The technique at the heart of the work is known as a light-induced self-written waveguide, or LISW. The concept dates back to the 1990s and relies on a deceptively simple property of photocurable resin: the material hardens preferentially where the optical field is most intense. When curing light shines out of the facet of an optical fiber or waveguide into a droplet of liquid resin, the resin solidifies along the beam path, forming a thin solid waveguide that extends outward from the facet. If two opposing waveguides face each other across a gap of resin, both emit curing light, and the two self-written guides grow toward each other until they meet. Crucially, if the two waveguides are laterally offset from one another, the resulting structure does not form a straight bridge but an elegant S-shaped curve that bends to connect the two misaligned cores. The waveguide essentially forgives the assembly error by reshaping itself around it.

Earlier demonstrations of LISW connections had already shown respectable alignment tolerances of roughly plus or minus 4 micrometers for standard single-mode waveguides with core diameters of 8 to 9 micrometers. Researchers also knew that widening the gap between the waveguides helped: at a gap of 500 micrometers, tolerances near 10 micrometers had been reported, because a longer connection gives the S-bend more room to curve gently. But stretching the gap creates its own difficulties. A longer waveguide means a longer propagation path and therefore more accumulated loss, and the curing light itself weakens as it travels through the resin through scattering and absorption. Beyond a certain length, the light may simply be too faint to keep polymerizing the resin, and the self-writing process stalls before a usable guide forms. The Japanese team set out to find whether a large alignment tolerance could be achieved without paying the price of a long gap.

Their starting point was geometry. Using an analytical expression relating the lateral offset, the gap, and the bending radius of an S-bend built purely from circular arcs, they showed that for a fixed gap, the maximum tolerable offset depends on how tightly the waveguide can bend before radiating light away. A waveguide that tolerates a smaller bending radius can absorb a larger misalignment at the same gap. That is precisely where the refractive-index contrast, denoted by the Greek letter delta, enters the picture. A high-delta waveguide, in which the core index differs strongly from the cladding index, confines light far more tightly and can sustain sharp curves with little radiation loss. In other words, the team identified a materials-based route to alignment tolerance: instead of stretching the gap to relax the bend, make the waveguide itself bend-tolerant by boosting its index contrast.

To test this idea quantitatively, the researchers ran two-dimensional finite-difference time-domain simulations using the open-source Meep software. They modeled an S-bent LISW connecting two waveguides with 3-micrometer cores and a delta of 2.2 percent, values chosen to resemble the spot-size converters found at the edges of silicon photonic chips. Sweeping the gap, waveguide width, and delta across a physically realistic range, they found a clear trend: larger gaps and higher index contrasts both reduced coupling loss at a fixed lateral offset of 8 micrometers. Just as importantly, at a delta of 1.3 percent or above, the coupling loss became nearly independent of the waveguide width, meaning the connection would be forgiving of fabrication variations as well as assembly errors. A delta of 1.3 percent is well within reach of commercial photocurable resins, whose index rises during photopolymerization as the material’s molecular structure and density change.

The experimental demonstration used a commercial acrylic resin with a refractive index of 1.49 after curing and 1.47 before, at a wavelength of 1.31 micrometers, corresponding to a delta of about 1.3 percent. The team connected two high-numerical-aperture fibers, each with a mode field diameter of only 3.4 micrometers and an estimated core diameter near 3 micrometers, a deliberately punishing test case since small cores make alignment exquisitely sensitive. Across a gap of just 250 micrometers, the bare fiber-to-fiber coupling loss was a dismal 19.4 decibels, dominated by light diffracting away into the gap. After dispensing resin and irradiating both fibers simultaneously with 405-nanometer laser diodes at minus 15 dBm for 2.5 seconds, the loss collapsed to 0.7 decibels. The self-written bridge had effectively erased the gap.

The headline result came when the researchers deliberately misaligned the fibers. By fitting their loss measurements with a quadratic curve and finding where the excess loss reached 1 decibel, they extracted a 1-dB alignment tolerance of approximately plus or minus 7.4 micrometers, achieved at the modest gap of 250 micrometers. Microscope images confirmed that an S-bent-like waveguide had indeed formed between the offset cores, curving smoothly from one fiber to the other. For small-core devices such as silicon photonic chips and high-delta multicore fibers, whose tiny mode field diameters normally impose brutal alignment demands, a self-forming connector that shrugs off several micrometers of offset at a short gap is a genuinely attractive proposition.

The study is also candid about where theory and experiment diverge. The simulations predicted losses below 1 decibel even at an 8-micrometer offset, yet the measured tolerance was somewhat tighter. The authors attribute the discrepancy to the simplified model: the simulation assumed a uniform refractive index and constant width along the LISW, while the real structure likely varies in both, and its trajectory may mix straight and bent sections that shrink the effective bending radius. The two-dimensional model itself may also shift absolute loss values. Characterizing the actual structure is hard, since the waveguide is minuscule and removing all uncured resin without disturbing it is difficult. The residual 0.7-decibel loss appears to stem mostly from waveguide-related factors rather than resin absorption; transmission measurements through bulk-cured resin samples 200 to 400 micrometers thick differed by less than 0.1 decibel, and three-dimensional simulations of mode-field mismatch at the two fiber interfaces suggest a contribution of roughly 0.4 to 0.5 decibels that higher delta could further reduce.

Gap-dependence experiments added a further caution. At a fixed 8-micrometer offset, losses fell as the gap grew from small values toward about 250 micrometers, matching the simulations, but beyond 300 micrometers the loss climbed again, contrary to the model. At 350 micrometers the loss spiked sharply, driven by severe waveguide defects caused by insufficient curing power; raising the curing power from minus 15 to minus 12 dBm partially rescued the connection, though the loss still exceeded that at 250 micrometers. The physical picture is a tug-of-war: a larger gap gives diverging writing beams more room to overlap geometrically across a given offset, but it also dilutes their intensity, and past a threshold the light can no longer sustain polymerization. The maximum writable offset is therefore set jointly by the numerical aperture, gap, optical power, and resin photosensitivity, meaning that simply lengthening the gap is not a reliable recipe for both low loss and high tolerance.

The takeaway is a new design rule for the interconnects that future AI data centers will depend on. Rather than fighting for ever-finer alignment or stretching gaps until curing light fades, engineers can raise the index contrast of the self-written waveguide itself, letting tight optical confinement buy bending tolerance at short gaps. The work remains an early-stage demonstration on fiber pairs rather than full chip assemblies, and the authors note that optimizing formation conditions and resin parameters could push losses lower still. But the vision is compelling: optical connectors that assemble themselves out of liquid resin, guided by the very light they will one day carry, quietly bending around the imperfections of the machines that place them.

Subject of Research: High-refractive-index-contrast light-induced self-written waveguides for alignment-tolerant optical coupling

Article Title: High-alignment-tolerance optical coupling based on high-Δ light-induced self-written waveguides

Article References: Saito, Y., Shikama, K., & Sugihara, O. (2026). High-alignment-tolerance optical coupling based on high-Δ light-induced self-written waveguides. Results in Optics, Article 101179. https://doi.org/10.1016/j.rio.2026.101179

Image Credits: AI Generated

DOI: 10.1016/j.rio.2026.101179

Keywords: light-induced self-written waveguides, optical coupling, alignment tolerance, silicon photonics, multicore fiber, photocurable resin, refractive-index contrast, FDTD simulation, optical interconnects, mode field diameter, S-bend waveguide, fiber-to-chip coupling