electrons heavier.

This paper's contribution is to ask what happens when the electron temperature stops being a modest perturbation and is driven as far as the material will go. In the modelled structure it approaches the Fermi temperature itself — 11,584 K at this doping — at which point the electron distribution is no longer a cold Fermi sea with a warm edge but something closer to a classical gas.

At extreme electron temperature a second process turns on and works against the first, and the competition is what produces the oscillation.

Heavier electrons pull the plasma frequency down, which moves the ENZ wavelength and changes the index at whatever wavelength you are probing. At the same time, hot electrons lose momentum faster — through electron–phonon scattering, impurities, grain boundaries and surface roughness — so the Drude damping rate rises. The authors model this as Γ(Te) = Γ₀[1 + (Te/T₀)ᵐ], adopting a quadratic dependence, m = 2, as the baseline and running a sensitivity analysis across a range of m rather than presenting the exponent as settled.

Permittivity has a real part and an imaginary part, and these two influences do not act on them in step. In the paper's own framing, the refractive and absorptive contributions drive the optical response in competing directions, toward either a more dielectric and transmissive state or a more metallic and absorptive one. Because the electron temperature rises and falls on a timescale comparable to the optical period, the film swings between those two characters repeatedly inside a single relaxation event. The transmittance oscillates, and the sign of the index change reverses.

The analogy is a set of scales with weights being added to both pans at once, by two people working at different speeds. The pointer does not travel smoothly from one side to the other; it crosses back and forth several times before it settles. Prior ENZ experiments measured where the pointer settled, over a few hundred femtoseconds. This paper is about the crossings.

An ordinary TCO film on a glass substrate does not absorb enough of the pump to reach those temperatures, so the paper puts the film on a mirror and lets an algorithm choose the mirror.

The structure is a 10 nm AZO film on top of an inverse-designed Si/SiO₂ dielectric mirror of five layers with varying thicknesses. Inverse design here means the layer thicknesses are not taken from a quarter-wave textbook recipe but optimised numerically against a target — in this case maximum absorption in the TCO at the ENZ wavelength. The result concentrates the pump field in a film roughly one one-hundred-and-fortieth of the pump wavelength thick, and drives its electron gas far hotter than the same film on fused silica would get.

The simulated excitation conditions should travel with every number quoted from this work: a 10 fs pump pulse at 1425 nm, 2 mJ/cm² fluence, incident at 60° from normal — which is a peak intensity of about 200 GW/cm². The probe is p-polarised at 50° incidence, sampled off-ENZ around 1.2 to 1.3 µm, near-ENZ at 1.26 and 1.431 µm, and at 1.8 to 2.0 µm for the acceptor layer described below.

Under those conditions the modelled transmittance oscillates with a period of about 20 fs, approximately four optical cycles of the 1.431 µm probe. That is a more pronounced oscillatory response than has been observed experimentally in TCOs to date.The acceptor layer

Transmittance oscillation is a useful signature but not, by itself, a useful device. What a photonic engineer wants to modulate is the refractive index, and getting the index itself to oscillate needed a second element — and, this time, a genuine change in carrier density.

The authors add a 2 nm TCO electron-acceptor layer on top of the inverse-designed cavity, doped an order of magnitude lower at 0.5 × 10²⁰ cm⁻³. The doping difference sets up an energy barrier at the junction, and the very hot electrons in the absorber beneath have enough energy to cross it. That thermionic injection dumps carriers into the thin acceptor layer on the same timescale the absorber is heating, producing a carrier-density swing there that does not merely follow the pump envelope — and carrier density, unlike effective mass, pushes the plasma frequency the other way.

The magnitude is modest — an index modulation of around two per cent — and it is worth being clear-eyed about that. Two per cent is far below the near-unity index changes ENZ films show in slow measurements. The trade is depth for speed, and the paper's argument is that two per cent at three optical cycles is a different kind of object from two per cent at three hundred femtoseconds.

Modulate a material slowly compared with the optical period and you have amplitude or phase modulation: the carrier wave stays recognisable and you have imposed a slow envelope on it. Modulate on the scale of the period itself and the distinction between carrier and modulation collapses. The material is no longer a medium the light passes through with a property attached; it is a medium whose property is changing while the light is inside it.

That regime already has a body of theory attached, most of it under the heading of time-varying photonics and photonic time crystals — media whose refractive index is modulated periodically in time rather than in space. The predicted phenomena are genuinely unlike anything in ordinary optics: momentum bandgaps instead of energy bandgaps, parametric amplification of light drawn from the modulation rather than from a gain medium, reflection and refraction at a boundary in time. The obstacle has never been theoretical. It is that no material would change its index deeply enough, fast enough and repeatedly enough at optical frequencies. Most demonstrations of time-varying effects have been done at microwave frequencies, where the period is tens to hundreds of picoseconds and electronics can keep up. The paper notes that experimental realisation of a photonic time crystal in the visible to infrared has not yet been observed.

What this work argues is that a transparent conducting oxide, driven near its Fermi temperature on an inverse-designed mirror and topped with a thin acceptor layer, is a candidate building block for attempting it at 1.9 µm. Not a demonstration of a photonic time crystal — a plausible route to the modulation such a thing would require.

The experiment implied by the paper is demanding but not exotic. It needs a 10 fs source in the short-wave infrared with enough energy to reach 2 mJ/cm² on a small spot, a pump–probe geometry with two different non-normal incidence angles, and enough timing resolution to sample a 20 fs oscillation properly — sub-femtosecond delay steps and very good interferometric stability. Groups working on few-cycle and attosecond physics have all of that.The harder part is the sample: a five-layer silicon–silica mirror with a 10 nm AZO film at a specified carrier concentration and a 2 nm acceptor layer on top, with interfaces clean enough that the junction barrier is the designed one rather than whatever the deposition happened to produce. That is a coatings problem as much as a physics problem — ångström-level thickness control, doping uniformity in a film a few tens of atoms thick, interface quality that survives 200 GW/cm² — and it is the kind of coatings problem that decides whether a prediction like this becomes a measurement or stays a figure in a journal.

There is also a durability question the paper does not address, and it is the one an engineer asks first. Electron temperatures approaching 11,584 K are survivable precisely because the lattice never gets there; the electrons cool into the phonons long before thermal equilibrium would melt anything. But that energy does reach the lattice eventually, and at a repetition rate high enough to be useful for data modulation, the average power handling of a ten-nanometre film under 200 GW/cm² peak becomes the limiting specification. Nobody has published that number for this structure.

Fast index modulation in ENZ materials has been a known effect for a decade, and the number everyone quoted was a full relaxation in under 500 femtoseconds. This paper argues that the sub-500-femtosecond relaxation is not the fundamental limit but an envelope, and that inside it there is structure at the optical period — structure that shows up, unexplained and largely unremarked, in data already collected.

If that reading is right, the useful timescale of a transparent conducting oxide has been misstated by more than an order of magnitude, and a class of time-varying optical phenomena that has been theoretically available and experimentally out of reach acquires a specific material system and a specific set of conditions to attempt. If it is wrong, the cost of finding out is one difficult sample and a few weeks of a good ultrafast lab's time. Either way, what has changed is that the measurement is now defined. The preprint has been available since December 2025; peer review has now put the claim on the record, and the sample it calls for is a coating problem somebody will take on.